1
00:00:00,480 --> 00:00:01,600
I'm David Pinsov.

2
00:00:01,840 --> 00:00:03,120
I'm David Petrozowski.

3
00:00:03,440 --> 00:00:07,360
And this is Evolutionary Psychology, the podcast, Actual Evolutionary Psychology.

4
00:00:07,680 --> 00:00:10,320
By actual evolutionary psychologists.

5
00:00:11,920 --> 00:00:15,520
Wow, that was worthy of an Oscar.

6
00:00:15,760 --> 00:00:16,000
Okay.

7
00:00:30,890 --> 00:00:36,570
David today's guest is uh Tidek Quillian and he studies really

8
00:00:36,810 --> 00:00:42,650
complicated things that are abstract and difficult to grasp from an adaptationist

9
00:00:42,730 --> 00:00:43,210
perspective.

10
00:00:43,840 --> 00:00:47,360
Well yeah, you make it sound really exciting and appealing for the listener.

11
00:00:47,840 --> 00:00:52,080
Well I I honestly this is if you if you're if you're if you feel like you want a mental

12
00:00:52,080 --> 00:00:54,080
Rubik's Cube, this is the episode for you.

13
00:00:54,320 --> 00:00:57,600
This this is uh a a mentally challenging episode for sure.

14
00:00:57,680 --> 00:01:00,640
It certainly um expanded both of our minds.

15
00:01:00,880 --> 00:01:05,920
Um so yeah maybe if you're sleepy this isn't the best time to to dive into this meaty

16
00:01:05,920 --> 00:01:06,160
podcast.

17
00:01:06,560 --> 00:01:12,450
I did I did uh uh I did rant about uh what a domain is though so there's yeah we

18
00:01:12,610 --> 00:01:17,490
we had some good rants in there about uh simplicity and domain specificity and what it

19
00:01:17,490 --> 00:01:18,850
means to be computational.

20
00:01:19,250 --> 00:01:25,010
Uh though this this one if if you're you know alert, you got a cup of coffee with you and

21
00:01:25,010 --> 00:01:31,010
you're and you think you can tackle this, this is a really um fascinating, insightful, mind

22
00:01:31,170 --> 00:01:35,410
expanding episode about the nature of causality, yeah.

23
00:01:35,570 --> 00:01:40,020
Uh which touches on literally everything the mind does, which is a point we get to in the

24
00:01:40,020 --> 00:01:40,500
episode.

25
00:01:40,820 --> 00:01:44,260
We could not do anything if we had no understanding of causality.

26
00:01:44,500 --> 00:01:50,100
Uh and and Taydeg just has a really uh elegant, insightful way of

27
00:01:50,100 --> 00:01:53,140
thinking about this complex topic.

28
00:01:53,300 --> 00:01:57,550
Um and I I found a lot of value and learned a lot from uh talking to Tay Deg.

29
00:01:57,790 --> 00:01:57,950
Yeah.

30
00:01:58,030 --> 00:02:03,710
Uh it's a really cool example of cool uh work at the cutting edge of uh computational

31
00:02:03,950 --> 00:02:08,510
modeling and um sort of philosophy of causation.

32
00:02:08,910 --> 00:02:12,750
Um and uh I think it's a challenge.

33
00:02:12,830 --> 00:02:17,710
It's this is you should take this as a challenge to um uh grapple with these issues.

34
00:02:17,790 --> 00:02:20,110
These are these are chewy issues to deal with.

35
00:02:20,840 --> 00:02:21,160
Absolutely.

36
00:02:21,240 --> 00:02:24,120
Very chewy, very crunchy, um, very meaty.

37
00:02:24,280 --> 00:02:26,600
Uh I'm just gonna run through the food metaphors.

38
00:02:27,400 --> 00:02:28,520
Uh this is great.

39
00:02:28,680 --> 00:02:33,800
Um but uh but yeah, without further ado, here is Tidek uh and uh David and I talking about

40
00:02:33,800 --> 00:02:35,880
all things easy like causality.

41
00:02:40,680 --> 00:02:44,520
Tadek, tell us about where you are and how long you've been there.

42
00:02:45,610 --> 00:02:46,970
I'm in Scotland.

43
00:02:47,290 --> 00:02:48,730
I'm at the University of Edinburgh.

44
00:02:48,890 --> 00:02:51,050
And it's been about five years now.

45
00:02:51,690 --> 00:02:56,490
There's like a storied history of Edinburgh and um computer science.

46
00:02:56,650 --> 00:02:59,450
And um yeah, is that true?

47
00:02:59,930 --> 00:03:05,130
So the I think the the first center for co cognitive science uh maybe in the world was in

48
00:03:05,130 --> 00:03:06,170
Edinburgh.

49
00:03:07,120 --> 00:03:13,120
So we had uh people like Nick Chater were here, uh people like Jeff Hinton, Peter

50
00:03:13,120 --> 00:03:14,000
Diane.

51
00:03:14,800 --> 00:03:18,880
Why Edinburgh and then they chose to dissolve the center for some reason.

52
00:03:20,080 --> 00:03:21,920
Why why Edinburgh?

53
00:03:22,000 --> 00:03:27,440
I mean not to be US centric about it or continental centric about it, but like why

54
00:03:27,440 --> 00:03:28,480
Edinburgh?

55
00:03:29,410 --> 00:03:34,610
Um how do you mean why uh such a what is the why the

56
00:03:35,250 --> 00:03:41,170
the the the answer I like is that maybe in the it's a city that has a long intellectual

57
00:03:41,170 --> 00:03:46,530
history and we have people like um people like David Hume that's yeah I was gonna say David

58
00:03:46,610 --> 00:03:48,930
I was gonna do David Hume joke actually.

59
00:03:50,450 --> 00:03:54,610
So I I like to think that that is where the cognitive science energy is coming from.

60
00:03:54,850 --> 00:03:55,170
Yeah.

61
00:03:55,560 --> 00:03:56,440
David Hume.

62
00:03:57,880 --> 00:03:59,320
Have you read David Hume?

63
00:04:00,200 --> 00:04:00,760
I have.

64
00:04:01,240 --> 00:04:05,800
It's actually I started as a political science major.

65
00:04:06,600 --> 00:04:10,680
And there was a political philosophy class I had to take.

66
00:04:11,240 --> 00:04:17,080
And almost kind of randomly, I trust you to read a book by David Hume because we had to

67
00:04:17,080 --> 00:04:19,400
kind of pick a book and write a report about it.

68
00:04:19,720 --> 00:04:24,610
And that's, I think that was my gateway direct towards uh cognitive science.

69
00:04:24,850 --> 00:04:25,250
Yeah.

70
00:04:25,570 --> 00:04:31,410
Um just to be on this for a second, um, there are two opinions about David Hume, and I

71
00:04:31,410 --> 00:04:32,050
think both are right.

72
00:04:32,130 --> 00:04:33,330
And I'm curious where you stand.

73
00:04:33,490 --> 00:04:37,250
One is that he was not an empiricist, and one is that he was.

74
00:04:37,410 --> 00:04:38,450
Where do you stand?

75
00:04:39,970 --> 00:04:45,170
Um I think I I stand with the kind of classical view that he was an empiricist.

76
00:04:45,730 --> 00:04:51,280
Um but I mean I can see the the pull of the opposing argument

77
00:04:51,840 --> 00:04:57,840
because uh so Immanuel Kant said that he was woken from his dogmatic slumber

78
00:04:57,840 --> 00:05:01,840
by reading David Hume, and Kant is famously not an empiricist.

79
00:05:02,000 --> 00:05:02,160
Yeah.

80
00:05:02,320 --> 00:05:08,130
Um and so I kind of read David Hume as um like someone who

81
00:05:08,130 --> 00:05:13,970
pushes empiricism to to its logical limits, and you find that if you push empiricism

82
00:05:13,970 --> 00:05:19,570
to its limits, then you're forced to uh quite an extreme degree of skepticism.

83
00:05:19,730 --> 00:05:23,970
Uh and then that's a reason in itself to be skeptical of empire of empiricism.

84
00:05:24,450 --> 00:05:29,570
For people who might not know, can we just make this kind of grounded and concrete?

85
00:05:29,730 --> 00:05:35,700
Like um just like you know, what what was the basic argument This is gonna turn into uh

86
00:05:35,780 --> 00:05:40,340
uh um uh what what do you know about David Hume dial a friend?

87
00:05:40,420 --> 00:05:44,740
Uh well I mean at least for for my purposes, yeah.

88
00:05:44,980 --> 00:05:49,780
Um what I find interesting is his argument about about causality.

89
00:05:49,940 --> 00:05:50,180
Yeah.

90
00:05:50,420 --> 00:05:56,260
Um because um so David Hume was concerned with why it is that

91
00:05:56,260 --> 00:05:58,740
we can come to have causal knowledge at all.

92
00:05:59,060 --> 00:06:04,830
Okay, and so he observed that well, we can come to know that there is an

93
00:06:04,830 --> 00:06:10,590
association between uh the presence of clouds in the sky and then

94
00:06:10,590 --> 00:06:11,230
rain.

95
00:06:11,550 --> 00:06:17,550
And I know day after day we see that when there is when there are clouds in the

96
00:06:17,550 --> 00:06:19,150
sky, rain follows.

97
00:06:19,870 --> 00:06:24,750
And so we form by by induction this kind of sort of a strong association between those two

98
00:06:24,750 --> 00:06:25,150
things.

99
00:06:25,790 --> 00:06:31,710
But then it seems that in order to really uh come to the conclusion that there is a

100
00:06:32,320 --> 00:06:38,320
causal relationship between those things, that the the rain causes the that the cloud

101
00:06:38,320 --> 00:06:42,400
is the cause of the rain, then there's an extra step that we need to take.

102
00:06:43,120 --> 00:06:48,000
And it is not clear that this take can be justified in in a rational way.

103
00:06:48,560 --> 00:06:53,520
So the idea is that so I've always seen clouds, then rain, clouds and rain.

104
00:06:53,680 --> 00:06:57,280
And then I say, look, clouds cause rain, and somebody asks me, Well, how do you know that?

105
00:06:57,440 --> 00:07:00,880
And I say, Well, they've always preceded that.

106
00:07:01,130 --> 00:07:06,650
And the argument is that well, yeah, that's your data, but where do you get

107
00:07:07,130 --> 00:07:09,370
where do you get the conclusion that that caused that?

108
00:07:09,450 --> 00:07:10,330
Maybe it was something else.

109
00:07:10,730 --> 00:07:11,210
Yeah, exactly.

110
00:07:11,290 --> 00:07:13,850
And it's quite difficult to justify that.

111
00:07:15,050 --> 00:07:17,930
Is this just the correlation causation fallacy?

112
00:07:18,170 --> 00:07:22,410
Because you could say, well, no, clouds don't cause rain, rain causes clouds.

113
00:07:22,570 --> 00:07:27,610
Or maybe some other thing, maybe uh Zeus getting angry causes both, right?

114
00:07:27,900 --> 00:07:28,700
Um right.

115
00:07:29,020 --> 00:07:33,180
So how do we know that that it's called the causal arrow is going from clouds to rain?

116
00:07:33,260 --> 00:07:36,540
Is it is that it or is is is the the skepticism deeper than that?

117
00:07:36,940 --> 00:07:42,700
So I think this might be pushing beyond my ex philosophical expertise, but uh I I would say

118
00:07:42,700 --> 00:07:45,580
that maybe the worry is even a little bit deeper.

119
00:07:45,820 --> 00:07:51,570
Um the idea is that even if you can um sort of control exclude those

120
00:07:51,570 --> 00:07:52,450
alternative explanations.

121
00:07:52,530 --> 00:07:57,410
So even if you can definitely exclude that the causal direction is not from the rain to the

122
00:07:57,410 --> 00:07:57,890
clouds.

123
00:07:58,450 --> 00:08:04,450
Then you still have the question of what justifies me to think that there is a causal

124
00:08:04,450 --> 00:08:05,250
relationship.

125
00:08:06,210 --> 00:08:12,140
And uh we're kind of getting into the territory of um it is

126
00:08:12,140 --> 00:08:15,900
a bit weird to posit something like a causal relationship.

127
00:08:16,060 --> 00:08:17,980
This is not something that we can see.

128
00:08:18,380 --> 00:08:18,700
Yes.

129
00:08:19,260 --> 00:08:20,460
It's not so there is a data.

130
00:08:20,860 --> 00:08:23,660
Yes, yes, there is a lip of safe that we have to take regardless.

131
00:08:23,980 --> 00:08:28,140
And I think like this is something I was worried about before we had this conversation, and

132
00:08:28,140 --> 00:08:33,180
I want to like clear away because when I when I I remember when I first started learning

133
00:08:33,180 --> 00:08:38,430
about like causation and things like that as an early student, I found it very obtuse, like

134
00:08:38,430 --> 00:08:40,830
this kind of thing where I'd be like, okay, yeah, whatever.

135
00:08:40,910 --> 00:08:42,670
Like, of course, clouds, clouds, rain, whatever.

136
00:08:42,750 --> 00:08:43,870
Why guys this is important?

137
00:08:44,030 --> 00:08:49,710
And I think it's worth like just taking a step back to just dispelling like why this is not

138
00:08:49,710 --> 00:08:51,470
obtuse and why it's actually really important.

139
00:08:51,630 --> 00:08:57,520
And I think the cloud thing is a good example where my understanding is and like why

140
00:08:57,520 --> 00:09:03,280
I now find it not at all obtuse, but like it's fundamentally like the same non-bullshitty

141
00:09:03,280 --> 00:09:05,680
thing as like kids asking, well, why, why, why?

142
00:09:05,760 --> 00:09:08,720
Like that's a simple non-bullshitty thing.

143
00:09:08,960 --> 00:09:14,400
And my understanding is that it is kind of like David's thing where it's like clouds rain,

144
00:09:14,560 --> 00:09:15,520
clouds rain.

145
00:09:15,760 --> 00:09:19,110
You've always seen that relationship, but you don't know.

146
00:09:19,350 --> 00:09:21,190
It might be something else.

147
00:09:21,430 --> 00:09:26,630
And the problem of induction is really just that it's always correlation.

148
00:09:26,790 --> 00:09:30,950
Like that was kind of my read of Hume was like, it's always correlation.

149
00:09:31,430 --> 00:09:35,830
And it's weird that people don't represent that.

150
00:09:35,910 --> 00:09:41,190
Well, I'm I have a tentative guess that it is a correlation, but I should reserve in my

151
00:09:41,190 --> 00:09:45,000
mind the fact that maybe that's just a correlation.

152
00:09:45,080 --> 00:09:50,760
And I don't have this idea that no matter what, and especially in the future,

153
00:09:51,160 --> 00:09:53,080
it's going to be the same way.

154
00:09:53,240 --> 00:09:59,240
Like I'm really taking a leap of faith in saying the way it is today is the same it's

155
00:09:59,240 --> 00:10:04,200
going to be tomorrow because I don't have all the information I need to make that leap of

156
00:10:04,200 --> 00:10:04,520
faith.

157
00:10:04,600 --> 00:10:06,440
And then therefore you could be wrong.

158
00:10:06,680 --> 00:10:12,190
And to my mind, like that's the problem of induction, is it's always

159
00:10:12,190 --> 00:10:18,190
correlation because there's an infinite number of causal relationships that the

160
00:10:18,430 --> 00:10:24,350
the correlation you've seen so far is not determinative of what the

161
00:10:24,350 --> 00:10:27,470
actual underlying causal thing, and that's the obtuse way.

162
00:10:27,630 --> 00:10:31,790
But the simple way of putting it is like if you don't know why something happens, you might

163
00:10:31,790 --> 00:10:32,750
be wrong in the future.

164
00:10:32,830 --> 00:10:34,430
And that that's my read.

165
00:10:34,670 --> 00:10:40,020
And that sort of grounds it in like non-bullshit, like this is a fitness problem because

166
00:10:40,180 --> 00:10:46,020
like you gotta know like did the tiger eat the my friend for X reason or Y reason

167
00:10:46,100 --> 00:10:48,260
and can I be go back there tomorrow?

168
00:10:48,420 --> 00:10:49,700
Like that's not bullshit.

169
00:10:49,780 --> 00:10:52,980
That's not like, you know, well is it causation or correlation?

170
00:10:53,060 --> 00:10:58,500
But we know it always like the this is like really like um real um stuff here.

171
00:10:58,660 --> 00:11:02,180
I just so that's my um yeah clearing of that.

172
00:11:02,580 --> 00:11:07,540
Yeah the the often I mean not often hopefully but sometimes it will be a a life or death

173
00:11:07,540 --> 00:11:07,780
matter.

174
00:11:08,030 --> 00:11:08,990
Yeah, exactly.

175
00:11:09,150 --> 00:11:09,470
Yeah.

176
00:11:09,710 --> 00:11:10,190
I don't know.

177
00:11:11,070 --> 00:11:11,230
Yeah.

178
00:11:11,710 --> 00:11:16,830
One one example to you, one reason why it's important to represent

179
00:11:17,070 --> 00:11:22,510
causality instead of just a pattern of correlations, is that um storing

180
00:11:22,990 --> 00:11:27,870
having a a predictive model of the world that is just a pattern of correlations um is not a

181
00:11:27,870 --> 00:11:30,190
very stable way of representing the world.

182
00:11:30,750 --> 00:11:36,350
So I mean you might um notice that there is a high correlation between um there being

183
00:11:36,350 --> 00:11:40,060
rain and then the grass in your in your garden being wet.

184
00:11:40,460 --> 00:11:43,260
And you store the fact that the correlation is maybe 0.3.

185
00:11:44,780 --> 00:11:45,980
So far so good.

186
00:11:46,540 --> 00:11:51,660
But maybe you install a sprinkler in your garden and to to to water the grass.

187
00:11:52,140 --> 00:11:56,300
Now once you do that, then the correlation between the rain and the grass being wet is

188
00:11:56,300 --> 00:11:57,740
going to be much lower.

189
00:11:57,980 --> 00:11:58,220
Yeah.

190
00:11:59,180 --> 00:12:03,370
And this is not something you know this is not something where you have a nice procedure

191
00:12:03,370 --> 00:12:08,250
where we can say, Well, I can just adjust the correlation based on this new information.

192
00:12:08,490 --> 00:12:08,810
Yeah.

193
00:12:09,130 --> 00:12:12,490
There's nothing in the correlation that really tells you how you sh you're supposed to do

194
00:12:12,490 --> 00:12:12,650
that.

195
00:12:13,370 --> 00:12:18,970
Instead, if what's if your representation is a representation of the is a causal

196
00:12:18,970 --> 00:12:24,970
relationship between the rain and the and the the grass being wet, then you can

197
00:12:24,970 --> 00:12:29,850
incorporate this new information that there is a sprinkler in a much easier way.

198
00:12:30,400 --> 00:12:30,640
Yeah.

199
00:12:30,880 --> 00:12:35,680
It's like Paul Smaldino calls it uh being brittle, and he uses the example of science,

200
00:12:35,760 --> 00:12:36,000
right?

201
00:12:36,080 --> 00:12:39,840
So, like if you have this thing of like, I think we use the example, I think Noam Chomsky

202
00:12:39,920 --> 00:12:45,680
talks about this, is you could have a model, uh a correlational non-causal model of leaves

203
00:12:45,680 --> 00:12:47,920
falling, just describing the data.

204
00:12:48,000 --> 00:12:51,680
And that's what like a lot of science is, like, especially social science, it's just

205
00:12:51,680 --> 00:12:52,640
describing the data.

206
00:12:52,800 --> 00:12:55,840
But then that's not going to work for bowling balls, how they fall.

207
00:12:56,010 --> 00:12:58,090
Um, that's not going to work for a lot of other things.

208
00:12:58,170 --> 00:12:58,970
It's very brittle.

209
00:12:59,050 --> 00:13:00,330
It only applies to that context.

210
00:13:00,410 --> 00:13:03,850
And you don't even know what the context is because you don't have a theory, because you

211
00:13:03,850 --> 00:13:07,610
don't have like um coefficients of drag and accelerations of gravity.

212
00:13:07,690 --> 00:13:10,650
You just have the descriptions of the relationships.

213
00:13:10,810 --> 00:13:16,490
And so the whole replication crisis, or like the this effect holds in this context, but not

214
00:13:16,650 --> 00:13:20,820
another one, is the scientific equivalent of exactly what you're talking about.

215
00:13:20,980 --> 00:13:25,620
Um that it's this, as Paul Smaldino says, it's this brittle thing that as soon as you try

216
00:13:25,620 --> 00:13:29,700
to pick it up and move it, it's like a tower of blocks and it just falls apart.

217
00:13:30,580 --> 00:13:35,380
Yeah, and I mean to be clear, uh having a description is a a good starting point.

218
00:13:35,860 --> 00:13:41,830
Sometimes you sometimes you can't really start by having a nice theory, you need to gather

219
00:13:41,830 --> 00:13:42,870
a lot of data.

220
00:13:43,030 --> 00:13:48,150
Um, but it certainly cannot be the end goal of uh of a science.

221
00:13:48,310 --> 00:13:53,830
Um and yeah, and because we we should ideally be extracting the invariants, the things

222
00:13:53,910 --> 00:13:59,510
such that if uh if you look at a different context, you know that the invariants are still

223
00:13:59,510 --> 00:14:04,550
there and you understand how they work, um, and then you can make new new predictions.

224
00:14:07,660 --> 00:14:13,660
So I uh to me this is uh a lovely example of uh instinct blindness and

225
00:14:13,660 --> 00:14:16,460
and making the natural seem strange.

226
00:14:16,620 --> 00:14:22,060
Uh this idea that causality is not in the input itself, it's something that the mind

227
00:14:22,060 --> 00:14:26,780
constructs in the same way that you know uh the mental states of other people, there's

228
00:14:26,940 --> 00:14:32,830
there's nothing in you looking uh in one direction that necessarily says

229
00:14:32,830 --> 00:14:35,710
that you are that you know what's over in that direction, right?

230
00:14:35,790 --> 00:14:37,710
That's something that I'm projecting onto you.

231
00:14:37,870 --> 00:14:42,830
Uh, if you're moving really uh quickly toward a particular object, I'm projecting the fact

232
00:14:42,830 --> 00:14:44,270
that you want that object, right?

233
00:14:44,350 --> 00:14:45,870
That that's not in the data.

234
00:14:46,030 --> 00:14:49,950
Um, in the same way as you know, I'm picking up this coffee mug and drinking it, and when I

235
00:14:49,950 --> 00:14:55,680
put it down on the table, there's nothing in the sensory stimulation on the coffee cup that

236
00:14:55,680 --> 00:14:57,680
says it's my coffee cup, that it belongs to me.

237
00:14:57,840 --> 00:15:01,040
Ownership is another thing that we project onto the world.

238
00:15:01,200 --> 00:15:06,480
And what's uh lovely about your work is that it's we're doing the same thing with

239
00:15:06,480 --> 00:15:07,440
causality, right?

240
00:15:07,600 --> 00:15:10,080
And in some sense, we're constructing it.

241
00:15:10,320 --> 00:15:14,880
Um so to me, that leaves the question of how we are constructing.

242
00:15:15,120 --> 00:15:20,160
So, given that all we're getting is just uh, you know, photons hitting the retina,

243
00:15:20,320 --> 00:15:26,090
vibrations hitting our eardrums, how how do we take that information and

244
00:15:26,090 --> 00:15:29,770
construct causal relationships out of that data?

245
00:15:30,250 --> 00:15:30,410
Yeah.

246
00:15:30,490 --> 00:15:32,410
And can I add to that question really quick?

247
00:15:32,570 --> 00:15:34,090
What is it that we're constructing?

248
00:15:34,250 --> 00:15:35,050
Like what is it?

249
00:15:35,130 --> 00:15:38,090
Uh software, like what does it mean that we have a causal reper?

250
00:15:38,490 --> 00:15:39,050
That would be great.

251
00:15:39,130 --> 00:15:39,450
Thank you.

252
00:15:39,530 --> 00:15:39,770
Yeah.

253
00:15:39,850 --> 00:15:40,010
Yeah.

254
00:15:40,330 --> 00:15:40,650
Yeah.

255
00:15:40,730 --> 00:15:41,050
Yeah.

256
00:15:41,130 --> 00:15:45,610
And that's uh that's that's last question is the first question that we should ask,

257
00:15:45,770 --> 00:15:46,410
obviously.

258
00:15:46,650 --> 00:15:52,080
Um so I mean I I take it and and a lot of people who are working in the field of causal

259
00:15:52,080 --> 00:15:57,920
cognition take it that the target, the goal is to construct a causal model uh of

260
00:15:57,920 --> 00:16:02,640
the world, or at least a causal model of the particular system that we're interested in.

261
00:16:03,120 --> 00:16:08,960
And so this causal model uh will be made up of of variables uh and then the

262
00:16:08,960 --> 00:16:10,880
causal relationships between them.

263
00:16:11,200 --> 00:16:16,720
So um if I have a a causal model of of the rain then maybe

264
00:16:17,190 --> 00:16:20,710
one of my variables will be the presence of clouds in the sky.

265
00:16:21,190 --> 00:16:24,710
Another variable will be the presence of rain.

266
00:16:24,950 --> 00:16:29,270
And then we will have another variable that is is the grass wet, for example.

267
00:16:30,310 --> 00:16:34,550
And then we can express the causal relationships between them.

268
00:16:34,790 --> 00:16:40,310
So what is the color the causal relationships between the rain and the clouds?

269
00:16:40,470 --> 00:16:46,100
And then what is the causal relationships between the whether the grass is wet

270
00:16:46,260 --> 00:16:47,940
uh and and the rain.

271
00:16:48,340 --> 00:16:53,220
And there's various formalisms for how you might represent those causal relationships.

272
00:16:53,540 --> 00:16:56,740
And we don't need to get into the the details.

273
00:16:57,620 --> 00:17:03,300
But basically something that is quite um important and that is kind of foundational to the

274
00:17:03,300 --> 00:17:08,870
way that we think about causality is that the relationship between two variables is

275
00:17:08,870 --> 00:17:10,150
asymmetric, right?

276
00:17:10,310 --> 00:17:16,230
So it's not just that we know that there is an association between uh clouds and

277
00:17:16,230 --> 00:17:16,630
rain.

278
00:17:16,950 --> 00:17:21,830
It's just that there's this asymmetry that it is the clouds that are causing the rain and

279
00:17:21,830 --> 00:17:23,270
and not vice versa.

280
00:17:23,990 --> 00:17:29,960
Um and so one way to actually represent that pact concretely is by

281
00:17:30,520 --> 00:17:35,640
building into that the model this commitment that if we were to intervene on the on the

282
00:17:35,640 --> 00:17:39,400
clouds, then this will have a downstream effect on rain.

283
00:17:39,800 --> 00:17:45,400
So if by some if I had some magical powers and I was able to to

284
00:17:45,400 --> 00:17:49,480
prevent the clouds from forming, then there would not be any rain.

285
00:17:50,040 --> 00:17:55,550
But if I were to kind of uh if I were to build a giant dome over the city that

286
00:17:55,790 --> 00:17:59,870
shelters us from the rain, this would not make the clouds go away.

287
00:18:00,350 --> 00:18:06,030
And so that is the way that we kind of build in this important distinction between

288
00:18:06,350 --> 00:18:09,870
something like X causing Y and Y causing X.

289
00:18:10,670 --> 00:18:14,430
So just to we can keep going on, but just to clarify.

290
00:18:14,590 --> 00:18:18,910
So what I'm hearing is because it's hard to talk about this, because this is again where

291
00:18:18,910 --> 00:18:23,720
natural language fails us, is that if we snip out the word causal, if we're trying to

292
00:18:23,720 --> 00:18:27,640
explain what a causal relationship is and we snip out because we don't want to use the term

293
00:18:27,640 --> 00:18:28,760
in the explanation.

294
00:18:29,000 --> 00:18:34,920
It's a relationship that excuse me is um it's a relationship

295
00:18:34,920 --> 00:18:39,000
that has like directionality and then also

296
00:18:40,920 --> 00:18:46,830
it sort of picks out um what what is likely

297
00:18:46,830 --> 00:18:52,430
to if one thing is present and another thing will be present in sort of a compositional

298
00:18:52,590 --> 00:18:54,110
order-dependent way.

299
00:18:54,510 --> 00:19:00,350
So it's sort of saying, you know, like when this when this bit of the world is there.

300
00:19:01,230 --> 00:19:03,470
I don't even know how to talk about it, but yeah, right?

301
00:19:03,630 --> 00:19:04,830
Like, is that is that fair?

302
00:19:04,990 --> 00:19:08,270
Like it's sort of that, you know, this this is necessary.

303
00:19:08,840 --> 00:19:14,360
I don't even like using the word necessary, but it's just it's these placeholders for

304
00:19:15,080 --> 00:19:20,840
um you can't change the order, and then if you

305
00:19:21,080 --> 00:19:25,400
manipulate this thing, this other thing, they're kind of attached to each other in terms of

306
00:19:25,560 --> 00:19:26,040
how comes of the thing.

307
00:19:26,760 --> 00:19:30,280
This this notion of manip manipulability is quite important.

308
00:19:30,360 --> 00:19:33,960
So sometimes we talk about interventions of manipulate manipulability, etc.

309
00:19:34,280 --> 00:19:40,150
But it's this idea that if if someone could kind of reach into the system from the outside

310
00:19:40,630 --> 00:19:46,150
and change the state of change what happens to uh at some node in the

311
00:19:46,150 --> 00:19:49,510
network, then you will have particular effects somewhere.

312
00:19:49,830 --> 00:19:49,990
Right.

313
00:19:50,070 --> 00:19:53,590
I like that we not example yeah, I like the example of moon and tides.

314
00:19:53,750 --> 00:19:56,950
Like nobody's ever nobody's ever grabbed the moon.

315
00:19:57,110 --> 00:20:02,070
But it but if but the theory is the the causal theory of moon and tides is if somebody

316
00:20:02,070 --> 00:20:03,910
grabbed the moon and moved it, the tides would be different.

317
00:20:04,150 --> 00:20:08,960
Yes, and it's quite remarkable that we can entertain this despite the obvious imp

318
00:20:09,120 --> 00:20:11,680
impracticality of actually moving the moon.

319
00:20:12,080 --> 00:20:15,040
It's something that's quite easy to imagine to think about.

320
00:20:15,360 --> 00:20:20,560
And if you were to explain to someone how tides work, then that would be one way to to do

321
00:20:20,560 --> 00:20:20,640
that.

322
00:20:20,800 --> 00:20:25,280
You can say, well yes, if we move the moon then the tide would happen in a different way.

323
00:20:25,760 --> 00:20:30,720
And I guess you're the person that you're trying to teach to would not be bewildered at at

324
00:20:30,720 --> 00:20:31,360
the idea.

325
00:20:32,690 --> 00:20:33,010
I don't know.

326
00:20:33,090 --> 00:20:37,810
Maybe maybe in certain cultures this this is a bit strange to teach things in that way.

327
00:20:38,290 --> 00:20:41,170
But I guess the general yeah, the general idea is quite intuitive.

328
00:20:41,650 --> 00:20:46,850
So it seems it seems like what you're saying is that causality is ultimately about

329
00:20:46,850 --> 00:20:48,370
hypothetical actions.

330
00:20:48,770 --> 00:20:54,210
And that if I think A causes B, what that means is just uh I'm just translating that into

331
00:20:54,210 --> 00:20:56,130
the language of hypothetical things I can do.

332
00:20:56,370 --> 00:20:58,290
If I add A, I'm gonna get B.

333
00:20:58,530 --> 00:21:01,490
If I take away A, uh then B goes away.

334
00:21:01,740 --> 00:21:02,060
Right?

335
00:21:02,460 --> 00:21:07,420
These are hypothetical, but e even if I'm in incapable of taking away A or adding A, if I

336
00:21:07,500 --> 00:21:10,540
you know, if if I don't have magical powers, I can't do anything about the clouds.

337
00:21:10,620 --> 00:21:16,460
Uh merely the the ability to entertain uh the the possible action of adding or

338
00:21:16,460 --> 00:21:19,180
removing A will have an effect.

339
00:21:19,420 --> 00:21:21,900
I'm using up the word effect, but will change B.

340
00:21:22,220 --> 00:21:24,530
Um, that is ultimately what causality is.

341
00:21:24,610 --> 00:21:29,170
It's framed in terms of hypothetical interventions or actions or impacts you can have on

342
00:21:29,170 --> 00:21:33,010
the world, uh, and what is going to be changed in the world as a result.

343
00:21:33,330 --> 00:21:36,850
Again, I'm using the word as a result, which gets us back into causality.

344
00:21:37,010 --> 00:21:40,050
But what so it's a it's about hypothetical actions.

345
00:21:40,450 --> 00:21:40,770
Right?

346
00:21:41,090 --> 00:21:41,410
Yes.

347
00:21:41,650 --> 00:21:42,610
Uh yes, yes.

348
00:21:42,690 --> 00:21:45,330
That's uh it's a great way of explaining it.

349
00:21:45,490 --> 00:21:51,440
Um And I think that the the fact that we are we we

350
00:21:51,440 --> 00:21:55,840
are quite uh able to entertain those hypothetical actions.

351
00:21:56,080 --> 00:22:00,720
Um I think this is one of the best psychological evidence we have that people are

352
00:22:00,720 --> 00:22:02,640
constructing causal models of the world.

353
00:22:03,120 --> 00:22:08,880
They're not just passively trying to store patterns of correlations in in

354
00:22:09,670 --> 00:22:12,070
that that results from their experience, right?

355
00:22:12,310 --> 00:22:17,910
Because I have never experienced a state of the world where the moon has

356
00:22:18,150 --> 00:22:20,550
been kind of changed course.

357
00:22:21,510 --> 00:22:24,390
But I can imagine a manipulation that would do so.

358
00:22:24,870 --> 00:22:30,150
Maybe a better example is I have never imagined the state of the world where the sun

359
00:22:30,150 --> 00:22:31,750
suddenly disappears.

360
00:22:31,910 --> 00:22:37,420
But I can quite easily imagine what would happen if a sort of an action, a

361
00:22:37,420 --> 00:22:40,380
hypothetical action, was taken and the sun goes out.

362
00:22:41,580 --> 00:22:46,700
And this, I mean, if you are if you are looking at the mind from a sort of behaviorist

363
00:22:46,700 --> 00:22:52,620
perspective, where all you're doing is keeping track of associations, maybe keeping

364
00:22:52,620 --> 00:22:56,940
track of the results of your own actions, you would not really predict that people have the

365
00:22:56,940 --> 00:22:58,940
ability to engage in that kind of thinking.

366
00:22:59,340 --> 00:23:04,050
And you'd you don't you would even less predict the fact that people actually quite like

367
00:23:04,050 --> 00:23:05,810
thinking about those those sorts of things.

368
00:23:06,130 --> 00:23:06,290
Yeah.

369
00:23:06,530 --> 00:23:10,370
It's also a point like I I don't know who made this point, or maybe other whatever.

370
00:23:10,530 --> 00:23:15,890
Um, is uh the associationism never really talked about what is done with those

371
00:23:15,890 --> 00:23:17,090
representations either.

372
00:23:17,250 --> 00:23:21,570
Like it was very much uh um explain the data given the study.

373
00:23:21,810 --> 00:23:26,210
But but yeah, like the problem, like a good example also is the desert ant where you know

374
00:23:26,290 --> 00:23:31,180
it goes out, it has polar light sensitive thing, it does dead reckoning where it

375
00:23:31,980 --> 00:23:37,180
encodes distance by steps and uh angle of the sun as it moves.

376
00:23:37,340 --> 00:23:41,820
And then it does that integration when it finds prey and goes back to the nest.

377
00:23:42,620 --> 00:23:47,820
You know, if you just say, well, it's just associating, um, it's just associating stuff,

378
00:23:47,980 --> 00:23:51,100
like there's no notion of like what the association is for.

379
00:23:51,180 --> 00:23:54,700
But the second you have this thing of like, well, the whole point of doing the association

380
00:23:54,700 --> 00:24:00,590
is creating some representation of where the nest is, That's very much like a model, right?

381
00:24:00,750 --> 00:24:06,190
Um, that's integrating stuff for a purpose and it allows you to predict and go on a

382
00:24:06,190 --> 00:24:11,950
novel d dead point uh line to where the nest is so you don't fry

383
00:24:11,950 --> 00:24:14,990
yourself in the desert, which is why it's a selection pressure.

384
00:24:15,150 --> 00:24:20,940
But no, I like that that this is very evolutionary um um in

385
00:24:20,940 --> 00:24:26,380
a way is in is this causal model uh literature

386
00:24:26,620 --> 00:24:32,380
evolutionary and what does evolution buy you that may not be in the literature

387
00:24:32,540 --> 00:24:37,980
or or is it i don't know yeah is that a coherent yeah no that that that's that's a great

388
00:24:37,980 --> 00:24:43,910
question i i think um i would say that the causal cognition literature is implicitly

389
00:24:43,910 --> 00:24:44,950
evolutionary.

390
00:24:45,350 --> 00:24:47,590
So it definitely has a functionalist band.

391
00:24:48,150 --> 00:24:53,430
So people who look at causal cognition, they they ask, well no, what is the purpose of the

392
00:24:53,430 --> 00:24:53,670
system?

393
00:24:53,990 --> 00:24:55,830
Why why should we have those computations?

394
00:24:55,910 --> 00:25:00,070
And and it's really coming committed to this idea that people are building representations.

395
00:25:00,950 --> 00:25:06,390
And so uh there's also this uh very strong commitment to to optimality,

396
00:25:07,060 --> 00:25:10,660
at least as a as a sort of methodological heuristic.

397
00:25:11,140 --> 00:25:16,340
So because I mean once you have this machinery of causal models, you can ask, well,

398
00:25:16,900 --> 00:25:20,020
what is the best way to work with a causal model?

399
00:25:20,100 --> 00:25:25,940
So for example, you're trying to learn well, given all the data that I have,

400
00:25:26,340 --> 00:25:32,180
what is the probability that um the what is the probability that it is

401
00:25:32,180 --> 00:25:36,420
raining right now, given that all I can observe is that the grass is wet.

402
00:25:36,670 --> 00:25:42,190
And this is something that you can solve using the mathematics of uh of

403
00:25:42,510 --> 00:25:43,550
probability theory.

404
00:25:43,950 --> 00:25:46,590
So it's basically a Bayesian inference problem.

405
00:25:46,670 --> 00:25:52,590
So you have to set up um Bayes' theorem, and then you can work out what is the probability

406
00:25:52,590 --> 00:25:54,430
of a cause given the effect.

407
00:25:54,670 --> 00:26:00,480
Um and so we can look at whether people actually um Compute,

408
00:26:00,720 --> 00:26:02,640
engage in causal reasoning in that way.

409
00:26:03,120 --> 00:26:08,800
To what extent is it the case that when people sorry, when people make those

410
00:26:08,800 --> 00:26:13,600
inferences, there are actually reasoning in the normative way that is described by

411
00:26:13,760 --> 00:26:14,640
probability theory.

412
00:26:28,310 --> 00:26:30,550
have hypotheses that can be tested.

413
00:26:31,350 --> 00:26:36,790
And from my own point of view, as an evolutionary psychologist, I think that the

414
00:26:36,950 --> 00:26:42,870
another justification for the optimality analysis is that you should expect

415
00:26:43,030 --> 00:26:48,870
to have first order of approximation that people will solve those problems in uh in the

416
00:26:48,870 --> 00:26:53,830
normatively correct way because they have been a relatively high selection pressure for

417
00:26:53,830 --> 00:26:56,950
people to really um be good at causal reasoning.

418
00:26:57,800 --> 00:27:00,200
So that's the first part of the of the answer.

419
00:27:01,000 --> 00:27:06,840
The second part of the answer is that people who do causal recognition, they are

420
00:27:06,840 --> 00:27:12,840
often not um they're not quite concerned about domain specificity in the way

421
00:27:12,840 --> 00:27:15,160
that most of us would be concerned.

422
00:27:16,520 --> 00:27:21,560
So we're not going to to ask well no what are the the

423
00:27:22,200 --> 00:27:27,010
principles that people draw on when they think about biology, for example.

424
00:27:27,490 --> 00:27:33,250
And one, I guess, one motivation behind this choice is that there are

425
00:27:33,250 --> 00:27:37,010
general principles of causality that are supposed to apply across the board.

426
00:27:37,170 --> 00:27:43,010
And so this idea that causality is a matter of intervention, and that you need to

427
00:27:43,010 --> 00:27:47,170
engage in some sort of Bayesian reasoning in order to engage in causal reasoning, those are

428
00:27:47,170 --> 00:27:52,460
things that will apply regardless of whether you are thinking about psychological reasoning

429
00:27:52,460 --> 00:27:56,300
or biological biological reasoning or physical physical reasoning.

430
00:27:56,620 --> 00:28:02,460
And so I mean to us at least from the outside this can appear as

431
00:28:02,460 --> 00:28:04,300
something that is very domain general.

432
00:28:04,780 --> 00:28:08,060
But I would argue that there is a justification to that.

433
00:28:09,020 --> 00:28:14,940
Because you can think of the domain as being causality and you have certain

434
00:28:14,940 --> 00:28:20,110
invariants of uh of good causal reasoning that will apply across the board.

435
00:28:20,350 --> 00:28:20,590
Yeah.

436
00:28:20,990 --> 00:28:22,510
Yeah, don't yeah, yeah that's great.

437
00:28:22,590 --> 00:28:22,910
Yeah, yeah.

438
00:28:23,230 --> 00:28:24,590
It's I mean it's I mean it's interesting.

439
00:28:24,670 --> 00:28:30,590
If if um causality is ultimately about things I can do, like what will

440
00:28:30,590 --> 00:28:36,430
happen if I do this, uh well that's basically almost pretty much all of what the brain does

441
00:28:36,510 --> 00:28:38,350
is trying to figure out what I should do.

442
00:28:38,510 --> 00:28:39,470
What will happen if I do this.

443
00:28:39,630 --> 00:28:39,950
Yeah, you can't do it.

444
00:28:40,110 --> 00:28:42,350
Like it's a it's a machine for figuring out what to do.

445
00:28:42,510 --> 00:28:46,400
And it seems like causality should be literally a part of every system in the mind, because

446
00:28:46,400 --> 00:28:49,840
every system in the mind is ultimately about what should I do next, right?

447
00:28:50,000 --> 00:28:51,040
What's gonna happen if I do this?

448
00:28:51,200 --> 00:28:52,560
What's gonna happen if I do that?

449
00:28:52,720 --> 00:28:58,400
Um so I think you could think about it as you know ultra-domain general generality, but I

450
00:28:58,400 --> 00:28:59,840
don't think that's a really helpful way to think about it.

451
00:28:59,920 --> 00:29:02,480
I think it's it's better to think about in terms of functional specificity.

452
00:29:02,640 --> 00:29:06,880
We have systems that are functionally designed for the purpose of picking out causal

453
00:29:06,880 --> 00:29:07,760
relationships.

454
00:29:07,920 --> 00:29:13,890
Um, and that should by design affect every part of the mind that is involved in

455
00:29:13,890 --> 00:29:15,250
figuring out what to do next.

456
00:29:15,490 --> 00:29:15,650
Yeah.

457
00:29:15,730 --> 00:29:17,570
And and can I just like elaborate too?

458
00:29:17,650 --> 00:29:22,610
Like I think like domain generality is or just the whole issue of domain, like the word

459
00:29:22,610 --> 00:29:27,970
domain, is if you look at how people use it, I think it's exceedingly

460
00:29:28,210 --> 00:29:29,250
unclear.

461
00:29:29,410 --> 00:29:35,400
Um if you look at how Fodor uses it, it's very folk psychological, where like a domain is

462
00:29:35,400 --> 00:29:36,600
like a thing out in the world.

463
00:29:36,680 --> 00:29:39,160
Like, is it like cheese versus like cows?

464
00:29:39,320 --> 00:29:41,080
Right, like that's a domain.

465
00:29:41,640 --> 00:29:46,520
There's remarkably little like notion of like once you start thinking about the psychology,

466
00:29:46,680 --> 00:29:48,440
what do you mean by domain at all?

467
00:29:48,680 --> 00:29:54,570
And I would put it as like what David just described, which is exactly right, is

468
00:29:54,890 --> 00:29:59,210
you have all these systems for representing all the stuff out in the world.

469
00:29:59,610 --> 00:30:04,890
And all of those systems are all very specific for making all those representations

470
00:30:04,890 --> 00:30:10,010
possible while you're representing objects or the social world or physical relationships or

471
00:30:10,010 --> 00:30:13,690
emotion, mental states, like all these things in specialized machinery.

472
00:30:14,000 --> 00:30:20,000
once all those very specific systems are all working, they're also generating formats in

473
00:30:20,000 --> 00:30:25,840
this like event system that can take many of these inputs in because of all this very

474
00:30:25,840 --> 00:30:27,440
specific machinery working.

475
00:30:27,600 --> 00:30:33,200
And it's that event system being fed all of this content where

476
00:30:33,520 --> 00:30:39,410
that that has this rule, these rules of sort of like, um, you know, well, if I

477
00:30:39,410 --> 00:30:41,730
do this, this is the other thing that's going to come along with it.

478
00:30:41,890 --> 00:30:46,290
And then these other things don't matter, which applies to lots of relationships in the

479
00:30:46,290 --> 00:30:46,610
world.

480
00:30:46,770 --> 00:30:51,970
Yeah, that's common to all of them, but it's that there's more and more specialized

481
00:30:51,970 --> 00:30:52,450
machinery.

482
00:30:52,530 --> 00:30:54,290
There's even machinery for that job.

483
00:30:54,370 --> 00:30:57,090
I mean, it might get replicated across systems, who knows?

484
00:30:57,250 --> 00:31:00,930
But I think this is a thing where, like, if you say it's domain general in the sense of it

485
00:31:00,930 --> 00:31:05,660
applies to cows and cheese, it sort of sounds like, well, one mechanism does cows and

486
00:31:05,660 --> 00:31:10,060
cheese, but it's actually there's a ton of mechan, there's even more mechanisms than just

487
00:31:10,060 --> 00:31:12,940
if there was one mechanism for cows and a different mechanism for cheese.

488
00:31:13,020 --> 00:31:17,900
It's actually there's more mechanisms that allow something to be domain general in the in

489
00:31:17,900 --> 00:31:19,820
the unclear way than the domain specific way.

490
00:31:19,980 --> 00:31:23,980
I don't know if that's helpful for this, but I just always have this rant of like, my God,

491
00:31:24,140 --> 00:31:29,410
domain general versus domain specific is like it's it's I think it's even worse than innate

492
00:31:29,410 --> 00:31:32,210
and learned um or it's just as bad.

493
00:31:32,370 --> 00:31:33,330
Um yeah.

494
00:31:33,730 --> 00:31:36,050
Yeah yeah I mean yeah I I agree with that.

495
00:31:36,210 --> 00:31:42,130
Um I think to some extent this this was a thing that um it's

496
00:31:42,130 --> 00:31:48,050
a it's a nice it's a nice set of concepts to use if you're not already being

497
00:31:48,290 --> 00:31:52,160
uh in the in the process of thinking about things in uh in a functional way.

498
00:31:52,480 --> 00:31:53,200
Yes, hundred percent.

499
00:31:53,280 --> 00:31:54,960
It's kind of a gateway drug.

500
00:31:55,120 --> 00:31:56,960
Yeah, it's a gateway functionality.

501
00:31:58,000 --> 00:32:01,200
Where you can use people's intuitive understanding about the world 100%.

502
00:32:01,600 --> 00:32:02,000
Yeah, yeah.

503
00:32:03,200 --> 00:32:08,960
So uh you mentioned that um uh uh we uh have looked at how

504
00:32:09,270 --> 00:32:14,070
well people conform to normative standards of causal reasoning or Bayesian reasoning.

505
00:32:14,230 --> 00:32:18,310
And you mentioned that um there are some cases where we fall short of the normative

506
00:32:18,310 --> 00:32:18,790
standard.

507
00:32:19,030 --> 00:32:23,670
What what are some of those cases and and and what do you think might be going on there?

508
00:32:24,310 --> 00:32:25,190
Uh yeah.

509
00:32:25,350 --> 00:32:29,350
So there are many things we can could say about that.

510
00:32:29,430 --> 00:32:31,430
And this this could be its own discussion.

511
00:32:31,510 --> 00:32:35,830
But I think maybe I can talk about two case studies that I think are quite interesting.

512
00:32:35,990 --> 00:32:40,920
Um and so the first case study actually illustrates the fact that

513
00:32:41,720 --> 00:32:47,720
we are our brain is uh wired up to reason to represent the world in

514
00:32:47,720 --> 00:32:48,760
causal terms.

515
00:32:48,920 --> 00:32:54,840
Um so if you if you read psychology from the 70s, 80s, um there's a bit of

516
00:32:54,840 --> 00:32:59,800
a pessimistic picture that emerges about uh how good people are at at reasoning.

517
00:32:59,960 --> 00:33:03,400
Uh so I know there are studies by people like Tversky and Kahneman.

518
00:33:03,640 --> 00:33:09,390
Um, and one classic classic example of that is the um the diagnostic test problem.

519
00:33:10,030 --> 00:33:15,550
And so in that problem, you're told that there's this disease that maybe one in a thousand

520
00:33:15,870 --> 00:33:16,510
people have.

521
00:33:16,830 --> 00:33:19,550
And there's this test to see if you have the disease.

522
00:33:19,950 --> 00:33:25,550
And you're told that the test has a certain false positive rate and it has a certain false

523
00:33:25,550 --> 00:33:26,350
negative rate.

524
00:33:26,990 --> 00:33:32,140
And putting all this information together, you can in principle compute what is the

525
00:33:32,140 --> 00:33:37,180
probability that you have the disease, given that you were tested positive to the to the

526
00:33:37,180 --> 00:33:37,740
disease.

527
00:33:37,900 --> 00:33:43,020
Um and when you ask participants uh you know what is this probability, um

528
00:33:43,340 --> 00:33:46,380
typically they will say something like oh like 80%.

529
00:33:47,180 --> 00:33:52,570
They will say something pretty high, even if the numbers are set up in such a way that the

530
00:33:52,570 --> 00:33:55,130
answer is actually something like seven percent.

531
00:33:55,690 --> 00:34:01,050
And so that is a huge gap between the normative Bayesian answer and the correct answer.

532
00:34:01,210 --> 00:34:05,770
Um and so when you look at that, you can maybe conclude, well, it looks like people are

533
00:34:05,770 --> 00:34:08,250
actually terrible at probabilistic reasoning.

534
00:34:08,410 --> 00:34:14,330
Um, but then if you look at the way this problem is set up, you are never really given

535
00:34:14,330 --> 00:34:17,370
an explanation for why there might be false positives.

536
00:34:17,620 --> 00:34:21,380
You kind of just told that well, sometimes the test doesn't work very well.

537
00:34:22,260 --> 00:34:27,700
And so it is difficult to kind of for the participants to construct a causal model of the

538
00:34:27,700 --> 00:34:29,620
situation that they have to reason about.

539
00:34:30,180 --> 00:34:35,940
And so one hypothesis that that is just the explanation for why this is a

540
00:34:35,940 --> 00:34:36,740
hard problem.

541
00:34:37,380 --> 00:34:43,150
And so some psychologists have done an experiment where modify this

542
00:34:43,150 --> 00:34:47,790
problem in a slight way, and the only thing that they do is that they say, well, sometimes

543
00:34:47,790 --> 00:34:53,470
people, you know, like I don't know, maybe 70% of people have a benign uh keist

544
00:34:54,110 --> 00:34:56,750
that will cause a false positive.

545
00:34:57,070 --> 00:35:01,150
And when you reformulate this problem in that way, then people suddenly they they have a

546
00:35:01,150 --> 00:35:05,950
sort of a causal explanation that is available to account for the presence of false

547
00:35:05,950 --> 00:35:06,430
positives.

548
00:35:06,590 --> 00:35:10,030
And now their answers get much closer to the Bayesian optimum.

549
00:35:10,720 --> 00:35:16,720
And so it seems that people are able to engage in in good Bayesian reasoning, but they

550
00:35:16,720 --> 00:35:19,440
need to form a causal representation in order to do that.

551
00:35:20,000 --> 00:35:20,880
And so yeah.

552
00:35:21,200 --> 00:35:26,800
Yeah, sorry to in that example too, just really quick, I think there's also additional work

553
00:35:26,800 --> 00:35:31,920
showing that if you do this with frequencies, so you show people this is the number of

554
00:35:31,920 --> 00:35:36,690
people who have false positives, this is the number of people who have the correct hits, it

555
00:35:36,690 --> 00:35:42,050
also goes away in the sense of the Tversky and Kahneman results was actually very

556
00:35:42,050 --> 00:35:44,610
improbable and it was really designed.

557
00:35:45,170 --> 00:35:51,010
My friend and colleague Tomas Leharaga has a paper in Psych Bulletin about how people used

558
00:35:51,010 --> 00:35:56,770
to see people say that people were good statisticians, and Tversky Kahneman reversed that,

559
00:35:56,930 --> 00:36:02,290
but they were using very abstract, uh non-evolutionarily

560
00:36:02,500 --> 00:36:07,780
valid sort of cues of like, you know, not using whole objects or frequencies and sort of

561
00:36:07,780 --> 00:36:09,940
giving people uh counterintuitive things.

562
00:36:10,100 --> 00:36:14,980
So I just want to say like that's really interesting that even just changing the uh adding

563
00:36:14,980 --> 00:36:19,220
the causal model is like another problem with that methodology.

564
00:36:19,300 --> 00:36:20,100
So yeah, yeah.

565
00:36:20,660 --> 00:36:26,590
Um I mean and and to some extent I I think that First Ken Ganneman um they still did

566
00:36:26,590 --> 00:36:32,030
a service to the field by pointing out that there are those cases where people do really

567
00:36:32,110 --> 00:36:33,150
really badly, right?

568
00:36:33,230 --> 00:36:36,110
Because then see something that is crying for explanation.

569
00:36:36,350 --> 00:36:41,550
And then when we explain why it is it that people do very badly, we learn new things about

570
00:36:41,550 --> 00:36:42,910
how the mind works.

571
00:36:51,260 --> 00:36:56,220
principle, it's almost um well, it is actually an axiom of causal reasoning that is called

572
00:36:56,220 --> 00:36:57,740
the the Markov condition.

573
00:36:57,820 --> 00:37:03,500
So it's named after this mathematician uh from the early 20th century called

574
00:37:03,500 --> 00:37:09,340
Andrei Markov, who studied very simple systems where uh the state of

575
00:37:09,340 --> 00:37:15,180
the system in uh in a particular uh time just depends on the previous state of

576
00:37:15,180 --> 00:37:15,740
the system.

577
00:37:16,370 --> 00:37:21,570
And this is relevant to causality because if you have a causal system where

578
00:37:22,050 --> 00:37:27,490
suppose you have the again the clouds cause the rain and the rain causes the grass to be

579
00:37:27,490 --> 00:37:27,890
wet.

580
00:37:28,930 --> 00:37:34,850
If that is all there is to your causal system, then the clouds

581
00:37:34,850 --> 00:37:37,090
can have an influence on the grass being wet.

582
00:37:37,560 --> 00:37:40,840
So if if there are clouds, the the grass are more likely to be wet.

583
00:37:41,240 --> 00:37:47,080
But this the whole causal influence of the clouds can only go uh via

584
00:37:47,160 --> 00:37:48,120
the the rain.

585
00:37:48,280 --> 00:37:53,880
Okay so the the the only reason that the clouds can make the rain wet is can make the the

586
00:37:53,880 --> 00:37:56,920
grass wet is is via the is by making it rain.

587
00:37:57,560 --> 00:38:03,450
And so there's um from um like from from the sort of perspective of

588
00:38:03,450 --> 00:38:09,130
of causal reasoning um then the the Markov condition is just a fancy way of saying that

589
00:38:09,290 --> 00:38:15,210
once you know whether it has rained or not, then anything that you learn about whether the

590
00:38:15,210 --> 00:38:19,930
clouds are there or not should not give you any more information about whether the grass is

591
00:38:19,930 --> 00:38:20,170
wet.

592
00:38:21,210 --> 00:38:25,370
You got the most important piece of information you get all the information right.

593
00:38:25,610 --> 00:38:31,320
And sometimes we say that the the rain screens off the influence of the clouds.

594
00:38:31,880 --> 00:38:36,600
And so what this means is that if you asked well, what is the probability that the grass is

595
00:38:36,600 --> 00:38:41,640
wet, given that we know that it's it's been raining, you should give some answer, right?

596
00:38:41,960 --> 00:38:47,000
And then if later on the experimenter asks you, Well, what is the probability that the

597
00:38:47,000 --> 00:38:51,150
grass is wet given that it has rain and there are clouds.

598
00:38:51,310 --> 00:38:53,470
You should give the exact same answer.

599
00:38:55,150 --> 00:38:56,910
But people don't really do that.

600
00:38:57,310 --> 00:39:02,350
People will give a higher probability if you tell them that, in addition to the fact that

601
00:39:02,350 --> 00:39:05,230
there is rain, there are also clouds.

602
00:39:05,710 --> 00:39:11,470
So it seems like a blatant violation of the rules of how you should reason with causal

603
00:39:11,470 --> 00:39:12,110
networks.

604
00:39:12,560 --> 00:39:16,160
And so this has gotten people in the field quite interested because

605
00:39:21,680 --> 00:39:26,160
is consistent with the fact that people are actually reasoning with causal models.

606
00:39:35,070 --> 00:39:40,830
when they need to engage in probabilistic reasoning, um they make uh use of approximations.

607
00:39:41,470 --> 00:39:47,150
Um and the the reason why you need to engage in approximations is that

608
00:39:47,150 --> 00:39:52,750
computing a probability will in general be relatively um expensive in computational

609
00:39:52,990 --> 00:39:53,630
terms.

610
00:39:53,950 --> 00:39:59,950
Um right you need to uh you need to engage in quite complex uh computations um and

611
00:39:59,950 --> 00:40:05,900
there are actually shortcuts that people who do computational statistics and

612
00:40:05,900 --> 00:40:09,980
machine learning typically use in order to make the computations easier.

613
00:40:10,140 --> 00:40:15,980
And it turns out that if you assume that people are engaging in the same kind of shortcuts,

614
00:40:16,780 --> 00:40:20,540
then you predict that they will make that kind of mistake.

615
00:40:20,780 --> 00:40:25,580
And so the picture that emerges from that is not that people are don't know the rules of

616
00:40:25,580 --> 00:40:30,380
causal models, but they they use some approximations in order to compute things in an

617
00:40:30,380 --> 00:40:31,500
efficient way.

618
00:40:32,990 --> 00:40:35,310
I'm I'm a little bit confused.

619
00:40:35,710 --> 00:40:40,590
I mean I hear what you're saying, but let me just let me be devil's advocate for the um so

620
00:40:40,670 --> 00:40:45,310
like with the rain and the clouds, is the is the study designed that like you tell the same

621
00:40:45,310 --> 00:40:49,870
person rain and then the same person rain in clouds and which one do you think is more

622
00:40:49,870 --> 00:40:51,790
likely that the grass is wet?

623
00:40:52,830 --> 00:40:56,670
Um so there are various ways that the experiment has been done.

624
00:40:56,750 --> 00:41:01,380
And of course uh when people do those experiments, they they use a more controlled setup

625
00:41:01,460 --> 00:41:04,900
where you say, well, there's this machine, and then there's this component of the machine

626
00:41:04,980 --> 00:41:08,260
that can influence this component that then can influence that component.

627
00:41:08,980 --> 00:41:14,740
But why is that so why is it wrong for me to be more confidently

628
00:41:14,980 --> 00:41:20,580
grass is wet if I say uh clouds uh and rain as opposed to rain?

629
00:41:21,780 --> 00:41:26,510
Um so if I ask you, you know, suppose you know that there is rain, yeah, what is the

630
00:41:26,510 --> 00:41:28,110
probability that the grass is wet?

631
00:41:28,270 --> 00:41:34,110
Um, then the the answer you give to that should be the same as if I ask you, suppose you

632
00:41:34,110 --> 00:41:36,270
know that there is rain and there are clouds.

633
00:41:36,590 --> 00:41:42,270
Yeah, but I but I would push back on that because like that ignores that like is it's very

634
00:41:42,350 --> 00:41:46,750
like um uh I mean I don't know, I'm not saying you're doing this, but I'm saying like the

635
00:41:46,750 --> 00:41:50,590
causal model people, like there's a difference between like there's a little bit of rain

636
00:41:50,590 --> 00:41:52,430
versus like there's a storm, right?

637
00:41:52,590 --> 00:41:56,480
So like there's differing I mean, in some sense, people Yeah.

638
00:41:56,640 --> 00:42:00,880
I mean, just to take a step back, like causality doesn't exist.

639
00:42:01,360 --> 00:42:06,880
It's a representation of the mind of like how certain am I likely that this event will

640
00:42:06,880 --> 00:42:08,080
produce this event?

641
00:42:08,320 --> 00:42:14,240
And if you tell me rain versus rain and clouds, um, how likely is

642
00:42:14,240 --> 00:42:17,040
it that the r the water, the grass is drenched?

643
00:42:17,230 --> 00:42:18,430
how much water is there?

644
00:42:18,590 --> 00:42:22,910
How likely is it that that was rain and not like a bird peeing or something?

645
00:42:23,070 --> 00:42:25,550
You know, like that's r all relevant.

646
00:42:25,790 --> 00:42:30,350
And if you tell me like there there's a there was a little bit of rain versus like there's

647
00:42:30,430 --> 00:42:36,430
rain in clouds, like you just if there's all this common sense knowledge that I

648
00:42:36,430 --> 00:42:42,360
think a smart computer would say it's much more likely that whatever

649
00:42:42,360 --> 00:42:47,160
you're trying to communicate of like how likely is it the grass is wet, how wet it got, how

650
00:42:47,160 --> 00:42:49,960
confident you were that like, you know, it was a rainstorm.

651
00:42:50,280 --> 00:42:51,160
All these things make sense.

652
00:42:51,480 --> 00:42:51,640
Yeah.

653
00:42:51,720 --> 00:42:55,720
Or or even like the person who's telling you that it's raining without clouds, like maybe

654
00:42:55,720 --> 00:42:56,760
you should not trust that person.

655
00:42:57,000 --> 00:42:57,640
Maybe they were wrong.

656
00:42:57,800 --> 00:43:01,480
Like if you've ever spelt rain when there's no clouds, you're like, what is this?

657
00:43:01,640 --> 00:43:02,680
And it does happen, right?

658
00:43:02,760 --> 00:43:04,200
Because like the wind is strong.

659
00:43:04,440 --> 00:43:09,910
So like I I I don't I mean maybe maybe we're we're hanging a lot on this example, but like

660
00:43:10,550 --> 00:43:12,710
it seems what you're saying scientists are wrong.

661
00:43:13,910 --> 00:43:18,790
I think that those are those are good objections to sort of the experiment in the way that

662
00:43:18,790 --> 00:43:19,350
I described it.

663
00:43:19,590 --> 00:43:19,910
Yeah, exactly.

664
00:43:20,230 --> 00:43:24,470
Um of course in in in practice when people do those experiments you have a much more

665
00:43:24,470 --> 00:43:28,870
controlled setting where you really make it clear to people that well those are just binary

666
00:43:28,870 --> 00:43:34,620
variables and really the only way that this like the the end end variable,

667
00:43:34,860 --> 00:43:39,340
the only way that this could be affected is by the state of the previous variable.

668
00:43:39,500 --> 00:43:44,220
And so that way you you bypass that kind of concern about well what inferences people are

669
00:43:44,220 --> 00:43:47,980
making from the fact that you tell them many things and so on and so forth.

670
00:43:48,300 --> 00:43:53,340
But isn't it still rational in the sense of like shouldn't an evolved psychology represent

671
00:43:53,470 --> 00:43:58,110
because, like, I mean, from an evolved psychology perspective, it should acknowledge that

672
00:43:58,110 --> 00:44:02,030
all it is, all it can do is pick up on correlations, right?

673
00:44:02,190 --> 00:44:07,950
Like, like the evolved psychology is is only ever

674
00:44:08,110 --> 00:44:14,110
detecting co occurrences and is using the fact that some co occurrences

675
00:44:14,750 --> 00:44:20,500
are actionable for intervention and actually generally produce the right

676
00:44:20,500 --> 00:44:21,540
outcome, right?

677
00:44:21,860 --> 00:44:26,820
Like, so that's what we mean by causation, but it's all a guess.

678
00:44:27,220 --> 00:44:32,260
Um so I it strikes me as like a I mean maybe this is I maybe I should get more concrete,

679
00:44:32,340 --> 00:44:32,820
but I don't know.

680
00:44:32,900 --> 00:44:37,140
I just I feel like the the scientists are wrong in this case and that it's a smart system

681
00:44:37,300 --> 00:44:43,230
to to be uh if more if more predictors are present, you should

682
00:44:43,230 --> 00:44:44,350
increase your likelihood.

683
00:44:44,670 --> 00:44:44,830
Yeah.

684
00:44:45,070 --> 00:44:50,910
I mean I have I have a a version of arguments that I've made in in print, which is that um

685
00:44:51,390 --> 00:44:56,990
so I mean I I buy the this idea that I described that kind of if people engage with in in

686
00:44:56,990 --> 00:45:00,830
shortcuts in their computations, then they will make the kind of mistakes that have been

687
00:45:00,830 --> 00:45:01,470
observed.

688
00:45:01,630 --> 00:45:07,400
Um but a different way that those kinds of mistakes will be produced is that

689
00:45:07,400 --> 00:45:11,800
if people don't completely buy the story that has been given to them by the experimenter.

690
00:45:11,960 --> 00:45:15,800
So maybe they think, well, you know, the experimenter tells me what those variables are

691
00:45:15,800 --> 00:45:20,040
causally related, but I don't know, maybe they're not, or maybe they're related but in a

692
00:45:20,040 --> 00:45:22,920
different way than has been told to me.

693
00:45:23,160 --> 00:45:28,920
And so if that's what you think, then the smart move is to just maintain those different

694
00:45:28,920 --> 00:45:32,360
hypotheses in mind and then and then to average over them.

695
00:45:34,650 --> 00:45:38,090
So you're not you're telling me that people aren't assuming that experimenters and

696
00:45:38,090 --> 00:45:41,210
psychology experiments are infallible and uh omniscient.

697
00:45:41,450 --> 00:45:43,610
What if you let people do the experiment themselves?

698
00:45:43,690 --> 00:45:44,810
Like, has anybody ever done that?

699
00:45:44,890 --> 00:45:48,650
Where like you actually give people the ability to actually test the relationships

700
00:45:48,650 --> 00:45:49,130
themselves?

701
00:45:49,290 --> 00:45:54,650
Like the experience, the experiential format of teasing apart what causes what, as opposed

702
00:45:54,650 --> 00:45:56,440
to somebody telling you, because actually, thank you.

703
00:45:56,520 --> 00:45:59,160
I f I felt like I was humming and hawing, but that's exactly what I was doing.

704
00:45:59,240 --> 00:46:00,680
I was like, I don't believe you.

705
00:46:00,840 --> 00:46:02,120
That's exactly what I was doing.

706
00:46:02,600 --> 00:46:06,680
And and and what what we what we've have shown in in that paper is that well if you if you

707
00:46:06,680 --> 00:46:11,400
assume that people have this uncertainty and that they are averaging over that uncertainty,

708
00:46:11,560 --> 00:46:17,490
then you predict that the that they will make the kind of quote unquote mistakes

709
00:46:17,490 --> 00:46:19,650
that have been observed in these experiments.

710
00:46:20,050 --> 00:46:24,930
But to answer your question, I don't think that that experiment has been run where you

711
00:46:24,930 --> 00:46:30,370
really allow participants to manipulate the causal structure themselves.

712
00:46:30,530 --> 00:46:34,530
That's it, yeah, it would be interesting to perform that experiment, that's for sure.

713
00:46:34,850 --> 00:46:35,490
Yeah.

714
00:46:35,970 --> 00:46:41,360
So is your take that um this idea that humans are bad at causal reasoning or they're

715
00:46:41,360 --> 00:46:44,800
irrational, uh, that we deviate from normative standards.

716
00:46:44,880 --> 00:46:49,760
Is your take that this idea is overblown and that actually, if you dig into the details and

717
00:46:49,760 --> 00:46:54,080
operationalize things in a sensible way and in an ecologically valid way, that actually

718
00:46:54,080 --> 00:46:58,080
people are pretty good at causal reasoning and and and fit pretty closely to what you might

719
00:46:58,080 --> 00:47:00,880
expect uh an optimal system to look like?

720
00:47:02,800 --> 00:47:03,440
Yeah, yeah.

721
00:47:03,600 --> 00:47:07,550
And uh I think that's a good summary of of what I think.

722
00:47:07,790 --> 00:47:13,630
Um and if you were to try to to come up with a taxonomy of

723
00:47:13,630 --> 00:47:19,150
possible sources of mistakes that people make, um I would say that there's maybe

724
00:47:19,390 --> 00:47:24,750
two main sources of mistakes that people can for why people might make mistakes.

725
00:47:24,910 --> 00:47:30,510
Um and the first is this kind of representational issue where people just either don't

726
00:47:30,510 --> 00:47:36,320
really believe you or they don't uh they're not able to construct a causal

727
00:47:36,320 --> 00:47:42,160
model that's that would actually allow them to to solve the the to solve the

728
00:47:42,160 --> 00:47:43,360
statistical problem.

729
00:47:43,520 --> 00:47:47,520
Um there's a different way that they could be wrong is that if the problem is super

730
00:47:47,520 --> 00:47:52,480
complicated then then of course you I mean it's it's kind of it's a bit of a trivial point

731
00:47:52,560 --> 00:47:52,960
in a way.

732
00:47:53,120 --> 00:47:58,320
It's like um asking people if I if I ask you you know what is the 17th digit of pi

733
00:47:58,990 --> 00:48:03,550
and you give me an answer that is no better than chance, there's a sense in which you're

734
00:48:03,550 --> 00:48:04,510
not being normative.

735
00:48:05,150 --> 00:48:08,910
But it's just that the computational demands that I put on you are too high.

736
00:48:09,070 --> 00:48:11,790
And so that that is one way that that you can make mistakes.

737
00:48:12,510 --> 00:48:18,110
And I mean one one other domain in which I have shown uh that people are relatively

738
00:48:18,270 --> 00:48:24,110
quite they're quite good at following the normative principles of probabilistic inference

739
00:48:24,350 --> 00:48:27,200
is um to in the social domain.

740
00:48:27,360 --> 00:48:32,480
So when they have to infer the welfare trade-off ratio of other people, then they have to

741
00:48:32,480 --> 00:48:35,280
solve a problem that is actually quite sophisticated.

742
00:48:35,600 --> 00:48:40,400
If you were to think of, you know, they have to make probabilistic computations over a

743
00:48:40,560 --> 00:48:43,920
causal model of how other people make decisions.

744
00:48:44,800 --> 00:48:50,800
And yes, despite the complexity of the problem, they're actually quite good at

745
00:48:50,880 --> 00:48:54,000
uh inferring how much someone else values them.

746
00:48:54,170 --> 00:49:00,090
So this is a a series of experiments that we've done with uh with John and Lida, um, where

747
00:49:00,730 --> 00:49:06,730
the simplified version of this experiment is that you see someone kind of uh sharing a

748
00:49:06,730 --> 00:49:12,250
pot of money uh in a particular context uh and you see the decision that they have made.

749
00:49:12,330 --> 00:49:17,690
Uh and then there's a slightly different context where maybe the cost to them of giving you

750
00:49:17,690 --> 00:49:19,130
money is a bit higher.

751
00:49:19,380 --> 00:49:21,300
And I um and I asked.

752
00:49:25,300 --> 00:49:26,020
Something, yeah, yeah.

753
00:49:26,420 --> 00:49:27,220
It's a good picture.

754
00:49:27,700 --> 00:49:31,380
And so in that in that context, so you know what they have done in that previous situation.

755
00:49:31,540 --> 00:49:36,020
And I ask you, what is the probability that they will give the food to you?

756
00:49:36,660 --> 00:49:42,500
And so you you can compute what is the like given some assumptions, what

757
00:49:42,500 --> 00:49:47,190
is the like the normative probability that uh a Bayesian model would give.

758
00:49:47,350 --> 00:49:52,790
And what we find is that first people are quite good, the people's answers are quite

759
00:49:52,790 --> 00:49:56,470
aligned with the answers that are given by the Bayesian model.

760
00:49:57,270 --> 00:50:03,270
And the second cool thing that we have found is that we also asked people well

761
00:50:03,510 --> 00:50:08,710
how grateful are you when the person actually because you actually at the end you you learn

762
00:50:08,710 --> 00:50:10,230
what they have what they have done.

763
00:50:10,390 --> 00:50:15,840
And so if the if the person in the experiment gives you uh gives you something.

764
00:50:16,000 --> 00:50:19,360
We ask people, well how grateful are you that you are given that thing?

765
00:50:19,520 --> 00:50:24,080
And if they decide to take the selfish action, we ask, well how angry do you feel?

766
00:50:24,400 --> 00:50:29,760
And then what we found is that the best predictor of how angry or grateful the people feel

767
00:50:29,920 --> 00:50:34,000
is not the best predictor is not how much they get or how much they fail to get.

768
00:50:34,560 --> 00:50:36,480
Because we control for those things tightly.

769
00:50:36,560 --> 00:50:42,160
So like the the kind of the payoffs that are involved or controlled across trials um

770
00:50:42,400 --> 00:50:45,630
and the sort of opportunity cost is also controlled.

771
00:50:45,950 --> 00:50:51,870
But what we can do is we can actually compute the welfare trade-off ratio that

772
00:50:51,870 --> 00:50:57,710
an ideal Bayesian observer would infer the other person has on the basis of the decisions

773
00:50:57,710 --> 00:50:58,670
that you observe.

774
00:50:59,230 --> 00:51:03,710
And then this inferred welfare trade-off ratio is actually quite a good predictor of

775
00:51:03,710 --> 00:51:05,070
gratitude and anger.

776
00:51:05,380 --> 00:51:11,220
So basically if you infer that if even if a if an uh an ideal Bayesian observer infers that

777
00:51:11,300 --> 00:51:15,620
yes, this person does not care at all about me, then I'm going to be quite hang angry.

778
00:51:16,020 --> 00:51:18,020
Even if you end up getting a lot.

779
00:51:18,740 --> 00:51:19,300
Yes, yeah.

780
00:51:19,700 --> 00:51:19,860
Yeah.

781
00:51:20,180 --> 00:51:23,940
But yeah, so it's kind of the the intuition is that suppose at some point they give me a

782
00:51:23,940 --> 00:51:26,740
lot, but because it was trivial for them to give me.

783
00:51:26,900 --> 00:51:27,060
Yeah.

784
00:51:27,940 --> 00:51:30,100
The cost of was super super low.

785
00:51:30,270 --> 00:51:33,310
And then they were gonna throw it in the garbage or whatever, and you just happen to be

786
00:51:33,310 --> 00:51:33,390
there.

787
00:51:33,470 --> 00:51:33,870
Yeah, whatever.

788
00:51:34,030 --> 00:51:34,350
Yeah, exactly.

789
00:51:34,430 --> 00:51:34,910
Yeah, exactly.

790
00:51:35,230 --> 00:51:40,830
And then on the other occasion, they kind of they could have benefited me by a little bit,

791
00:51:40,990 --> 00:51:44,270
uh, at again a trivial cost to themselves.

792
00:51:44,430 --> 00:51:48,030
Um, and they decide, oh yeah, you know what, I don't really feel like giving it to you.

793
00:51:48,190 --> 00:51:52,190
I'm going to take the two cents or whatever, instead of giving you ten dollars.

794
00:51:52,430 --> 00:51:58,080
Uh, and here, you know, you just lost ten dollars, it's not that much, but like, ah, you

795
00:51:58,080 --> 00:52:00,480
did it just to save two cents.

796
00:52:00,640 --> 00:52:03,040
Like that, that's you just you just don't care about me at all.

797
00:52:03,200 --> 00:52:05,200
And so that is what elicits anger.

798
00:52:05,600 --> 00:52:10,480
And of course, this is not this is not my own theory, this is the work of uh of Aaron Cell.

799
00:52:10,720 --> 00:52:16,000
But what we were able to show in that paper is actually you can uh create a computational

800
00:52:16,000 --> 00:52:21,830
account of of how anger uh follows from people's

801
00:52:21,830 --> 00:52:25,830
from observations of of what the other person uh has done.

802
00:52:26,310 --> 00:52:28,150
What what does that mean?

803
00:52:28,390 --> 00:52:31,750
I mean, I think people might have two questions, and this is both hard.

804
00:52:31,910 --> 00:52:34,950
Um we we we you know you and I have talked about this today.

805
00:52:35,030 --> 00:52:41,030
Like what does it mean to offer a computational um account because you straddle two

806
00:52:41,030 --> 00:52:46,940
communities we've talked a lot about functionalism or like a software level description Um,

807
00:52:47,100 --> 00:52:52,940
but then there's also like a technical methodological use of the term

808
00:52:52,940 --> 00:52:54,300
computationalism.

809
00:52:54,620 --> 00:52:59,580
Uh and then there's like computers, like a laptop, like can for people, because there's

810
00:52:59,580 --> 00:53:04,140
different audiences, can you explain what what does that mean, both to like a general

811
00:53:04,140 --> 00:53:09,260
audience, but also like evolutionary psychologists who might be used to like functionalism,

812
00:53:09,580 --> 00:53:10,300
functionalism.

813
00:53:10,380 --> 00:53:12,060
Yeah, what does that mean, computational?

814
00:53:12,430 --> 00:53:13,230
Yeah, yeah.

815
00:53:13,390 --> 00:53:16,030
Um that's I think a very important point.

816
00:53:16,190 --> 00:53:20,910
And it's a bit of an unfortunate fact about the history of psychology that you have just

817
00:53:21,150 --> 00:53:23,230
two different meanings of computational.

818
00:53:24,030 --> 00:53:29,550
So the first meaning of computational is maybe the one that will be most familiar to people

819
00:53:29,550 --> 00:53:30,830
across the sciences.

820
00:53:30,910 --> 00:53:35,710
Um so it's the idea that we are going to use uh we're going to use mathematics or we're

821
00:53:35,710 --> 00:53:39,070
going to use computers in order to to help us do our science.

822
00:53:39,700 --> 00:53:45,460
And so within that meaning of computational, uh, you are engaging in in

823
00:53:45,460 --> 00:53:50,180
computational work every time you compute a linear regression, or you're when you compute a

824
00:53:50,180 --> 00:53:54,580
t-test or something like that, because you're using a computer, you're using mathematical

825
00:53:54,580 --> 00:53:56,900
principles in order to make inferences.

826
00:53:57,220 --> 00:53:57,380
Right.

827
00:53:57,540 --> 00:54:03,410
It was sort of just standard, it's just sort of like rules of deduction and just

828
00:54:03,410 --> 00:54:08,210
sort of force like you know if you know what the inputs are just some deterministic process

829
00:54:08,370 --> 00:54:13,890
produces some outputs that's that's compute that that's that you know right and it's just

830
00:54:13,890 --> 00:54:18,050
certain ones work for whatever reason you don't have to do it on computer to you can also

831
00:54:18,290 --> 00:54:20,690
exactly you can you could compute your T test on on paper.

832
00:54:20,850 --> 00:54:26,290
Yeah exactly dinosaurs roam the earth yeah and to do that.

833
00:54:26,370 --> 00:54:28,290
I see the I don't remember anything from it.

834
00:54:28,450 --> 00:54:34,400
Yes exactly sorry yes Uh I wrote a horse to the schoolhouse and played with

835
00:54:34,400 --> 00:54:35,120
an advocate.

836
00:54:35,200 --> 00:54:35,520
Yeah, exactly.

837
00:54:35,760 --> 00:54:40,080
When there was it to CSB, actually we had to compute matrix multiplications by hand in the

838
00:54:40,080 --> 00:54:40,080
case.

839
00:54:40,240 --> 00:54:42,480
Yeah, we it's probably the Greg Ashby.

840
00:54:47,040 --> 00:54:47,280
Okay, yeah.

841
00:54:47,440 --> 00:54:49,120
Uh for that one semester because he wasn't live.

842
00:54:49,360 --> 00:54:52,800
But yeah, I mean and we'll we were giving given no explanations for why it was.

843
00:54:53,120 --> 00:54:53,360
Exactly.

844
00:54:53,440 --> 00:54:55,040
I did the same thing, yes, exactly.

845
00:54:55,360 --> 00:54:56,590
Um anyway.

846
00:54:56,750 --> 00:55:02,350
But yeah, so that's the everyone understands computation in this sort of sort of way of

847
00:55:02,350 --> 00:55:08,350
using either math or for more complicated systems, uh computer code, but it's still

848
00:55:08,350 --> 00:55:14,190
the same principle where it's just deterministic rules, um, it's just it's just deduction

849
00:55:14,190 --> 00:55:17,550
in sort of a scripted way that is sort of known.

850
00:55:17,630 --> 00:55:17,870
Yeah.

851
00:55:18,190 --> 00:55:18,430
Yeah.

852
00:55:18,590 --> 00:55:24,460
And obviously this is a super powerful tool, uh and it's it it's helpful

853
00:55:24,460 --> 00:55:26,060
in in in quite a lot of ways.

854
00:55:26,140 --> 00:55:31,820
Um but the second meaning of computational that is more specific to I guess psychologists

855
00:55:31,820 --> 00:55:37,500
and neuroscientists is uh being interested in the computations that are performed by an

856
00:55:37,500 --> 00:55:41,740
organism like typically the computations that are performed in someone's brain.

857
00:55:42,140 --> 00:55:45,900
So it's really making theories about information processing.

858
00:55:46,540 --> 00:55:51,150
And the interestingly, the two meanings can be dissociated.

859
00:55:51,630 --> 00:55:57,550
So I can have a theory of the computation that is that someone is performing, or

860
00:55:57,550 --> 00:56:02,830
that a theory of the computation that a given part of the brain is performing, uh, that can

861
00:56:02,830 --> 00:56:05,150
be expressed in in verbal terms.

862
00:56:07,600 --> 00:56:12,080
the like uh another version of that is that I can have a very simple computational theory.

863
00:56:12,240 --> 00:56:18,160
So if you think uh if we go back to the theory of of anger uh that Aaron Cell gave, the the

864
00:56:18,160 --> 00:56:20,080
equation is is quite simple.

865
00:56:20,480 --> 00:56:26,000
It's just the I mean for those who don't know, it's an equation that determines how um

866
00:56:26,320 --> 00:56:30,640
someone decides whether to help based on this parameter that's called their wealth fair

867
00:56:30,720 --> 00:56:31,680
trade-off ratio.

868
00:56:32,230 --> 00:56:38,070
and then you you decide to help given that your welfare trade-off ratio is above uh

869
00:56:38,630 --> 00:56:44,550
so if the costs and the benefits of your action of the potential actions are related to

870
00:56:44,550 --> 00:56:46,630
the welfare trade-off ratio in a certain way.

871
00:56:46,790 --> 00:56:50,150
Um so this is a computational theory in both senses, right?

872
00:56:50,230 --> 00:56:52,630
Because it you can express it with an equation.

873
00:56:52,710 --> 00:56:58,440
Um, and this is you're also telling something about the the computation that is that is

874
00:56:58,440 --> 00:56:59,800
taking place in someone's brain.

875
00:57:00,040 --> 00:57:00,200
Right.

876
00:57:00,280 --> 00:57:05,320
So if Weber Trade F ratio is one, it's like, well, if you would get $10 and I would get um

877
00:57:05,320 --> 00:57:08,520
nine, I would just give it to you because that's more.

878
00:57:08,760 --> 00:57:10,200
They're that's just the same.

879
00:57:10,360 --> 00:57:14,120
But then it's like a par it's just it very it changes the cost and benefits.

880
00:57:14,200 --> 00:57:20,040
So like for my son, I might have a a a positive one where like even if if if

881
00:57:20,120 --> 00:57:26,090
even if I get uh ten but he gets seven, I still might give it to him because I care

882
00:57:26,090 --> 00:57:28,010
about him more than I care about me.

883
00:57:28,250 --> 00:57:32,010
And so it's it's like a very positive of our trade-off ratio.

884
00:57:32,170 --> 00:57:37,770
But then if it's like a negative one and yeah or a very negative value, then um

885
00:57:38,010 --> 00:57:43,930
I might, you know, um be willing to incur a cost to make sure you don't get a benefit.

886
00:57:44,090 --> 00:57:46,250
So just to make it slightly more concrete.

887
00:57:46,330 --> 00:57:46,490
Yeah.

888
00:57:46,650 --> 00:57:50,010
Sorry, but that's just Yeah no no of course that's a yeah that's a great summary.

889
00:57:50,250 --> 00:57:55,980
And so yeah and so uh uh I think and I mean a lot of evolutionary psychologists think that

890
00:57:55,980 --> 00:58:00,620
having those computational theories, you know, those theories of what is the computation is

891
00:58:00,620 --> 00:58:02,140
extremely valuable.

892
00:58:02,380 --> 00:58:07,660
Um and and this is valuable even when those computational theories are actually not

893
00:58:07,660 --> 00:58:10,540
expressed in uh in computational terms.

894
00:58:10,620 --> 00:58:15,260
I don't uh even if you don't have equations or computer code in order to explain the

895
00:58:15,260 --> 00:58:15,660
theory.

896
00:58:15,900 --> 00:58:16,060
Yeah.

897
00:58:16,590 --> 00:58:19,710
Of course, maybe the the ideal case is when you have both, right?

898
00:58:19,870 --> 00:58:24,910
When you have a computational theory that is expressed in equations or in computer code.

899
00:58:25,390 --> 00:58:29,070
But then there's another quadrant in the space of possibilities.

900
00:58:29,710 --> 00:58:34,350
There's a quadrant where you're using very complicated mathematics, but you're not saying

901
00:58:34,350 --> 00:58:35,710
anything about information processing.

902
00:58:35,950 --> 00:58:36,270
Exactly.

903
00:58:36,590 --> 00:58:39,070
And and that is, I mean, it's often very useful, right?

904
00:58:39,150 --> 00:58:44,220
Sometimes you might be doing sociology and don't care at all about how the mind works.

905
00:58:44,380 --> 00:58:48,780
Uh it's a very um valuable thing to do.

906
00:58:49,020 --> 00:58:49,260
Yeah.

907
00:58:49,900 --> 00:58:54,220
But then the the sort of problematic part of this quadrant is that when you are a

908
00:58:54,220 --> 00:58:59,260
psychologist and you say, look, I have a computational theory of X, and your computational

909
00:58:59,260 --> 00:59:03,180
theory is indeed there's a lot of very complicated mathematics.

910
00:59:03,260 --> 00:59:06,380
You can implement it, you can run nice simulations.

911
00:59:06,780 --> 00:59:11,490
But when you look at it a little bit better, then you're not actually really saying

912
00:59:11,490 --> 00:59:13,970
anything about information processing in the mind.

913
00:59:14,290 --> 00:59:20,290
You're just modeling how the observed variables are

914
00:59:20,290 --> 00:59:21,970
actually related to each other.

915
00:59:22,530 --> 00:59:25,010
Yeah, I think that's a really important profound point.

916
00:59:25,170 --> 00:59:26,290
Yeah, you and I were talked about this.

917
00:59:26,450 --> 00:59:29,730
We really, what does it mean to be computational is like a paper we should really do.

918
00:59:30,050 --> 00:59:33,250
I think a good concrete example of this would be just describing, like to make it super

919
00:59:33,250 --> 00:59:36,740
concrete for people, just a bicycle, like to the extent people know how bicycle works.

920
00:59:36,820 --> 00:59:40,100
Like you could have a verbal model of how bicycle works, like you know, you push the

921
00:59:40,100 --> 00:59:43,380
pedals, it turns the crank, it pulls the chain, it pulls the cassette, it turns the wheel.

922
00:59:43,540 --> 00:59:44,500
That's all verbal.

923
00:59:44,740 --> 00:59:49,620
I could have a very nice mathematical description of like how fast the tire rotates, and

924
00:59:49,700 --> 00:59:54,660
like sort of some other parameter, like I don't know how often the chain vibrates, that has

925
00:59:54,660 --> 00:59:58,500
nothing to do with explaining how the thing works.

926
00:59:58,970 --> 01:00:04,730
And if you just like if I typed up the functional description and if I made a beautiful

927
01:00:04,730 --> 01:00:10,650
graph of the of the uh um fluff and you just

928
01:00:10,650 --> 01:00:15,370
did superficially like which one looks science or people would be like oh the beautiful

929
01:00:15,370 --> 01:00:19,450
graph you have numbers like there's a pragmatic effect that like once you know the function

930
01:00:19,450 --> 01:00:25,050
of something so well you you can avoid being qualitative and you can become quantitative.

931
01:00:25,210 --> 01:00:30,980
But I believe there's been an enormous often cynical, frankly, thing where it's like

932
01:00:31,060 --> 01:00:34,980
I if I make something quantitative, I can b lead people into believing that we know

933
01:00:35,140 --> 01:00:39,780
understand it so well that we're beyond the qualitative phase and now we're doing

934
01:00:39,780 --> 01:00:40,500
quantitative science.

935
01:00:40,580 --> 01:00:43,780
It's the equivalent of like putting on a lab coat and being like I'm a scientist.

936
01:00:44,260 --> 01:00:49,860
Um and yeah, it's not a straw man to say like I have sat in on many, many, many, many,

937
01:00:49,940 --> 01:00:53,380
many, many, many, many, many talks that were this quadrant.

938
01:00:53,540 --> 01:00:54,180
Yeah.

939
01:00:54,930 --> 01:00:55,490
Yeah, yeah.

940
01:00:55,730 --> 01:01:00,850
And there's a the there's a version of that that ends up connecting back to the issue of

941
01:01:00,850 --> 01:01:06,370
correlation and causation, which is this kind of industry of mediation analysis and

942
01:01:06,370 --> 01:01:07,970
structural equation models.

943
01:01:08,050 --> 01:01:13,570
Uh you can have increasingly complicated mathematical models where um someone

944
01:01:13,890 --> 01:01:19,880
collects a lot of data um and then they put them inside the computer and then the

945
01:01:19,880 --> 01:01:21,960
computer crunches the numbers for hours.

946
01:01:22,120 --> 01:01:27,480
And then what comes back is a chart where you have uh you have little causal errors between

947
01:01:27,480 --> 01:01:28,520
one variable and another.

948
01:01:28,600 --> 01:01:30,440
And you can have a quite a complicated graph.

949
01:01:30,520 --> 01:01:35,240
And again, it looks like you have produced a description, a mechanistic description of the

950
01:01:35,240 --> 01:01:35,560
system.

951
01:01:35,800 --> 01:01:41,570
Um but very often those things um or kind of increasingly sophisticated

952
01:01:41,570 --> 01:01:46,610
ways of convincing yourself that you have causal insights, whereas you're you're just

953
01:01:46,610 --> 01:01:48,530
describing a pattern of correlations.

954
01:01:49,090 --> 01:01:49,410
Yeah.

955
01:01:51,010 --> 01:01:55,970
And we shouldn't, you know, I mean, oftentimes the the computational aspect of it can be

956
01:01:55,970 --> 01:01:57,890
quite simple, as in the case of anger.

957
01:01:57,970 --> 01:02:02,210
Like it's actually when you think about the design of anger, it's shockingly simple.

958
01:02:02,370 --> 01:02:07,730
You know, you're trying to figure out what uh cost a person would be willing to inflict on

959
01:02:07,730 --> 01:02:11,300
you if i if in exchange for whatever benefit to them.

960
01:02:11,460 --> 01:02:12,980
It's just a ratio, literally a ratio.

961
01:02:13,060 --> 01:02:14,020
You're trying to figure that out.

962
01:02:14,180 --> 01:02:17,620
And then your anger and gratitude should be calibrated in response to that.

963
01:02:17,780 --> 01:02:20,580
Like that's a pretty simple computational model.

964
01:02:20,740 --> 01:02:24,660
It doesn't require a huge page of equations or graphs.

965
01:02:25,060 --> 01:02:28,660
And so we should be open to the possibility that whatever system we're trying to explain in

966
01:02:28,660 --> 01:02:33,430
the mind should be as simple or as complicated as it needs to be for solving the particular

967
01:02:33,430 --> 01:02:34,150
adaptive problem.

968
01:02:34,230 --> 01:02:38,310
And we shouldn't have a prior that it necessarily needs to be very sophisticated.

969
01:02:38,630 --> 01:02:42,070
Or we should at least, you know, be open to the possibility that some parts of it might be

970
01:02:42,070 --> 01:02:44,870
actually very simple and some parts of it might be complicated depending on what

971
01:02:44,870 --> 01:02:46,950
perspective you're taking or what part you're looking at.

972
01:02:47,670 --> 01:02:52,070
In terms of the the possible inputs to the welfare trade-off ratio calculation, there might

973
01:02:52,070 --> 01:02:55,750
be many different inputs and many different contextual cues that that go into that

974
01:02:55,750 --> 01:02:56,230
calculation.

975
01:02:56,390 --> 01:02:57,830
And that part might be very complicated.

976
01:02:57,990 --> 01:03:02,060
But what what what is spit out in the end is actually very simple uh it's just a ratio of

977
01:03:02,060 --> 01:03:02,620
two numbers.

978
01:03:03,260 --> 01:03:09,180
Yeah um I think uh I think you could even go further than that and and say that

979
01:03:09,340 --> 01:03:13,420
uh even if you start with the baseline assumption that's everything will be quite

980
01:03:13,420 --> 01:03:17,820
complicated, um there is often value in having a simplified model.

981
01:03:17,980 --> 01:03:22,780
Yeah where you sort of say well of course we know that the system doesn't really work

982
01:03:22,780 --> 01:03:28,610
exactly like that but um it's actually the case that we have a very simple

983
01:03:28,610 --> 01:03:30,690
model that gives us a lot of traction.

984
01:03:30,850 --> 01:03:35,330
Um and I mean this is actually the if we look at the history of science, this is the way

985
01:03:35,330 --> 01:03:36,850
that that things have worked, right?

986
01:03:36,930 --> 01:03:42,530
I mean uh kepp Kepler's the Kepler's law of laws of motion, they're not they're not a

987
01:03:42,530 --> 01:03:45,570
complete description of kind of the way that the solar system works.

988
01:03:45,810 --> 01:03:50,840
We we didn't like it's not like the general theory of of

989
01:03:50,840 --> 01:03:55,320
gravitation uh and relativity sprung up fully formed.

990
01:03:55,480 --> 01:03:59,560
Um we had to start with very simple models and then we added a lot of complexity.

991
01:03:59,720 --> 01:04:04,360
And so yeah, probably in the sciences of the mind, there's no reason that things should

992
01:04:04,360 --> 01:04:05,880
work in the uh otherwise.

993
01:04:06,280 --> 01:04:08,600
This was a long oh sorry, David, go ahead.

994
01:04:09,000 --> 01:04:12,920
I I feel like we often confuse which parts should be simple and which parts should be

995
01:04:12,920 --> 01:04:13,320
complicated.

996
01:04:13,400 --> 01:04:15,880
The deeper principles should be simple.

997
01:04:16,770 --> 01:04:20,770
What results from the operation of those simple principles can be mind bogglingly

998
01:04:20,770 --> 01:04:21,090
complicated.

999
01:04:21,170 --> 01:04:24,290
I mean, the best example of this is the theory of evolution itself.

1000
01:04:24,450 --> 01:04:25,810
You know, it's extremely simple.

1001
01:04:25,970 --> 01:04:27,970
You can lay it out in a couple sentences.

1002
01:04:28,130 --> 01:04:30,690
Um, Darwin laid it out in a couple sentences.

1003
01:04:30,770 --> 01:04:35,970
Uh, and yet that simple idea, that simple principle, gives rise to the most mind bogglingly

1004
01:04:35,970 --> 01:04:40,370
complex forms of biological organization, including the brain that's producing these words,

1005
01:04:40,450 --> 01:04:42,930
is a is a product of the of that of those simple principles.

1006
01:04:43,220 --> 01:04:47,940
So that to me is the best example of an extreme difference between the simplicity of the

1007
01:04:47,940 --> 01:04:53,860
underlying causal process and the complexity of what root emerges from that causal

1008
01:04:53,860 --> 01:04:54,340
process.

1009
01:04:54,740 --> 01:04:58,740
And I think we should be trying to have the same kind of insights about the mind.

1010
01:04:58,820 --> 01:05:03,220
We should be trying to hone in on what the simple principles are that are producing the

1011
01:05:03,220 --> 01:05:05,700
complexity of mental and behavioral phenomena.

1012
01:05:05,860 --> 01:05:10,410
And I think evolution helps you focus your lens on what those simple principles ought to

1013
01:05:10,410 --> 01:05:10,570
be.

1014
01:05:10,730 --> 01:05:13,370
What are the underlying goals of an evolved organism?

1015
01:05:13,450 --> 01:05:17,370
Uh and and what what are the the parameters that we need to be looking for in order to

1016
01:05:17,370 --> 01:05:19,050
guide behavior adaptively.

1017
01:05:19,370 --> 01:05:23,450
Yeah, Tedek, I think I derailed I think you triggered me when you said computational.

1018
01:05:23,530 --> 01:05:25,290
I think this is like a long digression.

1019
01:05:25,450 --> 01:05:29,530
You were saying you have a computational model like ten minutes ago.

1020
01:05:29,770 --> 01:05:31,290
Do you remember what you were talking about?

1021
01:05:31,370 --> 01:05:34,000
If not, I have we can go on to something else.

1022
01:05:34,640 --> 01:05:40,160
I think I was just talking about the the computational model uh and the instantiation of

1023
01:05:40,160 --> 01:05:45,840
the the model of how anger follows from a sort of a Bayesian inference about

1024
01:05:46,000 --> 01:05:49,360
the um uh about the the person's welfare threat of ratio.

1025
01:05:49,600 --> 01:05:49,680
Right.

1026
01:05:49,840 --> 01:05:55,120
Because I guess in in previous work by Aaron Cell, there was this idea that yes, if you if

1027
01:05:55,360 --> 01:06:01,210
the cost and benefit ratio is such that um this gives you information about the person's

1028
01:06:01,210 --> 01:06:04,730
welfare total ratio, then you will calibrate your anger accordingly.

1029
01:06:05,290 --> 01:06:11,130
And so what what we did in that paper is kind of to really flesh out uh like how this would

1030
01:06:11,130 --> 01:06:12,410
work in in Bayesian terms.

1031
01:06:12,730 --> 01:06:12,890
Yeah.

1032
01:06:13,050 --> 01:06:16,410
In Bayesian terms, just so it's not like jargon or fancy.

1033
01:06:16,570 --> 01:06:21,450
Let me may let me see if I do a simple like roughly six year old level description that you

1034
01:06:21,450 --> 01:06:22,570
tell me if it's accurate.

1035
01:06:22,810 --> 01:06:27,940
So like one version of sort of like let's use the the cloud rain example, right?

1036
01:06:28,100 --> 01:06:31,460
You could do something where you're sort of like, well, every time it rains, how often are

1037
01:06:31,460 --> 01:06:32,260
there clouds?

1038
01:06:32,500 --> 01:06:35,940
And then you could just sort of say, Well, that's how I'm going to figure out how likely it

1039
01:06:35,940 --> 01:06:37,220
is it's going to rain.

1040
01:06:37,380 --> 01:06:38,820
That's one version.

1041
01:06:38,900 --> 01:06:43,860
Um, but a smarter version would you would say, Well, look, how how when it's raining, when

1042
01:06:43,860 --> 01:06:44,820
are there clouds?

1043
01:06:44,980 --> 01:06:48,980
But you should look at another category too, which is how often are there clouds and it's

1044
01:06:48,980 --> 01:06:49,940
not raining?

1045
01:06:50,180 --> 01:06:52,260
And then you should combine those two.

1046
01:06:52,470 --> 01:06:56,070
And roughly speaking, that's what being Bayesian is.

1047
01:06:56,310 --> 01:06:58,470
It's looking at both things and not just the one.

1048
01:06:58,550 --> 01:07:01,030
Is that a fair, super simple gloss?

1049
01:07:01,270 --> 01:07:01,910
Yeah, yeah, yeah.

1050
01:07:02,070 --> 01:07:08,070
If we want to go even more at a more foundational level, maybe the idea behind

1051
01:07:08,070 --> 01:07:13,960
Bayesian thinking is that um there's a lot of uncertainty in the world, And so,

1052
01:07:14,120 --> 01:07:20,040
I mean, with some exception, with some exceptions, we can never be certain of most

1053
01:07:20,040 --> 01:07:22,840
things because of the uncertainty.

1054
01:07:23,400 --> 01:07:29,320
And so the best that we can do is to assign some number to our belief that something

1055
01:07:29,320 --> 01:07:30,200
is the case.

1056
01:07:30,520 --> 01:07:36,120
So maybe I look out the window and it's I see clouds in the sky, and

1057
01:07:36,120 --> 01:07:40,150
maybe maybe it's going to rain, but I'm not quite sure.

1058
01:07:40,230 --> 01:07:44,470
Maybe I will say that there's a 0.9 chance that it's going to rain.

1059
01:07:44,630 --> 01:07:50,150
And so this idea is that we can quantify our belief in something by assigning

1060
01:07:50,150 --> 01:07:50,710
numbers.

1061
01:07:51,190 --> 01:07:57,030
And then the the good way to do this, if we are going to engage in sound reasoning, is for

1062
01:07:57,030 --> 01:07:58,870
those numbers to be probabilities.

1063
01:07:59,580 --> 01:08:04,300
Because when we have probabilities, then we can work with them and we can combine them in

1064
01:08:04,300 --> 01:08:05,260
particular ways.

1065
01:08:12,780 --> 01:08:17,420
Is that you should really uh work with probabilities in certain instead of just truth

1066
01:08:17,420 --> 01:08:17,980
values.

1067
01:08:18,300 --> 01:08:23,900
And then this and this kind of framework is called uh Bayesianism after Bayes theorem,

1068
01:08:23,980 --> 01:08:29,890
because Bayes theorem tells you what is the right way to update your the number that you

1069
01:08:29,890 --> 01:08:34,050
assign to a particular proposition based on the evidence that you have.

1070
01:08:34,450 --> 01:08:34,930
Yeah.

1071
01:08:35,170 --> 01:08:39,330
It's uh we can do yeah, we could do a whole two hours on what is a probability.

1072
01:08:40,050 --> 01:08:40,770
We should not.

1073
01:08:41,250 --> 01:08:44,530
Um I wanted to I wanted to slightly move direct, change direction.

1074
01:08:44,690 --> 01:08:45,650
Oh, David, do you have something?

1075
01:08:45,730 --> 01:08:50,530
I don't want to of just like we've talked a lot about causation versus correlation, and

1076
01:08:50,530 --> 01:08:53,490
we've sort of been like, well, look, the mind is actually pretty good at like picking out

1077
01:08:53,490 --> 01:08:56,680
causation, But there's another sense in which causation doesn't matter at all.

1078
01:08:56,760 --> 01:09:00,680
And I think this gets into your free will work, which is imagine you have a causal account.

1079
01:09:00,760 --> 01:09:03,640
Like, I don't know, let's imagine you're up on a stage, it's made out of wood, and you slip

1080
01:09:03,640 --> 01:09:05,720
on a banana and you hit your head really hard.

1081
01:09:05,800 --> 01:09:09,240
And you're like, what caused me to hit my head?

1082
01:09:09,640 --> 01:09:14,120
There's a long stretch of causal sequences, including like the people who chopped down the

1083
01:09:14,120 --> 01:09:17,400
tree that made the stage, the person that grew the banana, right?

1084
01:09:17,640 --> 01:09:23,160
Like there, there's a long stretch of causality that all led up to you hitting your head.

1085
01:09:23,470 --> 01:09:29,390
But we, we meaning like psychological design, focuses in on

1086
01:09:30,030 --> 01:09:36,030
not all of that causation, but the part of the causation that has the highest effective

1087
01:09:36,030 --> 01:09:40,110
sort of turnover, where it's sort of like, where's the biggest bang for my buck?

1088
01:09:40,270 --> 01:09:44,350
Like, if I don't want to slip, should I like go to where they grow the trees and like

1089
01:09:44,430 --> 01:09:46,110
convince them to not grow the trees?

1090
01:09:46,350 --> 01:09:49,790
Should I like, you know, uh go to the banana grower and not do that?

1091
01:09:49,870 --> 01:09:52,430
No, it's like, well, that banana shouldn't have been on the ground there.

1092
01:09:52,510 --> 01:09:57,720
And so, like a notion of causation is also nar

1093
01:09:58,200 --> 01:10:03,240
it's not just all actual causation, it narrows down to like as David describes, sort of

1094
01:10:03,240 --> 01:10:07,400
like points of intervention or things I could do something about, which is what you would

1095
01:10:07,400 --> 01:10:09,400
expect an organism to do.

1096
01:10:09,880 --> 01:10:12,280
And I that's coming from your work, correct?

1097
01:10:12,440 --> 01:10:17,000
That like I think you're also making this point that sort of an interventionist view of

1098
01:10:17,000 --> 01:10:21,530
causation, I think there's other people doing it too, is also very evolutionary in the

1099
01:10:21,530 --> 01:10:27,370
sense of even if you grant causal views, it really is a a a shrinking

1100
01:10:27,370 --> 01:10:32,170
down of all possible causal sequences to those things you can functionally do something

1101
01:10:32,170 --> 01:10:32,490
about.

1102
01:10:33,290 --> 01:10:33,610
Yeah.

1103
01:10:33,690 --> 01:10:39,450
Uh and actually the first time I heard about this problem was uh a talk by by John

1104
01:10:39,690 --> 01:10:41,220
Toby at HBS.

1105
01:10:41,620 --> 01:10:47,540
And I don't think he ever published anything of the sort, but I

1106
01:10:47,540 --> 01:10:49,460
mean I think this is one of those things.

1107
01:10:49,860 --> 01:10:52,660
It's a great example of of instinct blindness.

1108
01:10:53,220 --> 01:10:56,580
This is a problem that we never think about.

1109
01:10:57,060 --> 01:11:02,980
Basically, I mean it's um it's it's something that's has been it has

1110
01:11:02,980 --> 01:11:08,020
taken uh quite a long time before people actually started to realize that well this is

1111
01:11:08,020 --> 01:11:10,110
actually an important psychological problem.

1112
01:11:10,350 --> 01:11:13,150
What is going on in people's brain uh when they do that?

1113
01:11:13,390 --> 01:11:18,670
Because this is one of those problems that we solve so effortlessly and unconsciously that

1114
01:11:18,670 --> 01:11:21,150
it just does not register that that we are doing it.

1115
01:11:21,870 --> 01:11:22,510
That sounds weird.

1116
01:11:22,670 --> 01:11:26,030
You have to like make effort to be like, well, it's the growers of the trees.

1117
01:11:26,110 --> 01:11:27,390
It's like weird to come up with that.

1118
01:11:28,270 --> 01:11:33,840
It's I mean that's why when I sometimes talk about my work, people look at me funny because

1119
01:11:34,400 --> 01:11:36,080
why is this kind of thing?

1120
01:11:36,880 --> 01:11:42,800
Um another kind of a classical example of of this kind of situation,

1121
01:11:43,040 --> 01:11:48,800
this kind of problem, is that suppose that you want to know what started a forest fire,

1122
01:11:48,880 --> 01:11:53,440
uh what was the cause, and that you know that well there were a lot of dry leaves on on the

1123
01:11:53,440 --> 01:11:58,080
ground and it was very hot on that day, there was a lot of wind, um, there was a lightning

1124
01:11:58,080 --> 01:12:02,550
bolt that struck a tree, there was some oxygen in the air, and so on and so forth.

1125
01:12:02,710 --> 01:12:08,230
Um it seems almost obvious that it was the lightning bolt that was the cause of the forest

1126
01:12:08,230 --> 01:12:08,630
fire.

1127
01:12:08,790 --> 01:12:11,270
Even though there are many things that also contributed.

1128
01:12:11,670 --> 01:12:17,590
So um the oxygen was in a sense also the the the cause of the fire because if we had

1129
01:12:17,590 --> 01:12:21,030
magically removed the oxygen from the air, then the fire would not have happened.

1130
01:12:21,320 --> 01:12:25,560
But it feels almost ludicrous to kind of ask the question why do we think it's the

1131
01:12:25,560 --> 01:12:27,240
lightning bolt and not the oxygen?

1132
01:12:27,400 --> 01:12:30,280
It's like, of course it's the it's not the oxygen.

1133
01:12:30,680 --> 01:12:36,520
This is um an intuition that our brain is giving is giving to us in such a vivid way that

1134
01:12:36,520 --> 01:12:41,720
it doesn't even feel that we had to do any computations in in the first place.

1135
01:12:42,600 --> 01:12:48,430
And so and I think that that's why people didn't really think a lot about

1136
01:12:48,430 --> 01:12:51,550
that problem, uh, and a lot of work that went into this.

1137
01:12:51,710 --> 01:12:54,110
So, this is called the problem of causal selection.

1138
01:12:54,270 --> 01:12:58,510
Sometimes people talk about the problem of causal responsibility and so on and so forth.

1139
01:12:58,670 --> 01:13:04,670
Um, so a lot of the work on this uh sprung up from issues in moral psychology.

1140
01:13:05,150 --> 01:13:10,270
So people say, Well, I know sometimes we have an intuition that some people are more

1141
01:13:10,590 --> 01:13:13,710
morally responsible for a bad outcome.

1142
01:13:13,950 --> 01:13:19,840
Um, and when we try to analyze this concept of moral responsibility, then we find

1143
01:13:19,840 --> 01:13:25,520
that well, causality has a lot to do with it, uh, which is itself uh its own

1144
01:13:25,600 --> 01:13:26,720
interesting thing, right?

1145
01:13:26,880 --> 01:13:32,800
Because uh, you know, you can imagine uh a thought experiment where I have an intent to do

1146
01:13:32,800 --> 01:13:38,790
something bad, I want to I don't I want to harm David, And then David

1147
01:13:38,790 --> 01:13:42,070
just gets harmed in exactly the way that I wanted, but actually, there was no connection,

1148
01:13:42,310 --> 01:13:44,950
causal connection between my intent and his harm.

1149
01:13:46,550 --> 01:13:52,310
People will not punish me as much because there was no causal connection between the bad

1150
01:13:52,310 --> 01:13:54,390
intent and the bad outcome.

1151
01:13:55,590 --> 01:14:01,110
And so this is a bit of a tangent to say that in the process of analyzing how people uh

1152
01:14:01,110 --> 01:14:06,620
think about moral responsibility, uh philosophers and social psychologists and experimental

1153
01:14:06,620 --> 01:14:12,300
philosophers came to wonder well, why how is it that we attribute

1154
01:14:12,300 --> 01:14:14,620
causal responsibility in the first place?

1155
01:14:14,940 --> 01:14:20,220
Um, and so yeah, I think that is interesting because it's we had to

1156
01:14:20,460 --> 01:14:24,700
recognize that there is an information processing to be solved here, but we did not show in

1157
01:14:24,780 --> 01:14:28,220
the field did not really come to it in a very direct way.

1158
01:14:33,350 --> 01:14:39,030
So, um, how how do we figure out when someone is morally

1159
01:14:39,030 --> 01:14:39,510
responsible?

1160
01:14:39,670 --> 01:14:45,670
I mean, certainly, you know, the the causal part of things is is relevant, but

1161
01:14:45,670 --> 01:14:50,470
it do we figure it out in the same way that we figure out that the lightning bolt caused

1162
01:14:50,470 --> 01:14:53,990
the fire, or is it there's something different going on in the realm of moral

1163
01:14:53,990 --> 01:14:55,110
responsibility?

1164
01:14:57,080 --> 01:15:02,840
So uh maybe it it might be useful to talk about the the information

1165
01:15:02,840 --> 01:15:08,440
processing logic of causal responsibility in the first place before we extend it to the the

1166
01:15:08,440 --> 01:15:10,440
the realm of of moral responsibility.

1167
01:15:10,680 --> 01:15:15,880
Um so so one one one sort of obvious idea for how we

1168
01:15:16,120 --> 01:15:21,160
assign responsibility is that we engage in in counterfactual thinking.

1169
01:15:28,190 --> 01:15:33,550
Is that we imagine a counterfactual possibility where we remove the lightning bolt and see

1170
01:15:33,710 --> 01:15:35,470
and we see what happens to the forest fire.

1171
01:15:36,030 --> 01:15:39,550
And then we say, oh without the lightning bolt, then there's no forest fire.

1172
01:15:40,990 --> 01:15:43,870
And in intuitively it produces the correct result.

1173
01:15:44,350 --> 01:15:50,180
But one problem with with that is that we are now kind of forced to judge that the

1174
01:15:50,180 --> 01:15:54,100
presence of oxygen in the air was also a cause of the forest fire, right?

1175
01:15:54,180 --> 01:15:57,700
Because well, if there was no oxygen, then there was a forest fire.

1176
01:15:57,860 --> 01:16:03,780
Um an example that people often use is so you can have this situation where um you

1177
01:16:03,780 --> 01:16:07,140
ask me to water your plants while you're away on a trip.

1178
01:16:07,220 --> 01:16:11,700
Okay, so I'm the one that's supposed to water you, and I kind of forget uh, and then your

1179
01:16:11,700 --> 01:16:13,380
plants wither and die.

1180
01:16:13,860 --> 01:16:19,850
So you can say, well, um my my my omission from watering

1181
01:16:19,850 --> 01:16:24,010
the plants was the cause of your plants dying, which seems like a fine thing to say.

1182
01:16:24,170 --> 01:16:29,290
And this is because if I actually had decided to water your plants, then they would have

1183
01:16:29,290 --> 01:16:29,850
survived.

1184
01:16:30,170 --> 01:16:33,370
So in that counterfactual uh the outcome has changed.

1185
01:16:33,530 --> 01:16:39,050
But then you can also say, well, no, I think it's the Pope that caused your plants to die.

1186
01:16:39,130 --> 01:16:44,170
Because if the Pope had come to your flat and watered your plants, then they would not have

1187
01:16:44,170 --> 01:16:44,490
died.

1188
01:16:44,810 --> 01:16:46,520
Yeah, there's a lot of counterfactuals.

1189
01:16:46,600 --> 01:16:49,240
If you hadn't plant if you didn't have plants, you wouldn't have killed them.

1190
01:16:49,320 --> 01:16:53,080
I mean if there was no water, you wouldn't have killed the space of possible

1191
01:16:53,080 --> 01:16:55,880
counterfactuals that are relevant is is just infinite, right?

1192
01:16:56,120 --> 01:17:02,040
And so and so people need to have a way of just focusing on the counterfactuals that

1193
01:17:02,040 --> 01:17:06,120
are actually relevant to the problem of causal attribution.

1194
01:17:06,280 --> 01:17:11,850
Um and so a lot of what is going on is this that some counterfactual possibilities are just

1195
01:17:11,850 --> 01:17:13,370
more relevant than others.

1196
01:17:14,090 --> 01:17:19,850
And there is some work that has looked at what makes a counterfactual relevant for the

1197
01:17:19,850 --> 01:17:21,690
purpose of causal attribution.

1198
01:17:21,930 --> 01:17:27,930
And so roughly um a counterfactual is relevant if it it was kind of high

1199
01:17:27,930 --> 01:17:29,530
probability to to begin with.

1200
01:17:29,850 --> 01:17:35,290
So the pope coming to your to your flat and watering your plants is just was so unlikely

1201
01:17:35,290 --> 01:17:38,260
that this would happen that this is not a relevant possibility.

1202
01:17:38,420 --> 01:17:44,020
And in the same way, the present like the absence of oxygen in the air is such a remote

1203
01:17:44,020 --> 01:17:46,020
possibility that we're not going to consider it.

1204
01:17:46,260 --> 01:17:52,020
Um, also, people don't really focus on counterfactual possibilities that deviate too much

1205
01:17:52,020 --> 01:17:53,540
from what actually happened.

1206
01:17:54,740 --> 01:18:00,420
So the idea is sorry, like just to make so like in an alien world where oxygen was variable

1207
01:18:00,580 --> 01:18:06,260
but lightning was not, those aliens would think that oxygen, the oxygen caused the fire.

1208
01:18:06,670 --> 01:18:08,430
Yeah, yeah, yeah, absolutely.

1209
01:18:08,670 --> 01:18:14,430
Um and then I mean where it gets maybe a little bit more spooky is that um

1210
01:18:14,590 --> 01:18:20,190
kind of social and and moral norms also seem to have an influence on what we think

1211
01:18:20,350 --> 01:18:23,950
what or the the relevant kind of kind of actual possibilities.

1212
01:18:24,270 --> 01:18:29,940
So it might be that I don't know, um I I every day

1213
01:18:30,020 --> 01:18:34,340
like uh Dave and I kind of come to water your plants.

1214
01:18:34,420 --> 01:18:37,380
But you kind of repeatedly ask Dave not to come in.

1215
01:18:37,540 --> 01:18:40,820
So it's kind of like he's not supposed to be here, but Dave comes in anyway.

1216
01:18:41,220 --> 01:18:44,420
And I'm really the one that's supposed to come in and water your plants.

1217
01:18:44,660 --> 01:18:50,660
Um so then if kind of suddenly none of us come, neither of us come and your plants

1218
01:18:50,660 --> 01:18:55,380
die, it seems that I like I'm the person that is responsible for your plants dying because

1219
01:18:55,380 --> 01:18:57,220
I'm the one that was supposed to be here.

1220
01:18:57,470 --> 01:18:59,550
That's like the counterfactual possibilities.

1221
01:19:00,030 --> 01:19:05,630
The counterfactual possibility where I am I actually show a patcher apartment seems to be

1222
01:19:05,630 --> 01:19:07,550
the most relevant counterfactual.

1223
01:19:09,070 --> 01:19:14,510
And this is this is something for which I kind of I kind of wish there was a strong

1224
01:19:14,510 --> 01:19:19,710
adaptationist answer for why this should be a functional requirement of causal attribution.

1225
01:19:20,590 --> 01:19:25,320
But I don't I don't have super well formed ideas about why this is the case.

1226
01:19:25,720 --> 01:19:30,360
Well, wouldn't it be that I mean what unpacking relevant it comes back to interventionism,

1227
01:19:30,440 --> 01:19:30,600
right?

1228
01:19:30,760 --> 01:19:36,680
So like if the point is what is what's the thing that David can do

1229
01:19:37,000 --> 01:19:42,920
going forward, it's gonna be it's gonna be weird to say, well,

1230
01:19:43,080 --> 01:19:44,120
Dave, I was wrong.

1231
01:19:44,280 --> 01:19:46,360
You should have been coming all along.

1232
01:19:47,010 --> 01:19:52,370
Uh uh like he has to recalibrate himself and me, but just you, right?

1233
01:19:52,530 --> 01:19:56,850
So, like, you know, he asked you to water the plants, but you didn't.

1234
01:19:57,090 --> 01:19:59,650
He asked me to not be there, but I was there.

1235
01:19:59,970 --> 01:20:02,050
And we didn't water the plants.

1236
01:20:02,290 --> 01:20:08,130
The the most efficient, the the biggest bang for your buck causal interventionist thing is

1237
01:20:08,130 --> 01:20:14,120
to recalibrate how much you are taking into account what he's saying

1238
01:20:14,120 --> 01:20:17,160
about um the right about that.

1239
01:20:17,320 --> 01:20:20,200
Like you you just fail to do something you heard him say.

1240
01:20:20,440 --> 01:20:25,480
So that's like the I mean it's not exactly couldn't it couldn't it just be reframed in

1241
01:20:25,480 --> 01:20:31,330
terms of the more likely counterfactual um being a

1242
01:20:31,330 --> 01:20:36,450
more being more relevant because if you think about people responding to social

1243
01:20:36,450 --> 01:20:41,970
expectations or me doing what you request of me, like the probability that I will do what

1244
01:20:41,970 --> 01:20:44,210
you request of me is very high, right?

1245
01:20:44,770 --> 01:20:48,850
Um or or the the probability that I I'd forget is also pretty high, right?

1246
01:20:49,010 --> 01:20:55,010
Um and so it may it may be that um responding to requests or social expectations

1247
01:20:55,010 --> 01:21:01,000
is just is very relevant because um the uh the the degree to which we take them into

1248
01:21:01,000 --> 01:21:05,720
account or not is just at baseline a very high likelihood thing to uh consider.

1249
01:21:06,120 --> 01:21:07,160
Is that yeah, yeah.

1250
01:21:07,240 --> 01:21:09,400
I think that that's that's one possible explanation.

1251
01:21:09,480 --> 01:21:14,120
It's a little bit of a a deflationary explanation that all people are actually just

1252
01:21:14,120 --> 01:21:15,320
tracking probability.

1253
01:21:15,560 --> 01:21:20,120
Um I think I think that there is some evidence against this where people have done

1254
01:21:20,120 --> 01:21:24,890
experiments where they really emphasize well no like the the kind of frequency of those

1255
01:21:24,890 --> 01:21:26,650
things is is really the same.

1256
01:21:26,810 --> 01:21:31,930
Um, but then you could counter argue that you know to what extent the participants actually

1257
01:21:31,930 --> 01:21:37,610
believe what you're telling them, and maybe they're still um using those social

1258
01:21:37,610 --> 01:21:41,690
norms uh as a as a QT probability, that's for sure.

1259
01:21:41,850 --> 01:21:47,240
Um but I I'm I'm I'm kind of inclined to believe that there is There is

1260
01:21:47,240 --> 01:21:52,680
still an effect of kind of social expectations on the relevance of

1261
01:21:52,680 --> 01:21:53,720
countervactuals.

1262
01:21:53,800 --> 01:21:59,240
And one possible explanation to that is what Dave suggested about identifying good points

1263
01:21:59,240 --> 01:22:00,120
of intervention.

1264
01:22:00,520 --> 01:22:05,720
Maybe another possibility is that when we engage in causal attribution, we are trying to

1265
01:22:05,720 --> 01:22:11,720
communicate things about our beliefs, about our values, about our

1266
01:22:11,720 --> 01:22:13,000
own model of the world.

1267
01:22:13,240 --> 01:22:19,190
And so one way that we can communicate what we expect of others is

1268
01:22:19,190 --> 01:22:24,950
to actually uh make causal attributions that will contain cues about

1269
01:22:25,190 --> 01:22:27,430
what's what we expect them to do.

1270
01:22:27,910 --> 01:22:28,150
Yeah.

1271
01:22:28,390 --> 01:22:33,270
So that would predict, for example, like let's say there's um I don't know, we're holding

1272
01:22:33,270 --> 01:22:35,190
up a I'm trying to think of a good example.

1273
01:22:35,270 --> 01:22:39,030
This is a dumb example, but let's imagine like we're holding up a wall that might fall

1274
01:22:39,030 --> 01:22:39,350
over.

1275
01:22:40,620 --> 01:22:43,020
and or or we need to, right?

1276
01:22:43,100 --> 01:22:45,740
Like we're we've noticed this wall we've all worked hard on it.

1277
01:22:45,900 --> 01:22:46,300
We need to.

1278
01:22:46,380 --> 01:22:52,380
And I I tell like 10 people, the wall's falling over, it's gonna like um cause our uh

1279
01:22:52,460 --> 01:22:56,060
it's gonna like hit, it's gonna ru hurt us in some way, right?

1280
01:22:56,220 --> 01:22:58,620
And then another 10 people I don't tell.

1281
01:22:59,020 --> 01:23:01,740
And all else equal, the wall falls over.

1282
01:23:01,900 --> 01:23:05,980
And then you just say, like, well, who's responsible for the wall falling over?

1283
01:23:06,220 --> 01:23:08,940
My intuition is it's the 10 people I told.

1284
01:23:09,100 --> 01:23:15,030
And that's just a readout of like who's more likely to if I add a little bit

1285
01:23:15,030 --> 01:23:19,510
of information or who's easier to convince um, you know, like the

1286
01:23:20,630 --> 01:23:23,430
that's the that's more and more responsibility.

1287
01:23:23,510 --> 01:23:26,790
So I think it comes back to the that's the interventionist account right there.

1288
01:23:27,190 --> 01:23:32,390
That um that it's going to be easier to intervene if I've already told somebody and there's

1289
01:23:32,390 --> 01:23:37,840
common knowledge that there's a cost and there's a problem versus I haven't even said

1290
01:23:37,840 --> 01:23:39,280
anything yet to you.

1291
01:23:39,360 --> 01:23:42,960
Um that's enough to tip it towards the first 10 and not the second 10.

1292
01:23:43,600 --> 01:23:49,120
Yeah and and it is definitely the case that the the counterfactual where those people show

1293
01:23:49,120 --> 01:23:54,160
up is much more relevant than the counterfactual where it is the 10 people who have not

1294
01:23:54,160 --> 01:23:56,160
heard the information that that show up.

1295
01:23:56,650 --> 01:23:56,970
Yeah.

1296
01:23:57,050 --> 01:23:59,690
I mean, I I personally, like this is my own bias.

1297
01:23:59,770 --> 01:24:02,730
I don't like it relevance language because I find that obscuring.

1298
01:24:03,050 --> 01:24:08,410
Because I think it's sort of an outcome of the interventionist calculation.

1299
01:24:08,730 --> 01:24:14,730
Um, it doesn't like the relevance is sort of a placeholder for like this is what you think.

1300
01:24:15,290 --> 01:24:20,940
Not that relevance is calculated and then that's what causes it to be uh the outcome.

1301
01:24:21,020 --> 01:24:25,820
It's that it's relevant because it's the outcome calculated based on interventionist logic.

1302
01:24:26,380 --> 01:24:28,300
That's m that's my own personal bias.

1303
01:24:28,460 --> 01:24:33,420
But um interestingly the sort of the the relevance of something is not

1304
01:24:34,540 --> 01:24:39,820
there is not always a direct a direct mapping uh between the relevance of a counterfactual

1305
01:24:39,820 --> 01:24:41,740
and whether it is selected as the cause.

1306
01:24:41,990 --> 01:24:47,270
which was like I think one of the most interesting discoveries in this research on causal

1307
01:24:47,270 --> 01:24:48,070
attribution.

1308
01:24:48,230 --> 01:24:54,150
Um, and I can so I can first give a like an example of an experiment that shows this:

1309
01:24:54,310 --> 01:25:00,070
that the the most relevant thing is not always the um the what people

1310
01:25:00,070 --> 01:25:02,710
pick as a a causal explanation.

1311
01:25:02,870 --> 01:25:07,430
Um and then I can maybe say a few words about my own preferred explanation for for this

1312
01:25:07,430 --> 01:25:07,830
pattern.

1313
01:25:08,230 --> 01:25:13,760
And so you can imagine that there's this person, call them Bob, they're trying to

1314
01:25:14,000 --> 01:25:14,560
graduate.

1315
01:25:14,720 --> 01:25:18,480
And they're almost past all their requirements for graduating.

1316
01:25:18,640 --> 01:25:24,640
They just need to pass two different exams in order to graduate.

1317
01:25:24,880 --> 01:25:29,680
So they need to pass the history exam, which is famously very easy, and they need to pass

1318
01:25:29,680 --> 01:25:32,160
the math exam, which is famously very hard.

1319
01:25:32,320 --> 01:25:35,040
Everyone, almost everyone fails at the math exam, right?

1320
01:25:35,680 --> 01:25:38,560
And so they really need to pass both of them in order to graduate.

1321
01:25:38,830 --> 01:25:44,670
And suppose that Bob goes, he and he passes the math exam and the history exam.

1322
01:25:44,830 --> 01:25:46,510
Um, and therefore he graduates.

1323
01:25:46,590 --> 01:25:51,710
Uh, when you tell people that story, um, they will say, Well, I think Bob graduated because

1324
01:25:51,710 --> 01:25:53,310
he passed the math exam.

1325
01:25:53,950 --> 01:25:59,950
And so here we have a situation where um it seems that the most relevant counterfactual

1326
01:26:00,190 --> 01:26:05,560
is kind of failing the math exam because it was very likely in the first place that he

1327
01:26:05,560 --> 01:26:07,480
would fail this very difficult exam.

1328
01:26:07,720 --> 01:26:11,960
And also people think that yes, it was passing the math exam that was the cause.

1329
01:26:12,120 --> 01:26:17,480
So here we have we have an alignment between the what counterfactual is relevant and causal

1330
01:26:17,480 --> 01:26:18,200
attribution.

1331
01:26:18,440 --> 01:26:24,200
But you can also modify the story in a very uh slight way, in such a way that actually

1332
01:26:24,600 --> 01:26:26,360
you will get the reverse pattern.

1333
01:26:26,630 --> 01:26:32,390
So now if you say, well, um, and this is Alice, and she also needs to graduate.

1334
01:26:32,470 --> 01:26:37,910
Um, but Alice will graduate if she passes either the history exam or the math exam.

1335
01:26:38,390 --> 01:26:42,790
And again, the math exam is the most difficult thing, but she only needs one of them.

1336
01:26:42,950 --> 01:26:47,430
Um then again, Alice passes the math exam and the history exam and she graduates.

1337
01:26:47,510 --> 01:26:53,430
Uh, when you ask participants uh why did Alice graduate, they will tell you, Well, it is

1338
01:26:53,430 --> 01:26:55,750
because she passed the history exam.

1339
01:26:56,080 --> 01:27:00,000
I don't know, I don't know if you to what extent you share this intuition, but maybe it was

1340
01:27:00,080 --> 01:27:05,600
you know that's the she only needed to pass one of these.

1341
01:27:05,760 --> 01:27:09,840
Uh and since the history exam was the the easy one, that's kind of that's the one that

1342
01:27:09,840 --> 01:27:12,080
actually explains why she graduated.

1343
01:27:12,480 --> 01:27:17,520
Um and and in that case, it is still the case that if we think about what is the most

1344
01:27:17,520 --> 01:27:23,280
relevant counterfactual, um, the one that is easiest to come to mind in a way, um,

1345
01:27:23,440 --> 01:27:26,990
that is the counterfactual where she fails the mathematics exam.

1346
01:27:27,070 --> 01:27:31,230
Because again, the math exam is super difficult, and so it's quite likely that she would

1347
01:27:31,230 --> 01:27:32,270
have failed it.

1348
01:27:32,990 --> 01:27:38,670
But despite the fact that the most relevant

1349
01:27:38,670 --> 01:27:42,750
counterfactual is the one where she fails the math exam, people have this intuition that

1350
01:27:42,750 --> 01:27:46,350
it's really it was passing the history exam that caused her to graduate.

1351
01:27:47,080 --> 01:27:51,000
And really all that we did is change one word in the story.

1352
01:27:59,400 --> 01:28:01,400
I have intuitions about what might explain it.

1353
01:28:01,800 --> 01:28:02,360
I I do.

1354
01:28:02,680 --> 01:28:05,800
But m maybe Dave had a methodological comment.

1355
01:28:06,040 --> 01:28:06,440
No, no, no.

1356
01:28:06,520 --> 01:28:06,760
No, yeah.

1357
01:28:06,920 --> 01:28:07,400
Um okay.

1358
01:28:12,600 --> 01:28:18,210
Uh it might be that what we're simulating counterfactually is how

1359
01:28:18,210 --> 01:28:23,730
likely um she would be to study for the history exam versus the math exam.

1360
01:28:23,810 --> 01:28:27,010
Like whether she would she would really invest in trying to pass the history exam because

1361
01:28:27,010 --> 01:28:29,250
that's gonna get her the most bang for her buck.

1362
01:28:29,490 --> 01:28:35,490
And so um we uh so we're imagining that and that it would be really unlikely if she uh

1363
01:28:35,650 --> 01:28:39,650
if she invested all of her effort studying for the math exam.

1364
01:28:41,100 --> 01:28:42,860
So it's interventionless, sort of again.

1365
01:28:43,100 --> 01:28:46,540
Yeah, and it's again it goes back to this notion of what is the optimal point of

1366
01:28:46,540 --> 01:28:47,100
intervention.

1367
01:28:47,580 --> 01:28:49,420
Um yeah, I like this explanation.

1368
01:28:49,500 --> 01:28:52,940
Um however, I don't think it can explain like all possible cases.

1369
01:28:53,100 --> 01:28:53,260
Okay.

1370
01:28:53,580 --> 01:28:59,260
So you can you can run the same version of this experiment where the the two events are

1371
01:28:59,260 --> 01:29:01,020
completely out of your control.

1372
01:29:01,450 --> 01:29:06,650
So you have no in order to you're playing some sort of casino like chance game.

1373
01:29:06,810 --> 01:29:11,930
And in order to win the game you need to kind of draw a a ball from an urn.

1374
01:29:12,090 --> 01:29:13,930
And you kind of you're going to draw at random.

1375
01:29:14,170 --> 01:29:19,370
And so maybe you're going to win the game if you get a a green ball from

1376
01:29:20,650 --> 01:29:25,360
either one of two boxes in that would be one condition.

1377
01:29:25,520 --> 01:29:31,040
Um and in the other condition, you win if you get a successful draw from the earning

1378
01:29:31,280 --> 01:29:33,600
from uh from both both boxes.

1379
01:29:33,840 --> 01:29:37,600
And so in that kind of setup, you still find the same effects.

1380
01:29:37,840 --> 01:29:43,280
And then since the outcome of a given draw is a random one, you can't really uh

1381
01:29:43,920 --> 01:29:46,960
give an explanation in terms of like how much someone's going to prepare.

1382
01:29:47,570 --> 01:29:47,730
Okay.

1383
01:29:48,050 --> 01:29:48,370
That makes sense.

1384
01:29:48,690 --> 01:29:51,330
But isn't it still an intervention in the sense of like what's the outcome that has the

1385
01:29:51,330 --> 01:29:52,530
biggest bang for the buck?

1386
01:29:52,610 --> 01:29:55,090
Or like what do you need to what what goal do you need to do?

1387
01:29:56,210 --> 01:30:01,490
If if you were like, well, how was I going to do if I can manipulate the system, what's the

1388
01:30:01,490 --> 01:30:03,970
thing I need to do to get the outcome I want?

1389
01:30:04,450 --> 01:30:06,930
Yes, that that's definitely one one explanation.

1390
01:30:07,010 --> 01:30:12,580
And people have tried to formalize this insight in computational terms and it explains some

1391
01:30:12,580 --> 01:30:14,580
of the data but not all of it.

1392
01:30:17,540 --> 01:30:23,380
But I mean the the most parsimonious explanation that I came up with for that is

1393
01:30:23,380 --> 01:30:29,140
that um when people make a causal judgment they they imagine not

1394
01:30:29,140 --> 01:30:34,970
just one um counterfactual possibility but they they simulate many of them.

1395
01:30:36,250 --> 01:30:41,690
So they they kind of sample, simulate some number of counterfactuals.

1396
01:30:42,810 --> 01:30:48,490
And each counterfactual is simulated with a probability that is

1397
01:30:49,370 --> 01:30:53,610
that is proportional to kind of its what we would call its relevance.

1398
01:30:53,850 --> 01:30:59,600
So it is simulated in proportion to how uh to its prior probability, simulated

1399
01:30:59,760 --> 01:31:04,240
in proportion to how close it was to the the actual event and so on and so forth.

1400
01:31:04,720 --> 01:31:09,280
And then once you have this collection of different counterfactual possibilities, you can

1401
01:31:09,280 --> 01:31:14,400
compute the correlation between the cause and the effect across those counterfactuals.

1402
01:31:14,560 --> 01:31:20,110
Yeah, so if you're trying to think, oh is it the passing the history exam that caused Alice

1403
01:31:20,270 --> 01:31:21,150
to win?

1404
01:31:21,310 --> 01:31:27,230
Then you compute what is the correlation between passing the history exam and uh

1405
01:31:27,230 --> 01:31:30,510
Alice graduating in the worlds that you have simulated.

1406
01:31:30,750 --> 01:31:36,110
And so when you do that, you find that well, there is in almost all worlds, she she

1407
01:31:36,110 --> 01:31:37,070
graduates, right?

1408
01:31:37,150 --> 01:31:39,150
And she passes the history exam.

1409
01:31:39,230 --> 01:31:43,310
And it kind of doesn't really matter whether she passes the math exam or not.

1410
01:31:43,470 --> 01:31:49,040
And so the correlation between passing the history exam and graduating is quite high.

1411
01:31:49,280 --> 01:31:53,680
And that's why you think that it was passing the history exam that was the cause of her

1412
01:31:53,840 --> 01:31:54,720
graduating.

1413
01:31:54,960 --> 01:31:58,160
And now we can apply the same logic to the case of Bob.

1414
01:31:58,320 --> 01:32:04,080
So just as a reminder, Bob needed to pass both the math exam and the history exam in order

1415
01:32:04,080 --> 01:32:04,640
to graduate.

1416
01:32:04,800 --> 01:32:06,800
Like he really had to have both.

1417
01:32:07,280 --> 01:32:12,730
And so you kind of engage in the same process of simulating counterfactual possibilities.

1418
01:32:12,890 --> 01:32:18,570
And now you see that well, there's going to be many worlds in which Bob passes the

1419
01:32:18,890 --> 01:32:20,970
history exam, but he fails the math exam.

1420
01:32:21,050 --> 01:32:22,250
And so he graduates.

1421
01:32:30,540 --> 01:32:35,580
So it means that now you have a high correlation between passing the math exam and

1422
01:32:35,580 --> 01:32:36,300
graduating.

1423
01:32:36,540 --> 01:32:41,820
And this then explains why we think that, well, in the case of Bob, it was passing the math

1424
01:32:41,820 --> 01:32:43,020
exam that caused him to graduate.

1425
01:32:43,420 --> 01:32:43,980
Let me make sure.

1426
01:32:44,060 --> 01:32:48,140
So in other words, it's the crucial difference of all these counterfactuals are run

1427
01:32:48,220 --> 01:32:48,380
through.

1428
01:32:48,540 --> 01:32:54,140
It's sort of what's the crucial difference that leads to the outcome

1429
01:32:54,140 --> 01:32:56,380
desired versus not desired.

1430
01:32:57,350 --> 01:32:57,670
Yeah.

1431
01:32:57,910 --> 01:32:58,870
Yes, yes.

1432
01:32:59,350 --> 01:33:05,190
But this idea of, I think this idea of a crucial difference is a nice intuition for what is

1433
01:33:05,190 --> 01:33:05,830
going on.

1434
01:33:07,030 --> 01:33:12,310
And this kind of process of counterfactual simulation and computing correlation is an

1435
01:33:12,310 --> 01:33:15,590
explicit information processing description of what is going on.

1436
01:33:15,830 --> 01:33:16,070
Right.

1437
01:33:16,150 --> 01:33:19,270
It's a crucial difference right across those counterfactuals.

1438
01:33:19,540 --> 01:33:20,180
Yeah, yeah.

1439
01:33:20,340 --> 01:33:20,980
Yeah.

1440
01:33:21,860 --> 01:33:22,740
Interesting.

1441
01:33:22,980 --> 01:33:28,100
Um presumably these counterfactual simulations are happening unconsciously.

1442
01:33:28,180 --> 01:33:32,900
It it seems like it introspectively when when I was thinking about the the Bob problem or

1443
01:33:33,060 --> 01:33:37,940
or the the math test problem, like I didn't have a vivid, you know, simulation in my head

1444
01:33:38,020 --> 01:33:41,540
where you know he he passed in this case, he did he didn't in this case, he graduated in

1445
01:33:41,540 --> 01:33:41,940
this case.

1446
01:33:42,100 --> 01:33:43,620
It's just kind of I just had an intuition.

1447
01:33:43,700 --> 01:33:47,300
So presumably these simulations are happening unconsciously.

1448
01:33:47,970 --> 01:33:53,090
Yeah, and I I think I assume that's yeah, it is the case for most people.

1449
01:33:53,410 --> 01:33:58,450
I myself don't really have uh the phenomenology of running those simulations.

1450
01:33:58,610 --> 01:34:04,610
Um and so from uh from a cognitive perspective, there are two different

1451
01:34:04,610 --> 01:34:06,370
possible accounts of what is going on.

1452
01:34:06,610 --> 01:34:12,370
There's is an account where we can we do run those simulations uh but it's just implicit

1453
01:34:12,370 --> 01:34:13,730
and unconscious.

1454
01:34:13,890 --> 01:34:19,120
Um there is maybe another account where we don't even need to

1455
01:34:20,160 --> 01:34:22,080
to really engage in simulation.

1456
01:34:22,160 --> 01:34:28,080
It instead we our brain has found a way, kind of kind of a shortcut that allows you

1457
01:34:28,080 --> 01:34:33,120
to compute what would be the result of those um of those simulations.

1458
01:34:33,520 --> 01:34:39,440
Kind of maybe as an analogy for what this possibility might look like, you might think

1459
01:34:39,440 --> 01:34:42,800
about you know the process, the much simpler process of computing probabilities.

1460
01:34:43,450 --> 01:34:48,170
So I might want one way that you could compute the probability that

1461
01:34:50,490 --> 01:34:56,330
a dice is going to come up for is to just simulate many rows, many rolls of the

1462
01:34:56,330 --> 01:34:58,330
dice, and you count up the number of fours.

1463
01:34:59,690 --> 01:35:02,170
But you don't necessarily have to do that.

1464
01:35:02,250 --> 01:35:05,130
You can use the laws of probability theory to compute that.

1465
01:35:05,290 --> 01:35:10,650
And so there might be a way that our brain has also evolved some shortcut way of computing

1466
01:35:10,650 --> 01:35:13,920
what would be the outcome of such counterfactual simulations.

1467
01:35:14,400 --> 01:35:20,240
Would an example of that just be that like there's a representation of high variability and

1468
01:35:20,240 --> 01:35:25,920
low variability, for example, like associated with like you know doing good in

1469
01:35:25,920 --> 01:35:26,960
math versus this?

1470
01:35:27,120 --> 01:35:33,040
And so even just that common sense knowledge about the variability um

1471
01:35:33,840 --> 01:35:37,200
curtails some of the needing to run through the simulations.

1472
01:35:37,870 --> 01:35:43,630
Yeah, I mean it might be that if I kind of give you like a

1473
01:35:43,950 --> 01:35:49,870
novel scenario where you have an unfamiliar thing, then uh your brain

1474
01:35:49,870 --> 01:35:51,870
needs to engage in simulation.

1475
01:35:52,030 --> 01:35:57,870
Um but if I ask you about something that is much more common common knowledge,

1476
01:35:58,030 --> 01:36:02,910
I mean in the sense that those are familiar things, then you can um you can use a sort of a

1477
01:36:02,910 --> 01:36:08,520
pre-computed uh Yeah, or even just like if you have like imagine like a physics is a really

1478
01:36:08,520 --> 01:36:08,920
good example.

1479
01:36:09,080 --> 01:36:14,920
Like let's just say the physics, your folk the folk physics um system in your head, the

1480
01:36:14,920 --> 01:36:20,520
physics engine in your head, um just calculates based on certain

1481
01:36:20,520 --> 01:36:23,960
experiences, how precarious something is.

1482
01:36:24,200 --> 01:36:28,770
Um And then there's principles that then just can get extended automatically to new

1483
01:36:28,770 --> 01:36:32,290
instances based on the extracted cues.

1484
01:36:32,450 --> 01:36:36,690
So, like, you know, is a beam precarious, like when it's up straight up and down versus

1485
01:36:36,690 --> 01:36:37,810
like laying on the ground.

1486
01:36:37,970 --> 01:36:43,490
And there in that has been based on past runs of the of the

1487
01:36:43,490 --> 01:36:46,210
physics and seeing these things fall.

1488
01:36:46,370 --> 01:36:51,320
But now there's sort of that gets represented and attributed in new instances without

1489
01:36:51,320 --> 01:36:52,600
running the simulation.

1490
01:36:52,840 --> 01:36:58,840
That on its own, if you see a new instance where the the counterfactual

1491
01:36:58,840 --> 01:37:04,840
depends on the likelihood of stability versus falling, that involves no

1492
01:37:04,840 --> 01:37:07,720
running of the counterfactuals.

1493
01:37:07,800 --> 01:37:13,450
It's just that you already there's already a system there that plugs into one is more

1494
01:37:13,450 --> 01:37:18,330
likely, one is less likely, and you don't have to run which the counterfactuals.

1495
01:37:18,410 --> 01:37:20,410
It would be a concrete example of that.

1496
01:37:20,810 --> 01:37:23,690
Yeah, yeah, I definitely think that something like that could happen.

1497
01:37:23,850 --> 01:37:28,810
Uh I think yeah, I think it's it's an example, it's uh an illustration of the fact that the

1498
01:37:28,810 --> 01:37:34,800
model itself uh is really couched at the kind of the David Mars computational level of

1499
01:37:34,800 --> 01:37:40,080
of analysis where like okay, here is the computation that is that is that the system is

1500
01:37:40,080 --> 01:37:41,200
trying to implement.

1501
01:37:41,360 --> 01:37:46,160
Um when it comes to how the s the the mind is doing that, yeah, uh concretely it kind of

1502
01:37:46,160 --> 01:37:47,120
all backs are off current.

1503
01:37:47,360 --> 01:37:47,920
Yeah, yeah, exactly.

1504
01:37:48,000 --> 01:37:52,000
And the I think intuitions about what's easier and harder are really bad then too, because

1505
01:37:52,080 --> 01:37:57,920
you know, it's yeah, it's what it feels like to me to do something is not how hard

1506
01:37:57,920 --> 01:38:00,590
or easy it is for a computational system.

1507
01:38:00,750 --> 01:38:00,990
Yeah.

1508
01:38:01,870 --> 01:38:02,190
Yeah.

1509
01:38:02,350 --> 01:38:07,230
I mean, cause what's interesting to me is that there are cases where this is conscious that

1510
01:38:07,230 --> 01:38:08,830
I can think of pretty easily.

1511
01:38:08,910 --> 01:38:13,550
And it seems like when I do have a phenomenology of causal reasoning, it tends to fit your

1512
01:38:13,550 --> 01:38:14,110
model pretty well.

1513
01:38:14,270 --> 01:38:18,510
So like the example I'm thinking of is I'm playing chess and I really want to get

1514
01:38:18,510 --> 01:38:19,390
checkmate.

1515
01:38:19,620 --> 01:38:23,060
I might actually consciously simulate all the different moves and counter moves.

1516
01:38:23,140 --> 01:38:24,820
If I do this, you might do this in response.

1517
01:38:24,980 --> 01:38:27,140
If you do if you do that, I'll do this in response.

1518
01:38:27,300 --> 01:38:29,540
Uh and I might simulate a bunch of different possibilities.

1519
01:38:29,620 --> 01:38:34,100
You know, if I move this piece versus this piece, and I might look at the correlation it

1520
01:38:34,100 --> 01:38:37,860
consciously in my head of moving this piece and checkmate further down the line.

1521
01:38:38,020 --> 01:38:41,860
And if that correlation is high, like, oh, and all these different simulated possibilities,

1522
01:38:42,020 --> 01:38:45,860
moving this piece helps me get to checkmate, then I'll move that piece.

1523
01:38:45,940 --> 01:38:50,470
So it seems like uh there are some cases where we we are doing this uh in in in a

1524
01:38:50,470 --> 01:38:52,070
full-fledged conscious way.

1525
01:38:52,150 --> 01:38:56,070
And I'm curious if you if you have any thoughts about when we would need to do that

1526
01:38:56,070 --> 01:38:57,830
consciously and when we wouldn't.

1527
01:38:58,470 --> 01:39:03,670
Yeah, something that is related to to that point is that uh even though even if we assume

1528
01:39:03,670 --> 01:39:09,590
that the system is kind of mostly implicit, um it's not um I don't I I don't want to use

1529
01:39:09,590 --> 01:39:14,900
the term encapsulated here in in in Dave's presence because this we I'll leave.

1530
01:39:19,220 --> 01:39:23,380
You can definitely like uh like a verbally uh intervene on that system.

1531
01:39:23,620 --> 01:39:28,740
Um so you know it's people have done this experiment where you know again you have the case

1532
01:39:28,740 --> 01:39:31,460
of um Alice who needs to graduate.

1533
01:39:31,620 --> 01:39:35,460
And this time I I don't tell you anything about the difficulty of the exams, I just tell

1534
01:39:35,460 --> 01:39:39,170
you there's the history exam and the the the the math exam, right?

1535
01:39:39,330 --> 01:39:45,010
And I tell you, okay, like so she she passes the math exam, she passes the history exam.

1536
01:39:45,090 --> 01:39:50,210
And I tell you, well, just try to imagine you know what would have happened if she had

1537
01:39:50,210 --> 01:39:51,410
failed the math exam.

1538
01:39:51,650 --> 01:39:54,050
Kind of try to really simulate the counterfactual.

1539
01:39:54,130 --> 01:39:59,010
And then after you have done that, I ask you, well, what do you think was the cause of her

1540
01:39:59,010 --> 01:39:59,490
passing?

1541
01:39:59,740 --> 01:40:02,620
And then people will say, oh, this was the history exam.

1542
01:40:03,100 --> 01:40:03,340
Right.

1543
01:40:03,500 --> 01:40:09,340
And so you kind of sort of can artificially push people to to simulate one

1544
01:40:09,340 --> 01:40:10,700
counterfactual more.

1545
01:40:11,420 --> 01:40:14,380
And then this will have an effect on their causal attribution.

1546
01:40:15,020 --> 01:40:20,300
So you can sort of even if the process is implicit, you can sort of um use explicit

1547
01:40:20,300 --> 01:40:23,420
communication in order to influence what people are going to do.

1548
01:40:24,530 --> 01:40:25,970
That's fascinating.

1549
01:40:26,690 --> 01:40:32,290
And I think then the other part of your question is when are you going to engage in sort of

1550
01:40:32,290 --> 01:40:34,290
more conscious simulation?

1551
01:40:35,890 --> 01:40:41,810
I think that at least one part of the answer is that if you are in a novel

1552
01:40:41,810 --> 01:40:46,290
situation, maybe this is the first time that you've been playing this chess, this

1553
01:40:46,290 --> 01:40:49,020
particular kind of chess game.

1554
01:40:49,100 --> 01:40:52,780
I you I guess you you were never in that position before, then you need to engage in

1555
01:40:52,780 --> 01:40:54,140
explicit simulation.

1556
01:40:54,380 --> 01:40:59,900
Um but maybe if this is a situation that you're very familiar with, then all the

1557
01:40:59,900 --> 01:41:05,340
computation has just been done and you can rely on schemas that have been pre computed on

1558
01:41:05,340 --> 01:41:06,780
the basis of your past experience.

1559
01:41:07,100 --> 01:41:07,500
Interesting.

1560
01:41:07,580 --> 01:41:11,340
So maybe like the opening moves are just gonna be wrote, but like the end game, that's

1561
01:41:11,340 --> 01:41:13,340
where my conscious reasoning might kick in.

1562
01:41:13,740 --> 01:41:14,220
Yeah, yeah.

1563
01:41:14,530 --> 01:41:20,450
And I I think I think that uh to a large extent, you know, the people who develop uh AI

1564
01:41:20,450 --> 01:41:26,130
systems that play chess or go, they have this kind of hybrid architecture where they will

1565
01:41:26,370 --> 01:41:32,050
rely a lot on those kinds of uh cached pre-computed um

1566
01:41:32,770 --> 01:41:35,890
sort of calculations of what is the value of a particular move.

1567
01:41:36,050 --> 01:41:40,130
And then they will com combine this with more explicit simulations.

1568
01:41:40,370 --> 01:41:46,080
Um and it's quite likely that the human brain is designed along more or less the same

1569
01:41:46,080 --> 01:41:46,960
principles.

1570
01:41:48,640 --> 01:41:49,360
Interesting.

1571
01:41:49,920 --> 01:41:52,160
Um, let's talk about theory of mind.

1572
01:41:52,320 --> 01:41:57,600
Uh I mean, talking about nebula is hard to understand, um, obtuse stuff.

1573
01:41:57,760 --> 01:42:02,560
Um uh I I I have a very soft spot in my heart for theory of mind.

1574
01:42:02,640 --> 01:42:05,120
Theory of mind like the ability to attribute mental states.

1575
01:42:05,450 --> 01:42:10,090
Mental states are just like causation, as we talked about, like where in some sense they do

1576
01:42:10,090 --> 01:42:10,650
exist, right?

1577
01:42:10,730 --> 01:42:15,210
Like in the sense of it's a convention of like, of course, I have a desire and a belief.

1578
01:42:15,290 --> 01:42:21,050
And um, but there's another sense in which they're placeholders for common

1579
01:42:21,050 --> 01:42:22,170
outcomes in the world.

1580
01:42:22,330 --> 01:42:26,170
Um, so there it's very confusing and weird and abstract too.

1581
01:42:26,410 --> 01:42:31,160
So um, so where does theory of mind come in?

1582
01:42:31,240 --> 01:42:35,000
Um and is essentialism part of it, or is that a separate part?

1583
01:42:35,080 --> 01:42:39,640
Is is uh because because in addition to causation, you've tackled theory of mind and

1584
01:42:39,640 --> 01:42:40,680
essentialism.

1585
01:42:40,840 --> 01:42:46,360
Um uh so that's like the trifecta of of of of of abstraction.

1586
01:42:46,680 --> 01:42:51,560
Uh what what is the what is up with theory of mind and um and essentialism and are they

1587
01:42:51,560 --> 01:42:52,040
related?

1588
01:42:52,280 --> 01:42:52,520
Yeah.

1589
01:42:52,760 --> 01:42:57,690
Well, I think that the nice thing is as we could do like a a two hour discussion on each.

1590
01:42:58,090 --> 01:42:58,410
Exactly.

1591
01:42:58,570 --> 01:42:59,850
Uh yeah.

1592
01:43:00,490 --> 01:43:01,290
Get comfortable.

1593
01:43:01,450 --> 01:43:02,010
Get comfortable.

1594
01:43:02,650 --> 01:43:08,650
The bad news I maybe for the sake of kind of uh uh for the sake of tractability,

1595
01:43:08,810 --> 01:43:11,690
uh I'll focus on theory of mind uh for now.

1596
01:43:12,010 --> 01:43:12,170
Sure.

1597
01:43:12,410 --> 01:43:17,130
Um so theory of mind uh is yeah, I mean as you said it's quite interesting because there's

1598
01:43:17,130 --> 01:43:21,880
that's a lot of abstraction that's we can't really see someone's desires or someone's

1599
01:43:21,880 --> 01:43:27,720
beliefs um yet we have to attribute beliefs and desires and and so on and so

1600
01:43:27,720 --> 01:43:32,360
forth to other people in order to even start understanding their behavior.

1601
01:43:32,760 --> 01:43:37,720
If I were to just try to model you in physical terms

1602
01:43:38,850 --> 01:43:40,690
Then it would be very difficult.

1603
01:43:40,850 --> 01:43:46,370
I mean, I would predict maybe you're going to say to stay roughly at the same place over

1604
01:43:46,370 --> 01:43:51,730
time, that there you will be some sort of spatial temporal continuity to your behavior, but

1605
01:43:51,730 --> 01:43:54,130
that's roughly how far I'm going to go.

1606
01:43:54,370 --> 01:44:00,130
Whereas when I attribute to you mental states, then um the richness of my explanation of

1607
01:44:00,130 --> 01:44:05,330
your behavior is going to be much uh much uh much more effective.

1608
01:44:06,400 --> 01:44:11,840
And so this is something that has been of quite a lot of interest to psychologists because

1609
01:44:11,840 --> 01:44:16,400
obviously we want to know what is the functional logic of this system that has been so

1610
01:44:16,400 --> 01:44:16,720
important.

1611
01:44:17,760 --> 01:44:23,600
Evolutionarily, this is something that social cognition is, especially in a

1612
01:44:23,600 --> 01:44:26,080
species like ours, it is super important.

1613
01:44:26,160 --> 01:44:30,880
If you cannot predict the behavior of other people and if you cannot exert some degree of

1614
01:44:30,880 --> 01:44:34,160
influence on them, then you're not going to go very far.

1615
01:44:34,590 --> 01:44:34,990
Right.

1616
01:44:35,950 --> 01:44:41,710
And so a lot of a lot of research and theory of mind has looked at um our

1617
01:44:41,710 --> 01:44:43,790
ability to attribute beliefs.

1618
01:44:44,510 --> 01:44:47,070
And there are good reasons to to that.

1619
01:44:48,510 --> 01:44:54,270
Because in order to really predict

1620
01:44:54,430 --> 01:45:00,030
um the someone's behavior you need to engage in what people call uh a meta

1621
01:45:00,190 --> 01:45:01,070
representation.

1622
01:45:01,320 --> 01:45:06,840
A r meta representation is just a fancy term to say a representation of a representation.

1623
01:45:07,000 --> 01:45:07,240
Okay.

1624
01:45:07,640 --> 01:45:13,000
Um and it's because if I represent the representations that are inside your mind, then I

1625
01:45:13,000 --> 01:45:14,920
can I will know how you behave.

1626
01:45:15,160 --> 01:45:21,080
Uh and uh a standard test for whether someone has meta representations

1627
01:45:21,240 --> 01:45:26,840
is whether they are able to imagine that someone might uh represent the world in a

1628
01:45:26,840 --> 01:45:28,520
different way than than than your own.

1629
01:45:28,670 --> 01:45:28,910
Okay.

1630
01:45:29,070 --> 01:45:34,670
So and then to give a some simple example, um, there is a mug on on the table on my on my

1631
01:45:34,670 --> 01:45:36,430
office, and I assume you can see it.

1632
01:45:36,590 --> 01:45:42,590
And so I can represent that um that maybe I'll okay I can represent that

1633
01:45:42,830 --> 01:45:48,510
Dave has a representation of the fact that there is a mug on the table.

1634
01:45:49,220 --> 01:45:55,140
But to really have a handle on to really be able to prove that you have a representation, I

1635
01:45:55,140 --> 01:46:00,820
need to show that you are able to represent the world as being different from it actually

1636
01:46:00,820 --> 01:46:01,140
is.

1637
01:46:01,540 --> 01:46:07,460
And so maybe you might have uh I might have a sophisticated hologram of a

1638
01:46:07,460 --> 01:46:10,820
table, of a of a mug on the table instead of an actual mug.

1639
01:46:11,300 --> 01:46:15,620
If that is the case, I know that there is in fact no mug on the table.

1640
01:46:16,660 --> 01:46:20,490
But I know that you believe that there is a mug on the table.

1641
01:46:20,730 --> 01:46:26,730
Um and so a lot of research and theory of mind has been looking at this

1642
01:46:26,970 --> 01:46:30,490
ability to attribute false beliefs to to others.

1643
01:46:30,730 --> 01:46:36,410
Um and in in some of my recent work, I've been trying to to see how

1644
01:46:36,570 --> 01:46:41,860
uh maybe uh how we can think about theory of mind in that

1645
01:46:42,020 --> 01:46:47,780
involves uh representations that are not as uh as demanding as as that

1646
01:46:49,540 --> 01:46:55,380
what makes that demanding uh so so in in the

1647
01:46:55,380 --> 01:47:01,220
case of like representing that you know maybe you think that there is a a

1648
01:47:01,220 --> 01:47:05,810
mug on the table whereas Yeah, like let's say you move the mug and I didn't know that.

1649
01:47:05,890 --> 01:47:06,610
Like I just have a false.

1650
01:47:07,170 --> 01:47:11,090
So like you told me you showed me the mugs on the table, it's out of frame.

1651
01:47:11,410 --> 01:47:16,930
You are uh like a false belief, naughty character, you move the mug, I didn't know it.

1652
01:47:17,090 --> 01:47:22,130
So now you know where the mug is, uh, but I don't, but I so I have a false belief.

1653
01:47:22,210 --> 01:47:22,530
Yeah.

1654
01:47:22,850 --> 01:47:23,170
Yeah.

1655
01:47:23,330 --> 01:47:27,970
Um intuitively it's not necessarily demanding, but I mean if if you think about the the

1656
01:47:27,970 --> 01:47:33,960
logic of meta representation, uh it's if you really take it to the extreme

1657
01:47:34,040 --> 01:47:39,560
that in then it means that when i represent your mind then i will

1658
01:47:39,560 --> 01:47:45,320
construct a whole model of your own of your model of the mind

1659
01:47:45,800 --> 01:47:51,400
right and that's obviously that's a lot to represent right so i need to represent like

1660
01:47:51,560 --> 01:47:57,290
how i think the world is and then i need to have a another

1661
01:47:57,290 --> 01:48:03,050
representation that is basically the the world as I think uh Dave sees

1662
01:48:03,050 --> 01:48:03,370
it.

1663
01:48:04,490 --> 01:48:10,090
And so if you take on to the extreme it it's really um this idea of meta representation

1664
01:48:10,410 --> 01:48:16,410
really uh involves the commitments that I'm going to have one model of the world

1665
01:48:16,410 --> 01:48:22,140
for each person that I'm trying to to uh to as you interact with

1666
01:48:22,700 --> 01:48:28,540
so just to make sure i understand um the the you're arguing against a sort of an argument

1667
01:48:29,100 --> 01:48:34,860
which is the way that we represent other people's mental states is we create

1668
01:48:35,180 --> 01:48:40,700
an additional model of an entire other human brain

1669
01:48:41,500 --> 01:48:47,150
and what it currently is representing in addition to our own, and that we

1670
01:48:47,150 --> 01:48:51,710
multiply that every time for new states of the world and for new agents.

1671
01:48:52,190 --> 01:48:57,070
And that uh that sort of naive model is unlikely to be true.

1672
01:48:57,550 --> 01:48:57,950
Yes.

1673
01:48:58,110 --> 01:48:59,790
Yeah, that's uh that's a good summary.

1674
01:48:59,950 --> 01:49:03,470
And I don't think that's that anyone is really claiming that we do that.

1675
01:49:04,360 --> 01:49:10,360
But if you if you kind of look at the disclaim that theory of mind involves

1676
01:49:10,360 --> 01:49:15,240
meta-representation and that you run all the way with it, then it seems to be the logical

1677
01:49:15,240 --> 01:49:15,880
consequence.

1678
01:49:16,760 --> 01:49:22,600
And so what is your what's your claim that that is instead of that claim, uh what is your

1679
01:49:22,600 --> 01:49:23,000
argument?

1680
01:49:23,160 --> 01:49:23,800
What's going on?

1681
01:49:23,880 --> 01:49:24,120
Yeah.

1682
01:49:24,520 --> 01:49:24,760
Right.

1683
01:49:24,840 --> 01:49:29,990
And so an interesting question is indeed now no what are ways that the system could

1684
01:49:30,070 --> 01:49:34,710
actually work and not completely run out of memory uh every time you're trying to predict

1685
01:49:34,710 --> 01:49:35,590
someone's behavior.

1686
01:49:35,990 --> 01:49:41,430
Um and here the the idea that we have been putting forward is that you don't always need

1687
01:49:41,430 --> 01:49:42,710
meta representation.

1688
01:49:42,870 --> 01:49:46,790
And in fact in in most cases you you don't actually need it.

1689
01:49:46,950 --> 01:49:52,870
So in the case where I'm not playing any tricks on you and there is actually a mug on my

1690
01:49:52,870 --> 01:49:58,060
table and you see the mug and now you know that there is a mug on the table,

1691
01:49:59,020 --> 01:50:04,460
then I don't need to attribute to you the necessarily the a meta representation that there

1692
01:50:04,460 --> 01:50:08,860
is a mug on the table, that a kind of a belief that there is a mug on a table.

1693
01:50:09,020 --> 01:50:14,060
Instead I can just note that well you know that there is a mug on the table.

1694
01:50:14,220 --> 01:50:19,930
And so what this does um is that I have a physical description and a

1695
01:50:19,930 --> 01:50:22,330
physical I have my own physical model of the world.

1696
01:50:22,410 --> 01:50:25,930
So in my physical model of the world, there is a mug on the table.

1697
01:50:26,570 --> 01:50:32,330
And then in addition to that, I just add a very simple binary representation that oh Dave

1698
01:50:32,570 --> 01:50:34,650
knows this fact.

1699
01:50:35,290 --> 01:50:41,040
So now instead of having to duplicate my own model of the world in sort of the

1700
01:50:41,360 --> 01:50:44,160
meta representation that I attribute to you.

1701
01:50:44,400 --> 01:50:50,240
I can just uh have this one binary, one like on or off tag about whether you

1702
01:50:50,240 --> 01:50:50,560
know this.

1703
01:50:50,880 --> 01:50:56,320
And that the def what that on off tag means is uh all the

1704
01:50:56,320 --> 01:51:02,240
inferences and consequences that I have for the mug, the other agent has as well.

1705
01:51:02,630 --> 01:51:06,470
I'm not saying that that's uh the full model because of course it's not right, because like

1706
01:51:06,550 --> 01:51:09,430
I might see it from a different angle and you would need that additional.

1707
01:51:09,590 --> 01:51:14,870
But but just to be clear, like what it means is that there's no um

1708
01:51:15,830 --> 01:51:21,750
there's no difference between what you all the things that follow about the mug for you

1709
01:51:21,910 --> 01:51:24,950
are applied to me, but they don't have to get applied to me.

1710
01:51:25,030 --> 01:51:30,900
It's just that there's representation of everything that is represented in me

1711
01:51:30,900 --> 01:51:34,500
about the mug also applies to this other agent.

1712
01:51:34,740 --> 01:51:35,460
Yes, yes, exactly.

1713
01:51:35,860 --> 01:51:39,140
So it's like one piece of information instead of duplicating a brain.

1714
01:51:39,300 --> 01:51:39,380
Yeah.

1715
01:51:39,540 --> 01:51:39,620
Yeah.

1716
01:51:39,940 --> 01:51:40,020
Yeah.

1717
01:51:40,180 --> 01:51:44,260
In the same way that, you know, we scan the the the physical environment and we tag

1718
01:51:44,500 --> 01:51:46,420
physical objects with various properties.

1719
01:51:46,580 --> 01:51:52,020
We might say this, you know, coffee cup tag belongs to Dave, uh, contains hot liquid, um

1720
01:51:52,500 --> 01:51:55,540
is, you know, this sh shade of blue or whatever.

1721
01:51:55,620 --> 01:51:58,170
Um uh you know uh whatever.

1722
01:51:58,250 --> 01:52:00,970
And in addition to that, another tag is Dave knows it.

1723
01:52:01,130 --> 01:52:03,930
And we just oh or didn't Dave knows it's there.

1724
01:52:04,010 --> 01:52:08,570
And we just have that tag to, you know, everything we know, um, Dave also knows.

1725
01:52:08,730 --> 01:52:09,050
Yeah.

1726
01:52:09,210 --> 01:52:09,610
Right.

1727
01:52:09,930 --> 01:52:10,410
Yeah, yeah.

1728
01:52:10,490 --> 01:52:13,610
That that's yeah, it's such a great description of of the proposal.

1729
01:52:13,770 --> 01:52:13,930
Yeah.

1730
01:52:14,170 --> 01:52:16,810
Just one more feature of the of the object.

1731
01:52:17,660 --> 01:52:23,180
Um yeah so I mean this is this was just an idea um this this is not an idea that actually

1732
01:52:23,180 --> 01:52:29,180
came from from us this existed in in the literature um but this had been proposed as

1733
01:52:29,180 --> 01:52:33,420
just like one possibility for how theory of mind might work.

1734
01:52:33,740 --> 01:52:39,690
And what we we did actually in in our project was to try to take a

1735
01:52:39,690 --> 01:52:42,730
more adaptationist stance towards this proposal.

1736
01:52:42,970 --> 01:52:48,890
And we said, well, can we actually formalize what is the adaptive problem

1737
01:52:48,890 --> 01:52:49,130
here?

1738
01:52:49,770 --> 01:52:55,610
And we started with this fact that, well, we obviously we need to predict and explain the

1739
01:52:55,610 --> 01:53:01,460
behavior of others, but we also have a constraint on on

1740
01:53:01,460 --> 01:53:03,220
our information processing capacity.

1741
01:53:03,460 --> 01:53:07,300
Like we don't have infinite computation, we don't have infinite memory and so on and so

1742
01:53:07,300 --> 01:53:08,180
forth and so forth.

1743
01:53:08,580 --> 01:53:14,580
And so the design of our theory of mind system needs to somehow trade off between those

1744
01:53:14,580 --> 01:53:15,940
two different considerations.

1745
01:53:16,500 --> 01:53:21,780
We want to have a lot of predictive power while at the same time like not run out of

1746
01:53:21,780 --> 01:53:23,940
computation every time we interact with someone.

1747
01:53:24,330 --> 01:53:29,450
And this is this idea that computation is a constraint is really salient in the context of

1748
01:53:29,450 --> 01:53:34,410
theory of mind where it's very easy to run out of memory when you're predicting the

1749
01:53:34,410 --> 01:53:40,010
behavior of an entity that is as complex or even more complex as than than you are.

1750
01:53:41,130 --> 01:53:46,330
And so once you you set up this optimization problem where you try to trade off between

1751
01:53:46,330 --> 01:53:51,760
those two different things, The nice thing is that you can you can solve for it.

1752
01:53:52,640 --> 01:53:58,400
You can ask, well, what would be the optimal cognitive strategy for solving

1753
01:53:58,400 --> 01:53:59,040
that problem?

1754
01:53:59,440 --> 01:54:04,000
And we did that in a in a by setting up a toy model.

1755
01:54:04,560 --> 01:54:10,560
So we had sort of a simple agent-based model, and in that agent-based model,

1756
01:54:10,800 --> 01:54:16,610
the the agent, there's one agent that is basically doing a series of very

1757
01:54:16,610 --> 01:54:18,050
simple theory of mind tasks.

1758
01:54:18,130 --> 01:54:23,090
So there's a like a false belief task, there's a task when you need to predict what someone

1759
01:54:23,730 --> 01:54:27,090
who has knowledge of a fact is going to do and so on and so forth.

1760
01:54:27,570 --> 01:54:33,570
And then and this you can sort of uh then optimize the cognitive strategy that this

1761
01:54:33,570 --> 01:54:35,090
agent is going to use.

1762
01:54:36,320 --> 01:54:42,000
And when we did that, we find that indeed if your agent is sufficiently constrained in

1763
01:54:42,000 --> 01:54:47,840
terms of its information processing capacities, then it will use a strategy that is very

1764
01:54:47,840 --> 01:54:53,360
similar to the one that I just described, where it will kind of abandon any pretension of

1765
01:54:53,360 --> 01:54:54,880
creating meta representations.

1766
01:54:55,040 --> 01:55:00,670
Instead, it will just have its own physical description of the world, and then it will add

1767
01:55:00,910 --> 01:55:06,270
a kind of a binary tag about whether or like this fact is known or not known by the other

1768
01:55:06,270 --> 01:55:06,590
person.

1769
01:55:08,830 --> 01:55:14,350
Is one way to think about it then that um so like I don't I'm not a um

1770
01:55:15,470 --> 01:55:17,630
uh I don't know, I'm not like left-handed.

1771
01:55:17,710 --> 01:55:22,110
So like let's say there's a guitar a right-handed guitar and a left-handed guitar, and I

1772
01:55:22,110 --> 01:55:27,180
reach for the right-handed guitar and then I I somebody else, I see somebody else making

1773
01:55:27,180 --> 01:55:30,140
that same decision, but they're left-handed, they reach for the left-handed guitar.

1774
01:55:30,380 --> 01:55:36,220
My model of predicting what they would have done is wrong when I just use what I would

1775
01:55:36,220 --> 01:55:36,540
do.

1776
01:55:36,860 --> 01:55:37,180
Right.

1777
01:55:37,420 --> 01:55:42,780
So then I have to say, I have to have an entry for what's the minimal extra information I

1778
01:55:42,780 --> 01:55:48,690
need to add to still use me as the machine for

1779
01:55:48,690 --> 01:55:53,650
predicting behavior, but then change one variable to generate the right prediction.

1780
01:55:53,890 --> 01:55:59,570
Is one idea that, like, all these kinds of things of like what a false belief actually is,

1781
01:55:59,730 --> 01:56:05,650
or what a motivation is, or a personality variable, they're all just sort of settings

1782
01:56:05,810 --> 01:56:11,660
on what's the minimal thing I need to add to my own inference engine about

1783
01:56:11,660 --> 01:56:15,420
what I would do next to then just predict the behavior of others.

1784
01:56:15,580 --> 01:56:21,180
That in that sort of all these, all these representations that we think of are just sort of

1785
01:56:21,180 --> 01:56:26,140
uh just slight tweaks to this otherwise very complicated system that we don't have to

1786
01:56:26,140 --> 01:56:28,380
repeat and duplicate.

1787
01:56:28,620 --> 01:56:34,540
Um so it's just that for for for Tedig and for David, I just have my own pre,

1788
01:56:35,090 --> 01:56:39,570
it's like an equalizer, you know, and it's sort of like I have the presets for David such

1789
01:56:39,570 --> 01:56:45,490
that when I think of David or I see David, I can pull those up and I know my own

1790
01:56:45,490 --> 01:56:47,170
self-simulation system.

1791
01:56:48,130 --> 01:56:53,730
I need to add these 20 or 10,000 tweaks to all those little inference

1792
01:56:53,730 --> 01:56:57,250
engines, but that's only little 10 10,000 bits, right?

1793
01:56:57,330 --> 01:57:00,050
It's just small tweaks to the the big machine.

1794
01:57:00,290 --> 01:57:01,500
Is that a fair?

1795
01:57:01,980 --> 01:57:05,180
I mean, I realize it's sort of abstraction, but is that a fair way to characterize the

1796
01:57:05,740 --> 01:57:06,140
picture here?

1797
01:57:06,620 --> 01:57:11,740
I think it's it's uh what you are describing is is a more general extension of the general

1798
01:57:11,740 --> 01:57:17,020
principle that we should like to what the extent that we can, the the system should be

1799
01:57:17,340 --> 01:57:22,860
minimizing the computational complexity uh by using all tricks that it has available.

1800
01:57:23,260 --> 01:57:29,020
But the the the coincide behind our model is that um even before we are trying to predict

1801
01:57:29,020 --> 01:57:34,750
the behavior of others, We have our own model of the physical world

1802
01:57:34,910 --> 01:57:36,990
because it is, I mean, it's super useful.

1803
01:57:37,710 --> 01:57:41,630
It's very useful to know what the world looks like.

1804
01:57:43,070 --> 01:57:48,110
And the nice thing about other people is that they also live in the same world.

1805
01:57:48,590 --> 01:57:53,470
And so by virtue of the fact that you're living in the same world, you will end up creating

1806
01:57:53,470 --> 01:57:55,950
representations that are actually quite similar.

1807
01:57:56,680 --> 01:58:01,400
And so this is a kind of a nice fact about the problem of social prediction is that you're

1808
01:58:01,400 --> 01:58:06,040
predicting the behavior of agents that also live in the same the s the same world.

1809
01:58:06,440 --> 01:58:12,440
And so it means that to a large extent you can use your own understanding of w of what

1810
01:58:12,440 --> 01:58:15,480
the world looks like in order to predict their their behavior.

1811
01:58:16,040 --> 01:58:18,730
It also means you can tell when an error occurs, right?

1812
01:58:18,810 --> 01:58:22,890
Because that's one of the other problems, is like a lot of the machine learning, what the

1813
01:58:22,890 --> 01:58:27,050
pro the trick is to know when an error occurs, and that's a discrepancy between what they

1814
01:58:27,050 --> 01:58:31,050
do, which you have data for, and what you would do, which you also have data for.

1815
01:58:31,530 --> 01:58:32,090
Yes, yes.

1816
01:58:32,330 --> 01:58:37,130
And so really like really from this point of view, yes, the the main challenge is to uh um

1817
01:58:37,370 --> 01:58:43,130
to compute what what are those other those aspects of the world as I represent it that

1818
01:58:43,130 --> 01:58:45,130
other people also are aware of.

1819
01:58:45,980 --> 01:58:51,980
And this is a so I mean an interesting thing about that is that um this this model, even

1820
01:58:51,980 --> 01:58:55,900
though it is relatively simple, it makes quite interesting predictions.

1821
01:58:56,300 --> 01:59:01,740
Because you can look at the for a particular setting of the the parameters, you can look at

1822
01:59:01,740 --> 01:59:05,260
the optimal cognitive strategy that an agent comes up with.

1823
01:59:05,660 --> 01:59:11,260
And you can see well to what extent does this actually describe the behavior of

1824
01:59:11,420 --> 01:59:13,020
actual organisms.

1825
01:59:13,310 --> 01:59:18,510
And here a particularly interesting case is when we look at non-human primates.

1826
01:59:18,670 --> 01:59:24,670
Okay, because non-human primates um they are much more uh constrained in their

1827
01:59:24,670 --> 01:59:27,310
information processing capacities as we are.

1828
01:59:27,710 --> 01:59:33,070
And maybe they don't have to use theory of mind to such an extent that we do because um

1829
01:59:33,470 --> 01:59:39,080
the the sort of social demands of their ecology are are not as as

1830
01:59:39,080 --> 01:59:39,720
strong.

1831
01:59:41,000 --> 01:59:46,680
And so you can you can sort of set up an agent

1832
01:59:46,920 --> 01:59:52,040
that has not a lot of information processing capacity for the purpose of theory of mind.

1833
01:59:57,190 --> 02:00:02,150
And then you can look at the strategy that this simulated agent is using.

1834
02:00:03,110 --> 02:00:08,710
And one prediction that follows from that is that they will behave in a certain way

1835
02:00:09,270 --> 02:00:14,950
in tasks that are a little bit like a false belief task, except that you have

1836
02:00:14,950 --> 02:00:15,910
changed one thing.

1837
02:00:15,990 --> 02:00:19,510
So this task is called the accidental true belief task.

1838
02:00:19,750 --> 02:00:25,660
So this is a task where I so suppose I I hide

1839
02:00:25,660 --> 02:00:28,700
an apple in front of Dave.

1840
02:00:28,780 --> 02:00:34,540
So like you're aware that I'm hiding an apple in this in this one box.

1841
02:00:35,980 --> 02:00:36,940
And then you go away.

1842
02:00:37,340 --> 02:00:41,260
And while you're away, I will remove the apple from that box.

1843
02:00:41,420 --> 02:00:43,980
But then I put it back in the same box.

1844
02:00:44,190 --> 02:00:44,430
Okay.

1845
02:00:44,590 --> 02:00:49,870
So what it means is that you you have a belief that the apple is in that box.

1846
02:00:50,670 --> 02:00:54,670
But the there's a sense in which a belief is kind of lucky.

1847
02:00:55,230 --> 02:00:59,070
Like because it was just uh a chance event that I put it back in the same box.

1848
02:00:59,310 --> 02:01:03,550
Or um and to make it, let's say you know you you you remove it from the box and put it on

1849
02:01:03,550 --> 02:01:06,910
the windowsill and then a bird magically flies in and picks it up and puts it back in the

1850
02:01:06,910 --> 02:01:06,990
box.

1851
02:01:07,230 --> 02:01:07,390
Yes.

1852
02:01:09,380 --> 02:01:10,180
Another tree that's inner.

1853
02:01:12,500 --> 02:01:12,740
Yeah.

1854
02:01:12,900 --> 02:01:15,220
So we just really got lucky.

1855
02:01:15,700 --> 02:01:17,940
A forest fire that was caused by a lightning sunway.

1856
02:01:18,420 --> 02:01:18,660
Yeah.

1857
02:01:18,900 --> 02:01:23,060
Um and so in that situation something surprising happened, which was that kind of the

1858
02:01:23,060 --> 02:01:26,980
simulated agent um sort of stops making any prediction.

1859
02:01:27,220 --> 02:01:30,180
It was like, oh, like the situation is too complicated to me for me.

1860
02:01:30,260 --> 02:01:32,980
I'm just going to throw my hands up in the air.

1861
02:01:33,190 --> 02:01:38,950
and basically uh i'm going to predict that you you will look that i that dave will look in

1862
02:01:38,950 --> 02:01:44,790
any any of the boxes i'm not going to make any bet about where dave is going to look um

1863
02:01:44,950 --> 02:01:50,550
and it turns out that this is exactly how monkeys behave if you give them

1864
02:01:50,790 --> 02:01:56,680
uh that that kind of task And so so it so it seems in that

1865
02:01:56,680 --> 02:02:02,520
there's a lot of similar experiments that are done on non-human primates, uh, where they do

1866
02:02:02,520 --> 02:02:07,480
things that look a little bit weird, um, but that are quite well predicted by this kind of

1867
02:02:07,720 --> 02:02:11,800
version of the model where you assume that the agent does not have a lot of information

1868
02:02:11,800 --> 02:02:12,920
processing capacity.

1869
02:02:14,430 --> 02:02:15,390
Interesting.

1870
02:02:16,670 --> 02:02:18,510
Well, fascinating stuff.

1871
02:02:18,590 --> 02:02:22,990
I feel like uh I could keep picking your brain on this for hours, but we are um approaching

1872
02:02:22,990 --> 02:02:23,870
the end of our time.

1873
02:02:23,950 --> 02:02:28,590
And it and uh is there any topic that you wish we would have talked about that we that we

1874
02:02:28,590 --> 02:02:29,550
should have asked you about?

1875
02:02:29,790 --> 02:02:35,790
Anything that uh Oh yeah, um I think yeah, I think the another thing on which we could

1876
02:02:35,790 --> 02:02:41,640
have spent a lot of time is this concept of uh intentional action that is also related to

1877
02:02:41,640 --> 02:02:45,560
theory of mind that it's its own uh its own thing.

1878
02:02:46,200 --> 02:02:51,080
Well here's here's I would I'd like to ask how did you uh discover we haven't done origin

1879
02:02:51,240 --> 02:02:52,120
stories in a while.

1880
02:02:52,200 --> 02:02:58,200
Um how did you come across of psychological Yeah um so I've always been quite interested

1881
02:02:58,200 --> 02:03:00,840
in the general question of how the mind works.

1882
02:03:01,080 --> 02:03:05,990
Um but I I didn't really know it back then when I was in high school.

1883
02:03:06,230 --> 02:03:10,150
Um and so I decided to because I was also interested in in history.

1884
02:03:10,230 --> 02:03:14,310
Um and so I decided to to do political science as a major.

1885
02:03:14,550 --> 02:03:20,550
Uh and that's when in the process of majoring in in political science, I I read this

1886
02:03:20,550 --> 02:03:21,670
book by David Hume.

1887
02:03:21,990 --> 02:03:23,990
And back to Edinburgh.

1888
02:03:24,470 --> 02:03:25,190
Full circle.

1889
02:03:25,270 --> 02:03:25,430
Yeah.

1890
02:03:25,910 --> 02:03:31,200
And so I was kind of blown away by the fact that oh this guy was asking just deep questions

1891
02:03:31,200 --> 02:03:35,440
about about the mind and it made me realize well uh this is actually something that I find

1892
02:03:35,440 --> 02:03:41,280
super interesting and so I ended up double majoring in in philosophy um and

1893
02:03:41,280 --> 02:03:46,720
this was this is what led me to do uh then a master's in cognitive science in in Lyon.

1894
02:03:47,280 --> 02:03:52,480
Uh and and the the master's was was interesting there was a lot of interesting stuff about

1895
02:03:52,560 --> 02:03:56,640
you know behavioral neuroscience, cognitive psychology uh and so on and so forth.

1896
02:03:56,800 --> 02:04:02,790
Um but while taking those courses, I always had a nagging feeling that there was

1897
02:04:02,790 --> 02:04:05,910
a set of important questions that were not asked.

1898
02:04:06,310 --> 02:04:12,230
Um and then of course like now I know that I was um internally

1899
02:04:12,390 --> 02:04:16,230
complaining about the fact that there were there was not much of an evolutionary

1900
02:04:16,230 --> 02:04:18,230
perspective being being uh offered.

1901
02:04:18,390 --> 02:04:23,030
Um and then it came time to find uh an internship, like a research internship for my

1902
02:04:23,190 --> 02:04:23,590
master.

1903
02:04:23,740 --> 02:04:27,660
and there was this new researcher uh at the university called Pascal Boyer.

1904
02:04:28,140 --> 02:04:31,980
Uh so I reached out to him and like can I attend your lab meeting?

1905
02:04:32,140 --> 02:04:38,140
Um and then during the lab meeting he's sort of like laid out the perspective

1906
02:04:38,140 --> 02:04:42,700
for what his research over the next few few years what was going to be and I had this kind

1907
02:04:42,700 --> 02:04:46,220
of the this click of oh this this is what I want to do.

1908
02:04:47,180 --> 02:04:51,580
And and and uh a year later I was at UCSV working with John Anita.

1909
02:04:52,270 --> 02:04:53,070
That's fantastic.

1910
02:04:53,230 --> 02:04:55,150
So wait, Pascal was where?

1911
02:04:55,790 --> 02:04:57,790
So he he was in Lyon at the time.

1912
02:05:00,190 --> 02:05:02,430
He had gotten a grant to stay there for a few years.

1913
02:05:02,750 --> 02:05:03,310
Fantastic.

1914
02:05:03,470 --> 02:05:03,870
Yeah.

1915
02:05:04,110 --> 02:05:05,950
We have to have Pascal on.

1916
02:05:06,430 --> 02:05:07,870
Oh, I can't believe you haven't had it.

1917
02:05:08,270 --> 02:05:08,830
I know, right?

1918
02:05:08,910 --> 02:05:09,390
Yes.

1919
02:05:10,590 --> 02:05:10,910
Yeah.

1920
02:05:11,150 --> 02:05:14,430
Well, Latik, thank you so much for being on.

1921
02:05:14,750 --> 02:05:17,240
Um my brain is full.

1922
02:05:18,920 --> 02:05:22,600
I've I've exhausted the computational capacity in my head.

1923
02:05:25,320 --> 02:05:26,040
Yeah, this is great.

1924
02:05:26,440 --> 02:05:27,240
This is our pleasure, David.

1925
02:05:27,320 --> 02:05:29,000
It was a really fun conversation.

1926
02:05:32,680 --> 02:05:35,240
David, what causes a good podcast?

1927
02:05:35,640 --> 02:05:36,600
Uh well, hold on.

1928
02:05:36,680 --> 02:05:40,280
Let me run through all the counterfactual worlds in my head.

1929
02:05:40,590 --> 02:05:41,630
Uh computing.

1930
02:05:42,350 --> 02:05:42,910
Copper wires.

1931
02:05:43,150 --> 02:05:44,030
Copper wires.

1932
02:05:44,750 --> 02:05:45,070
Yes.

1933
02:05:45,230 --> 02:05:49,310
Oxygen in the atmosphere is definitely a a critical component to a good podcast.

1934
02:05:49,390 --> 02:05:52,110
If there was no oxygen, then uh we wouldn't be alive.

1935
02:05:52,190 --> 02:05:58,190
So that so um yeah uh so leave us a kind review and

1936
02:05:58,190 --> 02:06:04,190
talk about um all the things all the things that could have gone wrong such that you

1937
02:06:04,190 --> 02:06:05,790
wouldn't have discovered this podcast.

1938
02:06:06,270 --> 02:06:09,300
Yeah that's a great content for your kind review.

1939
02:06:09,460 --> 02:06:15,060
Um, you can also visit us on Patreon and that might have a causal impact on the

1940
02:06:15,060 --> 02:06:16,900
continuation of this podcast.

1941
02:06:17,140 --> 02:06:18,260
Maybe we can afford some.

1942
02:06:19,700 --> 02:06:25,380
This is very uh it feels very Well, we're just giving you ways you can

1943
02:06:25,540 --> 02:06:26,580
intervene on the world.

1944
02:06:26,740 --> 02:06:27,300
Yeah, exactly.

1945
02:06:29,990 --> 02:06:32,470
you have you have free will to do any of these things.

1946
02:06:32,710 --> 02:06:35,990
Yeah, but it there's we unintended consequences to things too.

1947
02:06:36,150 --> 02:06:38,070
So you know, be careful what you do.

1948
02:06:38,470 --> 02:06:39,190
That's true.

1949
02:06:39,350 --> 02:06:40,070
That's true.

1950
02:06:40,310 --> 02:06:43,590
Um uh I'll see you next week, David.

1951
02:06:43,830 --> 02:06:44,790
Yep, sounds good.

1952
02:06:44,950 --> 02:06:46,310
All right, well see everybody.

1953
02:06:46,390 --> 02:06:46,550
Okay.

1954
02:06:46,950 --> 02:06:47,270
See ya.