WEBVTT

00:00:03.497 --> 00:00:10.497
This is the Convergent Science Network podcast. Leading researchers in the domain

00:00:10.497 --> 00:00:16.737
of neuroscience, brain theory and technology are interviewed by Paul Vichure and Tony Prescott.

00:00:21.297 --> 00:00:29.357
So this is Paul Vichure for the Convergent Science Network podcast together with Tony Prescott.

00:00:30.017 --> 00:00:34.617
Today, we're speaking with Aaron Scherner. Welcome, Aaron, to our podcast.

00:00:36.037 --> 00:00:42.837
You were speaking today in our BCBT summer school, and you very much delved

00:00:42.837 --> 00:00:47.917
into the physiology of behavior, of volition, of free will.

00:00:48.697 --> 00:00:52.217
So do you think we can say anything meaningful about free will at all from a

00:00:52.217 --> 00:00:53.237
sort of scientific perspective?

00:00:54.677 --> 00:00:59.277
I think in theory it's possible to do so, but I'm not sure if we've said anything meaningful yet.

00:01:00.997 --> 00:01:04.477
Okay um but now free will

00:01:04.477 --> 00:01:08.057
is a constant complex uh concept as

00:01:08.057 --> 00:01:11.697
she also indicated to your talk um so is

00:01:11.697 --> 00:01:14.997
that the construct is that is that the natural category of

00:01:14.997 --> 00:01:18.057
a follow ryle is that what we should be investigating is

00:01:18.057 --> 00:01:21.117
that the right starting point or would you rephrase it

00:01:21.117 --> 00:01:24.197
and use a concept like volition or decision

00:01:24.197 --> 00:01:27.057
making i i i tend to

00:01:27.057 --> 00:01:30.837
use the the the phrase conscious will uh

00:01:30.837 --> 00:01:33.897
because yeah i think i think free will

00:01:33.897 --> 00:01:37.217
is something that if you if you take the

00:01:37.217 --> 00:01:39.957
the hardcore definition of free

00:01:39.957 --> 00:01:43.717
will which is uh some uncaused cause uh

00:01:43.717 --> 00:01:49.157
that's not something you can really grapple with uh scientifically but you can

00:01:49.157 --> 00:01:55.257
ask questions about the relationship between conscious events and actions and

00:01:55.257 --> 00:02:00.697
ask if there's any kind of causal relationship between those two.

00:02:02.154 --> 00:02:06.854
But in some sense, then you first have to also grapple with the so-called heart

00:02:06.854 --> 00:02:10.114
problem, right? Because now you link it to conscious states.

00:02:10.274 --> 00:02:14.754
And so how do we access these conscious states in an independent way?

00:02:15.734 --> 00:02:20.394
Yeah, that's, I mean, that I think is a direction that we're going in.

00:02:21.094 --> 00:02:27.394
It's not something that anyone has done in the past for that very reason.

00:02:27.394 --> 00:02:30.414
Uh but we do now have some

00:02:30.414 --> 00:02:33.934
uh very reliable correlates

00:02:33.934 --> 00:02:37.054
of conscious perception and so we

00:02:37.054 --> 00:02:40.514
could potentially start looking at those uh as

00:02:40.514 --> 00:02:43.334
indicators of the the presence of a conscious

00:02:43.334 --> 00:02:47.234
intention and ask whether or not it has the right relationship with the subsequent

00:02:47.234 --> 00:02:51.594
action or at least the neural activity that that brings the action about it's

00:02:51.594 --> 00:02:54.994
still you're very cautious is right now yeah how you approach this and i understand

00:02:54.994 --> 00:02:59.874
that it's it's a conceptual minefield so exactly but still you have to leave

00:02:59.874 --> 00:03:02.794
some of the mystery to have something to explain.

00:03:03.614 --> 00:03:06.574
So have you just scraped off too much now of

00:03:06.574 --> 00:03:09.994
of the free will volition phenomenon well we

00:03:09.994 --> 00:03:13.554
we we scraped away maybe we

00:03:13.554 --> 00:03:16.434
scraped away the the mystery about the meaning of this

00:03:16.434 --> 00:03:20.554
uh signature the the readiness potential uh which

00:03:20.554 --> 00:03:24.534
we thought uh was a sign of a decision uh

00:03:24.534 --> 00:03:27.434
and uh what what we've shown is that uh

00:03:27.434 --> 00:03:30.394
this may not be the sign of a decision right it

00:03:30.394 --> 00:03:33.214
it it uh might be the sign of

00:03:33.214 --> 00:03:36.514
a random process uh and if we if if

00:03:36.514 --> 00:03:39.814
we're using this uh this phenomenon

00:03:39.814 --> 00:03:44.254
this readiness potential as a temporal marker against which to to to compare

00:03:44.254 --> 00:03:51.274
the time of other events well uh we we We may have been deceiving ourselves

00:03:51.274 --> 00:03:56.934
by interpreting it in that way, interpreting it as a decision.

00:03:57.874 --> 00:04:01.234
But now we, so we already raised that quite a bit, right? But just first,

00:04:01.314 --> 00:04:03.134
I don't want to be too flippant about it.

00:04:03.954 --> 00:04:06.414
You know, you might still believe it is flippant, but.

00:04:07.493 --> 00:04:11.193
You said, well, instead of free will, I want to speak of conscious will, right?

00:04:11.473 --> 00:04:15.913
But you left the notion will in there as a construct, right?

00:04:16.593 --> 00:04:21.673
Can we do anything with this construct will? Also from a physiological perspective,

00:04:21.913 --> 00:04:26.953
as a physiologist, is will a useful construct? Can we do anything with that?

00:04:27.793 --> 00:04:32.193
Well, if it maps onto something like an intention or an urge,

00:04:32.313 --> 00:04:35.553
something, then yes, I think so.

00:04:35.553 --> 00:04:46.233
So I use the word will in this context to be almost identical with an intention.

00:04:47.553 --> 00:04:53.913
Which is, you could say, a mental state or a neural state that represents a

00:04:53.913 --> 00:04:56.593
commitment to some course of action.

00:04:56.913 --> 00:05:03.173
An urge is sort of, in the way that I think people would think about it,

00:05:03.193 --> 00:05:04.433
is almost something which is unconscious.

00:05:04.433 --> 00:05:08.173
Unconscious you know i felt an urge you know i don't know where it came from but,

00:05:08.773 --> 00:05:12.633
something urged me to go to do this um so the

00:05:12.633 --> 00:05:15.933
origin of the urge the origin of the urge i guess is unconscious but

00:05:15.933 --> 00:05:19.173
the urge itself is so so the

00:05:19.173 --> 00:05:21.973
uh so there's an interesting question you

00:05:21.973 --> 00:05:25.233
could talk about talk about sort of unconscious will potentially as

00:05:25.233 --> 00:05:29.893
well yeah in fact that's something uh it's very interesting that you bring that

00:05:29.893 --> 00:05:34.213
up because it's just it's something that i I was just recently debating with

00:05:34.213 --> 00:05:41.553
colleagues and with some philosophers who really like this idea of the concept

00:05:41.553 --> 00:05:43.993
of an unconscious intention.

00:05:45.633 --> 00:05:49.673
But my problem with it is how would you know an unconscious intention if you saw one?

00:05:51.173 --> 00:05:56.153
I don't know. Behavior? Action? Arousal?

00:05:57.193 --> 00:05:59.753
Maybe. me well i think it's it speaks to the

00:05:59.753 --> 00:06:02.513
whole question of agency you know the

00:06:02.513 --> 00:06:06.473
fact that we have an experience of being agents is

00:06:06.473 --> 00:06:09.253
why we have this idea of free will in

00:06:09.253 --> 00:06:12.153
the first place you know if we did things without

00:06:12.153 --> 00:06:15.913
feeling we were making choices we we

00:06:15.913 --> 00:06:20.333
wouldn't need to have will but we will explains

00:06:20.333 --> 00:06:23.473
why we have a uh a feeling of agency so

00:06:23.473 --> 00:06:28.253
i guess and that feeling of agency is both unconscious and

00:06:28.253 --> 00:06:31.753
there is a conscious version of it i think it's

00:06:31.753 --> 00:06:34.933
important though to distinguish agency from volition

00:06:34.933 --> 00:06:41.633
uh where so age i think we can distinguish between the two i think we we can

00:06:41.633 --> 00:06:47.573
say that agency is the the uh that doesn't have to be the feeling but the knowledge

00:06:47.573 --> 00:06:52.993
that my action caused some effect in the world,

00:06:54.213 --> 00:06:59.853
whereas volition is that my thought caused my action.

00:07:01.273 --> 00:07:09.173
Uh my my desire or my intention uh was responsible for bringing that action about.

00:07:09.993 --> 00:07:13.133
You also usually know this distinction between let's say secondary and

00:07:13.133 --> 00:07:15.973
or primary secondary desires right so you you might

00:07:15.973 --> 00:07:20.533
have an urge but you must actually desire to execute or follow up on that urge

00:07:20.533 --> 00:07:27.493
but so it's a multi-level process that you commit yourself to and so so in an

00:07:27.493 --> 00:07:34.453
And unconscious or subconscious urge or desire would never be identified as such by the agent.

00:07:34.573 --> 00:07:36.693
It just plays out underneath the radar.

00:07:36.933 --> 00:07:41.753
Right. Right. So, yeah, you always require by necessity this idea of a secondary

00:07:41.753 --> 00:07:46.853
or a primary sort of level of urges or desires.

00:07:47.233 --> 00:07:52.773
Right. And then what you also need is an agent that actually owns these desires.

00:07:53.013 --> 00:07:55.693
But then there's also what's called the reason responsiveness.

00:07:55.813 --> 00:07:59.393
Right. so that you can deliberate on that desire.

00:08:00.153 --> 00:08:06.433
So these are ingredients you would need to then lead up to volition. To volition, right.

00:08:06.593 --> 00:08:11.033
Reason's responsiveness is an important one that I think in the past has been

00:08:11.033 --> 00:08:17.153
largely neglected, but that now everyone is talking about, which I think it's quite fair.

00:08:17.233 --> 00:08:21.913
It's true that if we want to give something this label of volition,

00:08:21.913 --> 00:08:28.573
it should have this property of responding to reasons this is right and of course

00:08:28.573 --> 00:08:33.533
this is also your trajectory right because you move to link the study of volition

00:08:33.533 --> 00:08:38.313
more closely with decision making which of course goes in the direction of.

00:08:39.453 --> 00:08:40.573
Deliberation and reasoning.

00:08:42.395 --> 00:08:45.315
But the other ingredient you need is the ability to do otherwise.

00:08:46.695 --> 00:08:53.535
So if I act following my urge and desire, I must also be able to understand

00:08:53.535 --> 00:08:56.235
that I have alternatives. I could have done otherwise.

00:08:56.575 --> 00:09:01.035
Right, I could have done otherwise. Yeah, this is an important one as well.

00:09:01.355 --> 00:09:06.275
Yeah, and one that's very important from a philosophical point of view,

00:09:06.275 --> 00:09:14.095
But it's very difficult empirically to show that you could have done otherwise.

00:09:15.015 --> 00:09:23.295
Absolutely. Yeah. Okay, so this is this conceptual environment of free will,

00:09:23.475 --> 00:09:26.255
which is complex and minefield, as you said.

00:09:26.255 --> 00:09:29.735
That, but it's now, now you, you cut through that with, with,

00:09:29.735 --> 00:09:34.695
with a laser in some sense, by really zooming in on a very specific physiological

00:09:34.695 --> 00:09:40.835
feature of, of voluntary action, which is this readiness potential that was discovered,

00:09:41.615 --> 00:09:46.995
in the sixties, um, so why, why do you think the readiness potential as such

00:09:46.995 --> 00:09:52.115
is, is actually telling us something about, about free will, um,

00:09:52.735 --> 00:09:55.895
I don't think it is. I think that's the issue, really.

00:09:56.035 --> 00:10:00.455
I don't think it tells us anything about free will one way or another.

00:10:01.935 --> 00:10:05.375
I think it tells us something about how the motor system works,

00:10:05.755 --> 00:10:12.035
some potentially interesting things, but I don't think it's a temporal marker

00:10:12.035 --> 00:10:16.835
for a decision or a temporal marker for a commitment. it.

00:10:17.595 --> 00:10:24.215
Because this would go back to these classical experiments on the Bonitov potential,

00:10:24.975 --> 00:10:29.835
that then fed into the Libet experiments, and you laid this out for us in quite

00:10:29.835 --> 00:10:33.715
some detail, the Cornelio-Brandicchi 60s experiments.

00:10:34.355 --> 00:10:40.275
And the big mystery that appeared with Libet was like, well, if I look at.

00:10:41.981 --> 00:10:46.101
The neural signals or the neural, the correlates, the neural correlates of decision-making,

00:10:46.281 --> 00:10:50.981
and I compare that to people's ability to declare where they are in the decision-making

00:10:50.981 --> 00:10:53.141
process in terms of making decisions or committing,

00:10:53.641 --> 00:10:57.261
to actually make the action, it is, there's, there's a delay, right?

00:10:57.301 --> 00:11:01.081
So, so the brain only knows what it's going to do before you know it,

00:11:01.161 --> 00:11:04.981
which is of course already implying a dualism in some sense, right?

00:11:05.541 --> 00:11:11.181
So, so are you saying it was a surprise? There was no surprise? rise. Come again?

00:11:12.361 --> 00:11:18.201
When you say it doesn't tell us much about evolution, people are tremendously

00:11:18.201 --> 00:11:20.021
impressed with this result, right?

00:11:20.381 --> 00:11:25.101
There have been many attempts to replicate it, and no one could actually have

00:11:25.101 --> 00:11:28.621
any successful replications of it, and it's very much accepted as a physiological

00:11:28.621 --> 00:11:30.981
feature of decision-making.

00:11:31.241 --> 00:11:34.521
Yeah, it replicates extremely well. Yeah, exactly. So in that sense,

00:11:34.621 --> 00:11:36.501
it's a physiological fact. Yeah.

00:11:36.661 --> 00:11:43.401
Yeah. But then a huge amount of trees have been killed to produce the papers,

00:11:43.601 --> 00:11:48.901
the paper on which many words have been written now about what this means with respect to free will.

00:11:49.021 --> 00:11:52.341
Because well, the brain knows what you're going to do before you know it,

00:11:52.381 --> 00:11:54.161
and therefore, there's no free will.

00:11:56.261 --> 00:12:00.981
So where did that interpretation go wrong? Because, okay, you have many methodological

00:12:00.981 --> 00:12:02.261
concerns, right, we can look at.

00:12:02.261 --> 00:12:08.661
But in some sense why were people so easily misled about this at the time,

00:12:09.461 --> 00:12:12.741
well because it it it reliably

00:12:12.741 --> 00:12:19.161
very reliably precedes self-initiated movement and so the the i think the first

00:12:19.161 --> 00:12:24.401
and it may be sensible conclusion to come to is that that this is uh this is

00:12:24.401 --> 00:12:30.921
the brain's way of getting ready to initiate a movement of of of planning and

00:12:30.921 --> 00:12:32.461
in preparation for movement.

00:12:33.181 --> 00:12:38.441
But of course, the kind of data that we use to get it out, it's all correlational.

00:12:39.961 --> 00:12:44.181
So you could say, well, we've made the mistake of, the simple mistake of confusing

00:12:44.181 --> 00:12:45.721
correlation for causation.

00:12:47.201 --> 00:12:51.581
It's a bit more than that, but it's that as well. Right.

00:12:51.981 --> 00:12:57.761
But isn't there also a hidden prior in this whole debate at this stage,

00:12:57.801 --> 00:13:00.121
right, before we delve into physiology physiology and the methodology.

00:13:01.822 --> 00:13:05.702
That there is an implicit dualism in the interpretation.

00:13:06.042 --> 00:13:12.642
There is this whole idea like, oh, every action at any timescale has to be caused

00:13:12.642 --> 00:13:16.562
by a conscious state, has to be preceded, therefore, by a conscious state,

00:13:16.742 --> 00:13:19.342
which seems an arbitrary assumption.

00:13:19.342 --> 00:13:23.162
Yeah, well, no, I think with the readiness potential, what people are saying

00:13:23.162 --> 00:13:26.982
is that every action, or at least every voluntary self-initiated action,

00:13:27.062 --> 00:13:33.162
has to be caused by this same preceding neural state,

00:13:33.362 --> 00:13:34.782
not necessarily conscious.

00:13:34.782 --> 00:13:37.682
Conscious right um so i think the conscious

00:13:37.682 --> 00:13:40.622
bit you can you can you can separate that out and in fact and

00:13:40.622 --> 00:13:44.362
and and we intentionally evaded the

00:13:44.362 --> 00:13:47.862
topic of consciousness when we first started working with this because we just

00:13:47.862 --> 00:13:54.342
wanted to ask is this signature this readiness potential the cause of the movement

00:13:54.342 --> 00:14:00.302
does this represent the brain getting ready to move uh and uh i think that's

00:14:00.302 --> 00:14:02.922
what we That's what we, if you want to say,

00:14:03.062 --> 00:14:05.242
debunked in a way.

00:14:06.302 --> 00:14:12.442
Okay, so here was the redness potential. It's a slow buildup of activity that

00:14:12.442 --> 00:14:14.642
peaks just before you initiate a movement.

00:14:16.562 --> 00:14:23.462
It precedes your ability to declare where you are in a position in the decision-making process.

00:14:23.462 --> 00:14:29.662
And so the first way you debunked that story is to say, well,

00:14:29.782 --> 00:14:32.282
there's no unitary process underlying this.

00:14:32.822 --> 00:14:38.962
In some sense, it's an artifact of just averaging many observations, right?

00:14:39.022 --> 00:14:44.582
Because the idea would be that if you have spontaneous fluctuations in the brain,

00:14:46.022 --> 00:14:50.222
and these spontaneous fluctuations are sort of slightly biased by some form

00:14:50.222 --> 00:14:56.802
of evidence, let's say, perceptual evidence about a task or an internally generated cue, then the,

00:14:57.880 --> 00:15:02.440
Over many trials, this will go to some average state. That will exactly look

00:15:02.440 --> 00:15:04.180
like what you call a maintenance potential.

00:15:04.580 --> 00:15:10.320
But at heart, it is essentially just an integration of a highly noisy signal.

00:15:10.620 --> 00:15:14.800
But even on single trials, it's the case. You don't really have to average trials

00:15:14.800 --> 00:15:15.920
to run into this problem.

00:15:15.920 --> 00:15:24.020
Problem um when you when you select a single trial based on the time of the movement itself,

00:15:24.900 --> 00:15:32.800
um you've in a way selected a biased sample right because that that little piece

00:15:32.800 --> 00:15:39.300
of data that you're looking at ends with a movement uh and if you want to understand

00:15:39.300 --> 00:15:42.240
how movement works You also want to know what happens when there is no movement.

00:15:43.540 --> 00:15:50.960
So, of course, in this case, because you have this biased sample,

00:15:51.280 --> 00:15:55.200
you see everything through the lens of, well, this is what happens before a movement.

00:15:56.420 --> 00:15:59.380
And you're you will if you

00:15:59.380 --> 00:16:02.360
if you look at that you'll recover in the average or on

00:16:02.360 --> 00:16:05.340
individual trials a tendency for there to be a a

00:16:05.340 --> 00:16:10.700
ramping phenomenon ramp up to in on the assumption that this is a threshold

00:16:10.700 --> 00:16:15.860
crossing type of phenomenon right and you also test that hypothesis right by

00:16:15.860 --> 00:16:21.500
asking your subjects to respond as quickly as they could where you would queue

00:16:21.500 --> 00:16:24.340
you would queue the response, right?

00:16:24.380 --> 00:16:25.680
And then you would sort of link

00:16:25.680 --> 00:16:29.980
the queuing to where they would be in that integration process, right?

00:16:30.040 --> 00:16:33.100
So you would assume there's some decision threshold, and if they're further

00:16:33.100 --> 00:16:35.820
away from the decision threshold, you would predict if you now force somebody

00:16:35.820 --> 00:16:39.960
to respond, the reaction time should be longer than they are forced to respond

00:16:39.960 --> 00:16:41.140
when they're closer in this direction.

00:16:41.340 --> 00:16:44.900
When they just happen to be closer, right? Of course, yeah. And this then just

00:16:44.900 --> 00:16:49.660
depends on this sort of highly variable fluctuation.

00:16:49.720 --> 00:16:52.440
Right, right, right. There's a lot of data fluctuation who sourced you,

00:16:52.540 --> 00:16:54.880
you don't know about. So-

00:16:56.449 --> 00:17:00.429
So with that, you came to this idea that you could explain the readiness potential,

00:17:01.229 --> 00:17:06.029
and also the performance from the perspective of a so-called drift diffusion model.

00:17:06.189 --> 00:17:09.289
Actually, I integrate this noisy signal.

00:17:09.489 --> 00:17:12.689
When I hit threshold, I'm going. It's highly variable.

00:17:13.309 --> 00:17:16.429
Averaging across trials gives me something that looks like this readiness potential.

00:17:16.429 --> 00:17:19.869
Potential but um then when

00:17:19.869 --> 00:17:24.609
you tested that on your on your subjects you could

00:17:24.609 --> 00:17:27.889
show that that your model that implements this kind

00:17:27.889 --> 00:17:30.869
of integration of a noisy fluctuation when you

00:17:30.869 --> 00:17:34.109
compared that to the eg signal you got from your from your

00:17:34.109 --> 00:17:36.949
subjects that it gave

00:17:36.949 --> 00:17:41.429
you gave you a similar kind of response right so you see that indeed if you

00:17:41.429 --> 00:17:46.249
compare fast and slow response trials you see that this integration process

00:17:46.249 --> 00:17:50.369
has reached a different level and has some offset that's the difference between

00:17:50.369 --> 00:17:54.129
these two traces that you extract yeah and importantly there's a difference,

00:17:54.769 --> 00:17:57.989
uh not just before the movement but before the queue you make

00:17:57.989 --> 00:18:01.009
the movement and and so these are these these uh

00:18:01.009 --> 00:18:03.849
interruptions are random i don't even

00:18:03.849 --> 00:18:06.749
know when the interruptions were going to happen nobody the computer only knows

00:18:06.749 --> 00:18:14.529
right so uh what you see as a as a as a difference in electrical potential preceding

00:18:14.529 --> 00:18:19.069
the cue can't possibly be a preparatory process because you can't prepare for

00:18:19.069 --> 00:18:21.689
a movement that you don't know you're going to make. Right.

00:18:23.032 --> 00:18:28.912
That's an important insight from this model that indeed there is no preparation.

00:18:30.912 --> 00:18:37.912
But if you compare the model to the actual physiology, there is indeed a difference in the offset.

00:18:38.612 --> 00:18:44.052
But in terms of the details of how the trace evolves, it's not identical.

00:18:45.012 --> 00:18:48.832
So if you would really lay the traces on top of each other, I would do some

00:18:48.832 --> 00:18:49.892
sort of correlation measure.

00:18:50.052 --> 00:18:53.492
I would not get one. Right. It's qualitative.

00:18:54.392 --> 00:18:59.052
What the model gives you relative to the reality is qualitative.

00:18:59.372 --> 00:19:06.252
Right. So for instance, the model gives a much stronger decay in the trace than

00:19:06.252 --> 00:19:09.332
you observe in your subjects.

00:19:09.452 --> 00:19:15.872
While for the fast trials, it seems very constant in the model,

00:19:15.872 --> 00:19:21.612
practically constant, the level, while in your subject it is still sort of sloping

00:19:21.612 --> 00:19:23.852
down. There's some form that you get, right? So the difference is here.

00:19:24.032 --> 00:19:28.172
Now, this in itself is not a criticism because model doesn't have to be identical

00:19:28.172 --> 00:19:30.972
to what you measure, but when is a model good enough?

00:19:31.232 --> 00:19:35.932
Right, so to what extent did you feel this model was good enough to explain that data?

00:19:36.232 --> 00:19:41.792
Well, I think it was good enough because it made a novel prediction,

00:19:41.792 --> 00:19:45.332
prediction uh and albeit qualitative

00:19:45.332 --> 00:19:47.952
but it it it told us we should find a

00:19:47.952 --> 00:19:51.672
difference between fast and slow responses uh in

00:19:51.672 --> 00:19:55.312
this particular direction uh and that

00:19:55.312 --> 00:20:00.732
we found now the fact that it doesn't map on to the reality exactly that leaves

00:20:00.732 --> 00:20:07.792
uh some more questions for us to grapple with uh in the next iteration of experience

00:20:07.792 --> 00:20:11.632
but it also leaves the door open and in some sense in the discussion part of

00:20:11.632 --> 00:20:12.592
your talk, as we came to that,

00:20:13.192 --> 00:20:16.852
I could also say, well, maybe the model that explains the data could even be simpler.

00:20:17.972 --> 00:20:22.652
Maybe it doesn't need to be drift-to-fusion. Right. I gave an example, yeah.

00:20:22.812 --> 00:20:28.232
Another example, maybe I could have just an oscillator that can exist in two modes, right?

00:20:28.452 --> 00:20:31.092
And it's a high-energy mode and a low-energy mode.

00:20:31.392 --> 00:20:33.852
And this would then account for the difference between the two traces.

00:20:34.572 --> 00:20:37.972
And it completely depends on intrinsic property and this flip it between the two.

00:20:38.852 --> 00:20:44.092
Right. It would be a completely solipsistic model, but in terms of the physiology,

00:20:44.392 --> 00:20:47.292
I could also account for this difference in offset, right? Right.

00:20:47.472 --> 00:20:50.432
So, why this drift-to-fusion? So, drift-to-fusion is not the simplest model.

00:20:51.741 --> 00:20:56.801
Um, well, you can go simpler. I mean, what you need, what I said,

00:20:56.881 --> 00:20:59.901
uh, after the talk, when we were, when we were discussing this,

00:21:00.001 --> 00:21:03.901
what you need at a minimum is pink noise, how you get that pink noise.

00:21:04.721 --> 00:21:08.801
Maybe it, maybe you can just say, well, I don't know how, I don't care how I got it. It's just there.

00:21:09.961 --> 00:21:14.561
Uh, the drift diffusion model gives you a sort of principled way that's grounded

00:21:14.561 --> 00:21:18.921
in prior research that gives, that gives you this pink noise. Right.

00:21:20.161 --> 00:21:23.601
Um, but then. Okay, so now we have an alternative explanation,

00:21:23.981 --> 00:21:27.721
right, of the retinence potential.

00:21:29.101 --> 00:21:35.361
So how many studies have really confirmed, you think, in your mind, the model you propose?

00:21:35.861 --> 00:21:43.841
Five or six different studies have come out that support the theory in one way or another.

00:21:43.841 --> 00:21:47.601
Another um so an important one

00:21:47.601 --> 00:21:50.981
was the 2014 study by murakami with rats

00:21:50.981 --> 00:21:54.501
who found a ramping like activity uh in

00:21:54.501 --> 00:21:57.101
the premotor cortex of rats when they were

00:21:57.101 --> 00:22:00.441
doing a task where they could basically spontaneously stop

00:22:00.441 --> 00:22:03.881
waiting for a big reward and just go immediately get

00:22:03.881 --> 00:22:06.961
a small reward correct yeah okay

00:22:06.961 --> 00:22:11.581
yeah you mentioned you mentioned that that result was um but

00:22:11.581 --> 00:22:14.361
ramping activity as such is that sufficient as a

00:22:14.361 --> 00:22:17.841
signature well it's okay so uh i

00:22:17.841 --> 00:22:21.001
spoke very quickly it was not just ramping activity but

00:22:21.001 --> 00:22:24.261
it was ramping activity that consistently reached

00:22:24.261 --> 00:22:29.601
the same level just at the moment that the that the rat left the waiting station

00:22:29.601 --> 00:22:36.341
and went to get the reward uh so it at least it's very consistent with the the

00:22:36.341 --> 00:22:41.441
idea of a of an accumulator of the this would be the output put of an accumulator, right?

00:22:41.521 --> 00:22:43.621
We would expect it to look just like that.

00:22:45.818 --> 00:22:51.818
I mean, you pointed to some interesting evidence from crayfish and rat as well,

00:22:52.618 --> 00:22:58.998
that I guess suggests that this readiness potential is something that's common to brains. Yeah.

00:22:59.278 --> 00:23:07.418
And therefore, probably if the crayfish is doing it, what we normally think

00:23:07.418 --> 00:23:10.598
of as conscious volition isn't going to be a factor.

00:23:11.478 --> 00:23:18.058
Yeah, I would agree with you. I mean, and you also talked about the idea that

00:23:18.058 --> 00:23:22.318
this experiment, actually, there is a weak imperative, you said, to me.

00:23:22.818 --> 00:23:25.618
So it's almost that if people are making a conscious choice,

00:23:25.698 --> 00:23:29.478
it's when they walk into the experiment and agree to do this task. Exactly.

00:23:29.598 --> 00:23:34.478
And the task is designed so that specifically you're asked to suspend volition

00:23:34.478 --> 00:23:37.138
and wait for that urge to come.

00:23:37.458 --> 00:23:42.278
So it's really a task that's weighted against having any conscious volition

00:23:42.278 --> 00:23:43.818
that we would normally think of.

00:23:44.018 --> 00:23:48.738
As conscious volition you know sort of choosing to

00:23:48.738 --> 00:23:51.658
buy a house or move country or something like these are

00:23:51.658 --> 00:23:55.178
conscious choices um whereas

00:23:55.178 --> 00:24:01.058
deciding to lift your finger in this task or whatever is is it's really designed

00:24:01.058 --> 00:24:06.498
to to remove anything but the most minimal of conscious choice yeah i think

00:24:06.498 --> 00:24:11.358
so i mean it's to that extent is it really that surprising i mean for you the

00:24:11.358 --> 00:24:15.038
result isn't uh surprising but i guess you thought though the.

00:24:16.398 --> 00:24:21.958
Why why had the scientific community become so obsessed with this yourself yeah

00:24:21.958 --> 00:24:27.538
well i think it i mean it is it is just a matter of lifting your finger it is

00:24:27.538 --> 00:24:30.458
a very yeah it is a very uh simple,

00:24:32.078 --> 00:24:33.078
act um.

00:24:37.022 --> 00:24:41.682
Yeah, I'm sorry, I lost my train of thought. Well, I mean, people have pointed

00:24:41.682 --> 00:24:44.782
to Labette as evidence against conscious control.

00:24:44.922 --> 00:24:48.382
But if you're a defender of conscious control, you can just say this experiment

00:24:48.382 --> 00:24:52.242
isn't representative of what we mean by conscious will.

00:24:52.502 --> 00:24:55.962
You know, it's the minimal amount of conscious will.

00:24:56.062 --> 00:25:01.422
So even if Labette was right and you were wrong, this doesn't really tell us

00:25:01.422 --> 00:25:04.362
anything about conscious volitional actions.

00:25:04.922 --> 00:25:10.402
Yeah, I would agree. I mean, you need a signature of a decision,

00:25:10.602 --> 00:25:16.402
and then you need to know the relationship between the time of that decision

00:25:16.402 --> 00:25:19.582
and the time of your conscious decision.

00:25:20.022 --> 00:25:24.542
And if that signature is not reliable or doesn't mean what you think it means,

00:25:24.702 --> 00:25:26.482
then you can't make those inferences.

00:25:26.482 --> 00:25:31.542
So given that that's the case, that the Lebet experiment isn't a useful experiment

00:25:31.542 --> 00:25:35.922
for doing this, and the readiness potential isn't a useful signal,

00:25:36.082 --> 00:25:41.542
what would be another way of getting at conscious volition that we could imagine

00:25:41.542 --> 00:25:44.062
experimentally that would be more powerful?

00:25:44.062 --> 00:25:49.042
I think a more powerful way that I alluded to at the end of the talk is to use

00:25:49.042 --> 00:25:55.862
a closed loop feedback system, or if you want to call it a brain-computer interface,

00:25:56.182 --> 00:26:03.382
to drive some external signal directly from cortical activity.

00:26:04.402 --> 00:26:10.882
And then you can ask questions about what happens when the feedback from,

00:26:10.942 --> 00:26:14.382
you might say an intention,

00:26:14.562 --> 00:26:22.162
if you will, comes far earlier than you expected it to, far earlier than your brain expected it to.

00:26:22.642 --> 00:26:28.142
And that can tell you some things about conscious, at least the conscious feeling of volition.

00:26:30.649 --> 00:26:34.389
But still, it would then depend on reportability, you know.

00:26:35.189 --> 00:26:39.069
You'd have to give a report, yeah. Yeah. Yeah, so it doesn't,

00:26:39.069 --> 00:26:44.529
in the context of saying no report paradigm, this doesn't work. Right.

00:26:46.049 --> 00:26:52.609
But the other thing is that, you like this more common result, which is interesting.

00:26:52.829 --> 00:26:57.169
Okay, we see some integration to threshold. We see how nicely lined up before

00:26:57.169 --> 00:27:03.889
this sort of self-initiated action to move from one port in the task as a mouse

00:27:03.889 --> 00:27:06.469
to the feeding port or the reward port, okay?

00:27:07.549 --> 00:27:13.949
But look at premotor core type, right? Well, usually also in this whole debate

00:27:13.949 --> 00:27:20.489
on demolition, people point more to, let's say, medial frontal structures, SMA.

00:27:20.829 --> 00:27:24.529
So more advanced in this hierarchy, right?

00:27:24.589 --> 00:27:28.349
So is it not an issue that we're getting a little bit, let's

00:27:28.349 --> 00:27:33.969
say unclear but also localization of these phenomena no I think the area of

00:27:33.969 --> 00:27:41.129
cortex that they were looking at in the rat m2 I think is the rodent is the

00:27:41.129 --> 00:27:45.989
rat homologue of the SMA if I'm not mistaken all right,

00:27:46.970 --> 00:27:53.770
so we have a reinterpretation of the Leavitt experiment and sometimes you're

00:27:53.770 --> 00:27:58.750
saying it's less magical than it looks like but still there is some source of

00:27:58.750 --> 00:28:00.930
let's say spontaneous activity that comes from somewhere,

00:28:01.810 --> 00:28:07.050
so is the spontaneous activity an echo of something that you might want to call,

00:28:07.690 --> 00:28:14.190
volition a sound agent that is trying to manipulate this or are you really thinking

00:28:14.190 --> 00:28:18.170
about but just noise that is floating around in neural circuits.

00:28:18.550 --> 00:28:21.010
I think it's the latter. I think it's just noise.

00:28:23.010 --> 00:28:29.510
I think you need more than that to explain and to account for volition.

00:28:30.270 --> 00:28:34.810
So what are the properties of neural noise in these carnival circuits?

00:28:36.150 --> 00:28:39.710
Well, one of the most important properties is that it's temporally autocorrelated.

00:28:39.710 --> 00:28:42.690
Uh and that uh that is

00:28:42.690 --> 00:28:45.770
key in order to give you the the kind

00:28:45.770 --> 00:28:49.090
of result that we it's temporary auto-correlated temporally auto-correlated

00:28:49.090 --> 00:28:51.730
in another way of saying it's pink it's pink

00:28:51.730 --> 00:28:56.330
noise it's not white noise uh so how much physiological evidence is there that

00:28:56.330 --> 00:29:02.250
there's pink noise in in these parts of the brain there's there's noise in the

00:29:02.250 --> 00:29:06.930
brain and in behavior tends to be pink uh in fact i would turn that around and

00:29:06.930 --> 00:29:09.790
challenge you to find white noise Anywhere in the brain.

00:29:10.390 --> 00:29:14.250
Sure. I would go more to the other direction. I'm not a big believer in noise.

00:29:14.630 --> 00:29:18.410
Noise just means there's a source of variability that you haven't identified yet.

00:29:18.630 --> 00:29:22.010
Ah, yes. So this is more where I was going with that.

00:29:22.230 --> 00:29:26.350
So to say pink noise, I think just means, well, we lump a lot of things together

00:29:26.350 --> 00:29:28.830
and it looks like pink noise.

00:29:29.010 --> 00:29:33.470
Right? But it either means there's some dynamical state that evolves with the memory.

00:29:35.052 --> 00:29:38.972
So as long as you have a dynamical system with some kind of memory,

00:29:39.172 --> 00:29:42.272
then you have your pink noise.

00:29:42.852 --> 00:29:47.652
So in the way you decompose the render's potential into, let's say,

00:29:47.692 --> 00:29:52.732
multiple highly variable traces, maybe they themselves can again be decomposed

00:29:52.732 --> 00:29:57.332
in something more mechanistic that we can understand that's more deterministic than pink noise itself.

00:29:58.032 --> 00:30:02.852
Is that reasonable? Or you think there's really a a pink noise source in the brain somewhere.

00:30:03.232 --> 00:30:06.972
Oh, I see. No, I wouldn't say that there's a pink noise source in the brain

00:30:06.972 --> 00:30:08.452
somewhere, but it isn't.

00:30:08.452 --> 00:30:13.172
I think one of the interesting things about the model is that this kind of simple

00:30:13.172 --> 00:30:20.312
accumulator, its output has a 1 over F power spectrum. It's pink.

00:30:22.172 --> 00:30:28.652
And that, as I mentioned before, that helps to sort of ground it in some neurophysiology

00:30:28.652 --> 00:30:35.172
that has been well characterized in perception research, in decision making research in general.

00:30:35.332 --> 00:30:39.612
So we have an instance where a very different kind of decision,

00:30:39.752 --> 00:30:44.372
a decision that is not made on the basis of a stimulus, at least not one that

00:30:44.372 --> 00:30:45.392
you have right there at hand,

00:30:45.912 --> 00:30:50.912
is governed maybe by the same kind of mechanism that all decisions are.

00:30:51.732 --> 00:30:55.932
The brain didn't have to reinvent the wheel for this kind of decision.

00:30:57.350 --> 00:31:01.310
So now we have a reinterpretation of the readiness potential.

00:31:01.790 --> 00:31:06.390
In some sense, you've deconstructed a notion of volition.

00:31:06.590 --> 00:31:09.930
And you said, well, we can just think about it in the same way we think about decision-making.

00:31:10.350 --> 00:31:13.190
And it doesn't really matter whether you want to call it volition or not.

00:31:13.210 --> 00:31:16.210
It's a decision-making process that we're looking at. Yeah, yeah.

00:31:16.790 --> 00:31:20.490
And then there are more recent results. And you point, for instance,

00:31:20.630 --> 00:31:25.350
to this work by Fried, but also Soon and others.

00:31:26.510 --> 00:31:32.330
Who then started to add more of the black box science to the whole story.

00:31:33.030 --> 00:31:38.810
So now we're going to say, okay, can I, as opposed to do a post-op analysis,

00:31:39.210 --> 00:31:45.510
I was going to say, can I predict decisions given some classifier extracting

00:31:45.510 --> 00:31:47.550
features from my physiology?

00:31:48.050 --> 00:31:51.170
And I couldn't do this in a free case on single cells. Right.

00:31:51.530 --> 00:31:56.750
And in the case of Soon, you can do that looking at f and y.

00:31:57.090 --> 00:32:00.850
Right. Right. So again, it was a bit of a game changer, right?

00:32:00.850 --> 00:32:01.910
It's this whole black box approach.

00:32:02.770 --> 00:32:07.570
And then surprisingly, these approaches seem to work.

00:32:07.690 --> 00:32:15.390
They could predict what the decision was going to be by the subject with some

00:32:15.390 --> 00:32:17.590
improved performance as compared to random.

00:32:18.990 --> 00:32:22.730
For Freed, it was 75% or something like this, right?

00:32:22.970 --> 00:32:28.450
And they could predict 500 milliseconds before the action was initiated of what the subject would do.

00:32:28.670 --> 00:32:31.630
But for Soon, it was even more extreme, that's with Dylan Hines.

00:32:32.750 --> 00:32:36.490
More extreme, it could be up to six or six to eight seconds or something like

00:32:36.490 --> 00:32:39.490
that, some extreme time window. With fMRI, yeah.

00:32:39.790 --> 00:32:44.390
And then the point was, okay, it almost seems to violate the basic principles

00:32:44.390 --> 00:32:46.030
of the universe, right? Because now I can.

00:32:48.168 --> 00:32:52.828
Look into the future. Yeah, I mean, what I think would have violated the basic

00:32:52.828 --> 00:32:58.288
principles of the universe would be if you couldn't classify what someone was

00:32:58.288 --> 00:33:00.068
going to do based on prior brain activity.

00:33:01.068 --> 00:33:07.808
That would be strange, right? As if that decision just emerged out of ether.

00:33:09.248 --> 00:33:13.148
But say, seconds might be a bit long. They can predict it before you even get the cue.

00:33:14.248 --> 00:33:17.728
Yeah, well, I mean, let's say you're talking about a right or a left-hand movement.

00:33:18.168 --> 00:33:22.148
I mean, it's, it's, it doesn't seem that unusual to me to think that several

00:33:22.148 --> 00:33:26.308
seconds beforehand, your brain might be in a state that, uh,

00:33:26.488 --> 00:33:30.168
will tend to bias it slightly towards one or the other.

00:33:31.288 --> 00:33:34.868
Um, so I, I, I don't find that to be too hard to believe.

00:33:35.368 --> 00:33:38.488
So also eight seconds, you would find not too hard to believe in,

00:33:38.488 --> 00:33:39.988
in the case of the SUD experiments.

00:33:40.828 --> 00:33:45.108
Yeah. I mean, no, I don't find it too hard to believe. Okay.

00:33:45.628 --> 00:33:47.408
Okay. Okay, but that's interesting, right? Because with that,

00:33:47.428 --> 00:33:52.148
you're saying that these fluctuations that would sort of bias you in one way

00:33:52.148 --> 00:33:57.088
or the other have really a rather long history in the dynamics of the brain.

00:33:57.408 --> 00:34:03.728
Yeah, I guess that's what that points to, is that if it is related to the phenomenon

00:34:03.728 --> 00:34:10.348
of autocorrelation, so it has a relatively sluggish time constant.

00:34:10.728 --> 00:34:14.228
That would also reduce the states that the system can then occupy.

00:34:16.068 --> 00:34:20.308
Yeah, although bear in mind that you're classifying slightly better than chance.

00:34:21.308 --> 00:34:27.908
I forget exactly how we're at 58% or 60% correct, where 50% is just a random guess. Right.

00:34:29.128 --> 00:34:34.348
It would be a very different story, I guess, if it was 80% or 90%. Right.

00:34:34.928 --> 00:34:38.628
So when you came across this free-to-result with the single cells using the

00:34:38.628 --> 00:34:40.188
support vector machine,

00:34:40.748 --> 00:34:45.168
or or soon with this fmi classifier at

00:34:45.168 --> 00:34:48.068
the time were you were you shocked by that or surprised or this

00:34:48.068 --> 00:34:50.988
was also already done for you within the realm

00:34:50.988 --> 00:34:54.668
of the expected um when i

00:34:54.668 --> 00:35:01.668
when i saw freed result uh i i wasn't too surprised uh i was already at that

00:35:01.668 --> 00:35:06.948
time i was already working on i was already doing this work uh and had already

00:35:06.948 --> 00:35:11.948
come to a lot of these conclusions when i When I saw the work of Soon,

00:35:12.248 --> 00:35:17.088
that was much earlier, that was in 2008, I think.

00:35:18.168 --> 00:35:20.508
Yeah, I was initially pretty blown away by that.

00:35:22.328 --> 00:35:23.288
And later...

00:35:25.861 --> 00:35:29.161
Well sort of revised uh revised my

00:35:29.161 --> 00:35:32.161
my way of interpreting uh the results

00:35:32.161 --> 00:35:34.981
yeah right because in the second in the last part of your talk

00:35:34.981 --> 00:35:39.281
you focus very much on also the methodological challenges that you face with

00:35:39.281 --> 00:35:44.521
this kind of analysis right and uh and you pointed out the number of caveats

00:35:44.521 --> 00:35:49.561
like for instance uh if you use a sliding time window to make your your estimates

00:35:49.561 --> 00:35:54.301
is you better be very clear where you put the reference to estimate.

00:35:54.541 --> 00:35:58.581
Yeah, you want to align your time axis to the leading edge of the window,

00:35:58.741 --> 00:36:03.141
not the middle or, God forbid, the tail end of the window, right?

00:36:03.181 --> 00:36:06.301
Because you don't want your classifier to be able to peek into the very future

00:36:06.301 --> 00:36:07.261
that it's trying to predict.

00:36:07.741 --> 00:36:10.581
Right, exactly. That would be cheating. Yeah, absolutely.

00:36:12.321 --> 00:36:17.521
And then you'll also look at setting up the right reference conditions,

00:36:17.521 --> 00:36:20.881
like look at conditions without movement, the width movement, right?

00:36:21.041 --> 00:36:25.541
Yeah, yeah. Those are all the old correlation in the signals that we're analyzing.

00:36:25.801 --> 00:36:32.641
So when all these caveats point to the fact that they're overestimating these time windows,

00:36:34.201 --> 00:36:40.661
if there's an error, the error is an overestimation of a time window in which

00:36:40.661 --> 00:36:41.521
you can reliably predict.

00:36:42.181 --> 00:36:45.021
You mean, if I understand you correctly, you're

00:36:45.021 --> 00:36:48.481
saying we're overestimating how far back yeah yeah

00:36:48.481 --> 00:36:51.961
i'm not sure it means that um because

00:36:51.961 --> 00:36:54.701
i think uh for example in the soon study they weren't really

00:36:54.701 --> 00:36:58.921
using a sliding window they were just classifying at each at each tr at each

00:36:58.921 --> 00:37:03.841
point in time before the movement um no but still they would have another correlation

00:37:03.841 --> 00:37:12.021
to then take care of yes although i believe they were they they did one reanalysis of some of those data,

00:37:12.121 --> 00:37:17.381
trying to rule out the possibility that this was just driven by autocorrelation.

00:37:18.541 --> 00:37:24.041
And at least their conclusion was that it couldn't be entirely accounted for by autocorrelation.

00:37:25.381 --> 00:37:30.061
But that, of course, leaves open the possibility that it could partly be. Right.

00:37:31.181 --> 00:37:38.461
But in your own analysis that you presented to also show the relevance of these

00:37:38.461 --> 00:37:40.101
adjustments to the analysis method,

00:37:40.381 --> 00:37:44.121
you showed that the whole shape of the, that you already called the scourge

00:37:44.121 --> 00:37:48.381
of the banana, the whole shape of this randomness potential starts to disappear, right?

00:37:48.441 --> 00:37:51.581
And at this transition point, at which you can start to predict,

00:37:51.921 --> 00:37:56.961
it looks much more discrete or very, very punctuated in time. Yeah.

00:37:57.161 --> 00:38:02.221
It's just less gradual. When you have properly controlled conditions where, as I mentioned before,

00:38:02.521 --> 00:38:08.361
if you want to talk about predicting the onset of movement, you want to have

00:38:08.361 --> 00:38:12.421
data that include movements and you want to also have data that don't include

00:38:12.421 --> 00:38:15.301
movements as a control, right?

00:38:15.301 --> 00:38:17.981
Right. And ideally, you'd want to have these two and compare them.

00:38:18.081 --> 00:38:25.061
So in a more recent experiment that we're now just working on getting published, we did just that.

00:38:25.141 --> 00:38:29.361
So we used an experimental paradigm where you end up with data epochs that either

00:38:29.361 --> 00:38:34.361
terminate with a movement or terminate without a movement, but are well matched in other respects.

00:38:34.821 --> 00:38:41.141
And when you apply a sliding window analysis using pattern classifier to that kind of data.

00:38:41.941 --> 00:38:45.221
You're using a very powerful

00:38:45.301 --> 00:38:49.001
classifier uh we couldn't tell apart movement from

00:38:49.001 --> 00:38:51.901
non-movement epochs until the very last moment

00:38:51.901 --> 00:38:54.941
until until basically the moment at which the movement was was

00:38:54.941 --> 00:38:58.121
was beginning uh and it certainly was was

00:38:58.121 --> 00:39:03.901
no fault of the classifier itself because at and after the time of movement

00:39:03.901 --> 00:39:08.821
the classifier was at at nearly perfect performance uh so you couldn't say oh

00:39:08.821 --> 00:39:12.001
well it's that's just because your classifier isn't good enough well appears

00:39:12.001 --> 00:39:15.941
to be good enough in some sense isn't the consequence of that.

00:39:17.796 --> 00:39:22.956
The shape that we gave to the randomness potential is more an artifact of,

00:39:23.156 --> 00:39:31.976
let's say, insufficiently tuned methods than that's really in the signal of the brain itself.

00:39:31.976 --> 00:39:37.176
Like also your examples, if you have a corrected signal,

00:39:37.836 --> 00:39:46.236
if you take out the autocorrelation effect, that suddenly the meaningful part

00:39:46.236 --> 00:39:51.496
of your signal starts to look very different.

00:39:51.676 --> 00:39:55.416
It's much more than, let's say, an S-shaped kind of response,

00:39:55.816 --> 00:40:00.296
a more compressed transition point than the traditional randomness potential.

00:40:00.296 --> 00:40:06.616
So having you with that sort of deconstructed the randomness potential as an

00:40:06.616 --> 00:40:09.836
artifact of insufficient methods. I think you can say that, yeah.

00:40:09.936 --> 00:40:17.316
I think you can call it an artifact. If you want an artifact of time-locking,

00:40:17.356 --> 00:40:22.936
at least a tendency to time-lock to crests in an autocorrelated time series. Right, yeah.

00:40:23.316 --> 00:40:26.436
And if you time-lock to crests in an autocorrelated time series,

00:40:26.596 --> 00:40:31.596
you recover the autocorrelation function, which looks like a slow curve. Right, exactly.

00:40:34.696 --> 00:40:38.376
So you've talked about having the right kinds of controls,

00:40:39.596 --> 00:40:46.316
Patrick Haggart who was here last year did a big experiment I'm sure you know

00:40:46.316 --> 00:40:53.096
his work on this has developed some quite beautifully controlled paradigms for comparing,

00:40:53.736 --> 00:40:55.456
volitional with stimulus driven,

00:40:56.716 --> 00:41:01.076
responses I think he was using sort of random noise movement and you had to

00:41:01.076 --> 00:41:02.916
say which correctional dots were going in.

00:41:03.496 --> 00:41:11.296
And his study, I think, also made it a bit more sort of mattered whether or

00:41:11.296 --> 00:41:15.036
not you, what response you gave, because there was actually a monetary reward.

00:41:15.316 --> 00:41:20.216
So there's a bit more of an incentive to do the right thing or to do the most rational thing.

00:41:20.536 --> 00:41:25.096
And his data suggested that in the case of volitional actions,

00:41:25.196 --> 00:41:27.736
you see a reduction in the amount of noise.

00:41:28.096 --> 00:41:34.376
That's right. You see a slow reduction in variability over time as you approach the movement.

00:41:35.236 --> 00:41:39.416
So what do you think of that result, and how would you interpret it?

00:41:39.696 --> 00:41:43.856
Well, I would really interpret it in the same way. That is to say that that

00:41:43.856 --> 00:41:48.396
signal, that variability,

00:41:49.616 --> 00:41:56.656
can vary in much the same way that the time series can vary in an autocorrelated fashion.

00:41:57.976 --> 00:42:04.716
So the same logic that gets you the readiness potential in the experiments I did could get you this.

00:42:06.957 --> 00:42:11.217
Let's say decrease in variability but why would it be different between the

00:42:11.217 --> 00:42:20.137
volitional and the instructed uh trials so well because the the uh in in the volitional case,

00:42:20.877 --> 00:42:28.377
uh the time the precise time of the movement right uh tends to be uh biased

00:42:28.377 --> 00:42:33.757
slightly by so it depends on this underlying process right whereas the cued

00:42:33.757 --> 00:42:36.037
in in the case of a instructed,

00:42:37.437 --> 00:42:38.357
uh, movement.

00:42:38.797 --> 00:42:47.937
That's not the case. So that's one of the key aspects of this explanation is that, uh, when the, uh,

00:42:48.177 --> 00:42:52.277
the way I say it is that when the imperative to move is weak or absent,

00:42:52.457 --> 00:42:57.597
meaning when this precise moment at which you move is not dictated by some stimulus,

00:42:57.797 --> 00:43:03.997
uh, then you have this question, okay, uh, I'm going to move approximately now, right?

00:43:04.237 --> 00:43:08.617
Uh, and so here goes, as I move, and you're left with this question of,

00:43:08.617 --> 00:43:15.537
well, why did I move just right then when I moved and not 200 milliseconds earlier or later, right?

00:43:15.597 --> 00:43:19.017
There is some freedom there to play with.

00:43:19.137 --> 00:43:24.017
And if the answer to that question is, well, there are spontaneous fluctuations

00:43:24.017 --> 00:43:28.177
in neural activity and they bias the precise moment at which you move,

00:43:28.317 --> 00:43:33.857
then if that's the case, then you're guaranteed to recover a slow gradual buildup.

00:43:35.999 --> 00:43:40.799
But I don't really see how this answers Tony's challenge because Tony's referring

00:43:40.799 --> 00:43:44.559
to the Peter Haggard claiming... Patrick. Patrick, sorry.

00:43:46.919 --> 00:43:52.699
Voluntary control, lower variability. Cube control, higher variability, right?

00:43:52.819 --> 00:43:55.259
Or no reduction in variability. Okay.

00:43:55.459 --> 00:44:00.699
But then cube control, then we can go back to the literal decision-making.

00:44:00.879 --> 00:44:04.599
When you start to drive, you sort of drive the integrators, right?

00:44:04.599 --> 00:44:08.239
And as soon as you start to add a drive to these integrators,

00:44:08.379 --> 00:44:11.379
you start to overcome the more spontaneous fluctuation.

00:44:11.459 --> 00:44:14.239
So you would expect to see lower variability there.

00:44:15.079 --> 00:44:19.059
So I don't really see how the explanation... Well, that's what you get.

00:44:19.139 --> 00:44:20.339
You get a decrease in variability.

00:44:21.439 --> 00:44:25.279
No, it was a voluntary case of this lower variability. Right, right. Yeah.

00:44:25.519 --> 00:44:28.299
Well, I was making the example, if you go to the decision-making literature

00:44:28.299 --> 00:44:30.599
where everything is cubed, right?

00:44:30.759 --> 00:44:35.259
Right. Then also for your model, if I start to drive your model with an external

00:44:35.259 --> 00:44:39.039
input coming from an external queue, I would start to reduce variability.

00:44:39.839 --> 00:44:47.819
Yeah, and you would abolish this early tail readiness potential.

00:44:48.159 --> 00:44:50.479
So this is, okay, my claim, and this one goes used now.

00:44:51.179 --> 00:44:54.979
If you take the Haggard challenge, you say, okay, volition, higher variability.

00:44:56.139 --> 00:44:58.999
No, volition, lower variability, queued, higher variability.

00:44:58.999 --> 00:45:02.479
But if you take your standard drift diffusion model with some,

00:45:02.599 --> 00:45:06.199
let's say, noise if I now start to get an external Q as an additional input,

00:45:06.359 --> 00:45:09.459
I start to reduce the noise effect right?

00:45:09.539 --> 00:45:13.239
So isn't that the encountering to it from this drift diffusion perspective?

00:45:14.679 --> 00:45:20.259
It's true that it does pose a bit of a challenge but I think it's not.

00:45:21.999 --> 00:45:26.899
I think the case is not at all closed by that evidence because again Again,

00:45:26.899 --> 00:45:31.919
that variability could itself be fluctuating in a random way.

00:45:34.159 --> 00:45:38.379
Okay, we're leaving a bit of a cop-out now. Well, essentially,

00:45:39.059 --> 00:45:46.079
I mean, that's one, I would say, to be fair, that's one of the weaknesses of

00:45:46.079 --> 00:45:49.719
the model or the explanation is that you can always keep saying,

00:45:49.859 --> 00:45:53.299
well, that could also vary, right?

00:45:53.339 --> 00:45:56.719
That could also fluctuate spontaneously. Simultaneously. So if somebody comes

00:45:56.719 --> 00:45:58.159
up with a new phenomenon.

00:45:58.279 --> 00:46:01.939
Oh, we just have another free parameter, right? And now we have the fluctuation we need.

00:46:02.139 --> 00:46:04.419
You said, well, that could fluctuate as well.

00:46:04.639 --> 00:46:09.079
But maybe another response could be, those to the Tony's challenge,

00:46:10.079 --> 00:46:12.639
in some of the people always ignore that, that,

00:46:14.033 --> 00:46:19.873
the agent itself is represented, is sending a signal to a decision-making stage.

00:46:20.313 --> 00:46:23.113
Mm-hmm. Right? So, you get a vision that the parietal areas,

00:46:23.473 --> 00:46:29.113
right, around the temporal parietal junction that are implicated in states of

00:46:29.113 --> 00:46:33.873
self and agency are generating internal cues, right?

00:46:34.093 --> 00:46:37.533
Right, right, right. So, in some sense, so the argument would then be internal

00:46:37.533 --> 00:46:41.373
cues have a higher gain, let's say, than externally generated cues.

00:46:41.553 --> 00:46:45.213
And then, that way, I might account for it. But that's what you said.

00:46:45.613 --> 00:46:52.873
But I think Patrick's result doesn't imply necessarily that that's due to conscious will, as you said.

00:46:52.933 --> 00:46:56.313
It could be due to some other internal attentional process.

00:46:56.373 --> 00:46:58.773
He's agnostic. Patrick would be agnostic about that.

00:47:00.893 --> 00:47:06.733
So it doesn't necessarily help the people who want to claim a role for conscious

00:47:06.733 --> 00:47:08.653
volition, which I think Patrick is one of those.

00:47:10.073 --> 00:47:18.633
So it could be some other unconscious that's happening, maybe it's linked to attention in some way.

00:47:19.493 --> 00:47:21.133
Is attention relevant there?

00:47:22.553 --> 00:47:26.773
I don't know, why do you bring up attention? It's very confusing.

00:47:26.973 --> 00:47:31.233
Because you're in a task where you're having to make a decision.

00:47:31.433 --> 00:47:36.073
So you're focusing on, am I going to, in his task, interrupt this trial,

00:47:36.173 --> 00:47:39.293
go to the next trial? Well, that's something you're fixating on.

00:47:39.613 --> 00:47:42.633
And also you're instructed. And you're instructed to be. Yeah,

00:47:42.633 --> 00:47:46.333
well, there's money involved. So you make more money if you make good choices.

00:47:46.593 --> 00:47:52.913
So that would be a reason for being more focused, which might reduce your neural noise.

00:47:54.473 --> 00:47:57.433
Okay, that's fair enough. But then...

00:47:59.639 --> 00:48:05.899
So now we're having a problem right because we have many problems but let's start with

00:48:06.199 --> 00:48:11.539
aaron's problem first and so

00:48:11.539 --> 00:48:14.339
so okay great we do fantastically well here's libid

00:48:14.339 --> 00:48:18.119
look very confusing conscious will is an illusion uh

00:48:18.119 --> 00:48:21.279
what's the causal relation between conscious states and

00:48:21.279 --> 00:48:24.479
action and so on right you resolve that conundrum

00:48:24.479 --> 00:48:27.139
by saying well you just missed you have misinterpreted your

00:48:27.139 --> 00:48:30.559
signal right essentially right yeah well what's uh what's the

00:48:30.559 --> 00:48:33.519
relation between the readiness potential and action yeah sure

00:48:33.519 --> 00:48:36.279
yeah so you have said look you have over

00:48:36.279 --> 00:48:39.799
interpreted an artifact yeah right

00:48:39.799 --> 00:48:44.559
and you can you can reduce the whole thing into very single model where we just

00:48:44.559 --> 00:48:48.679
have spontaneous fluctuations that that have a little memory so they can add

00:48:48.679 --> 00:48:53.019
up to to some decisions i showed right and it brings in the realm of the standard

00:48:53.019 --> 00:48:59.019
a decision yeah I mean it accounts for the data in a parsimonious way right but then.

00:48:59.879 --> 00:49:04.259
Then you start to get worried about all the methodologies involved and also

00:49:04.259 --> 00:49:09.779
these crazy long-range predictions people can make about decisions up to eight

00:49:09.779 --> 00:49:11.839
seconds before they happen which

00:49:11.839 --> 00:49:15.639
led to all sort of further speculation certainly in the popular press,

00:49:16.259 --> 00:49:21.659
and then you start to apply all sorts of methodological caveats to say well

00:49:21.659 --> 00:49:26.099
Well, actually, the real signal might look very different from the randomness potential.

00:49:26.919 --> 00:49:31.659
The real signal looks much more like a rather constant baseline state that shows

00:49:31.659 --> 00:49:37.539
a very rapid transition to a sort of an upstate shortly before the decision. Yeah. Correct. Yeah.

00:49:37.939 --> 00:49:42.679
But with that, you invalidate your earlier drift-diffusion model. That's now irrelevant.

00:49:44.263 --> 00:49:47.503
Doesn't invalidate it right in fact it it it the

00:49:47.503 --> 00:49:50.683
idea is that it confirms it uh so the

00:49:50.683 --> 00:49:54.003
the uh the you say two

00:49:54.003 --> 00:49:59.023
competing camps or two competing hypotheses are for an early decision which

00:49:59.023 --> 00:50:02.863
is you would might say i mean he's not here for us to ask him but i would say

00:50:02.863 --> 00:50:08.083
as libit's view the decision is early the the the neural what i call the neural

00:50:08.083 --> 00:50:12.543
decision right uh to it to contrast it with the conscious decision.

00:50:14.183 --> 00:50:18.783
This is an early decision, whereas our model says, well, no,

00:50:18.943 --> 00:50:24.243
the decision only happens, and Murakami used the same model,

00:50:24.343 --> 00:50:25.863
or the same kind of model, essentially.

00:50:26.923 --> 00:50:31.063
The decision doesn't happen until the threshold is crossed, and that happens

00:50:31.063 --> 00:50:34.463
very late in the game, very close to the time of the movement.

00:50:34.703 --> 00:50:39.543
And so that if you were, the upshot of that, one prediction that it makes is

00:50:39.543 --> 00:50:44.623
that if you were if you compare movement to no movement uh they should look very similar until,

00:50:45.383 --> 00:50:54.543
the very last moment and until uh that that threshold if you will has been crossed right so.

00:50:56.223 --> 00:51:00.703
We started with a very complex picture on free will and now we end up with something

00:51:00.703 --> 00:51:07.123
relatively pragmatic physiological controllable some sense right we're in the

00:51:07.123 --> 00:51:09.083
decision-making decision-making domain, yeah?

00:51:09.803 --> 00:51:15.043
Is that where you want to be? Or do you want to, again, jump out of that and

00:51:15.043 --> 00:51:20.303
go back to this fundamental question around conscious will, as you call it?

00:51:20.443 --> 00:51:24.383
Because sometimes, now it's out, it's out in the picture, right? Doesn't matter, do you?

00:51:25.223 --> 00:51:29.843
Now we're looking at decision-making about self-generated actions.

00:51:30.263 --> 00:51:34.343
Yeah, I think we should stay firmly grounded in the world of decision-making.

00:51:35.463 --> 00:51:37.123
And self-generated actions.

00:51:38.088 --> 00:51:43.588
And I think there's more that we can do with this model and models like it.

00:51:44.088 --> 00:51:48.608
I think even if ultimately the model is all wrong.

00:51:48.788 --> 00:51:54.828
I think just the idea of taking this kind of phenomenon and bringing it down

00:51:54.828 --> 00:52:00.468
to earth and grounding it in a computational model was an important step.

00:52:00.468 --> 00:52:05.168
And we've already gone beyond with this same model.

00:52:05.228 --> 00:52:10.048
We've done something very simple, which is just to add a second threshold,

00:52:10.148 --> 00:52:15.988
slightly lower than the threshold for activating movement, let's say.

00:52:15.988 --> 00:52:19.028
That we we say well this this lower

00:52:19.028 --> 00:52:22.868
threshold represents sort of self-monitoring process

00:52:22.868 --> 00:52:25.728
so that when when we

00:52:25.728 --> 00:52:28.728
cross that lower threshold some information

00:52:28.728 --> 00:52:31.408
is generated uh to the

00:52:31.408 --> 00:52:34.288
effect that well a movement is very likely to

00:52:34.288 --> 00:52:37.248
happen very soon because i'm very close to the

00:52:37.248 --> 00:52:40.528
threshold um and using that

00:52:40.528 --> 00:52:43.228
uh simple variant uh of the

00:52:43.228 --> 00:52:49.288
model we were able to confirm makes and confirm some predictions about uh the

00:52:49.288 --> 00:52:57.548
conscious urge to move what but a bit called w time uh for the will um so on

00:52:57.548 --> 00:53:02.368
uh on trials where the subject waited a longer time to produce a movement meant,

00:53:02.468 --> 00:53:12.028
the assumption is that the ramping signal was ramping not as steeply as in trials

00:53:12.028 --> 00:53:14.488
when the subject responded early.

00:53:14.808 --> 00:53:18.968
And so the delay between the crossing of those two thresholds would be longer.

00:53:19.208 --> 00:53:24.628
And that would tell us something about the relationship between subjects W time,

00:53:24.748 --> 00:53:29.148
when they felt they had the urge to move, and the movement itself, and that prediction.

00:53:29.468 --> 00:53:31.708
But that's something quite disturbing about.

00:53:32.867 --> 00:53:39.887
Our notion of free will, which what we have in Chobha Chitral now as my moment

00:53:39.887 --> 00:53:41.947
of choice, I decided to make a movement,

00:53:42.067 --> 00:53:47.807
turns out to be just some internal monitoring system reporting that a bit of

00:53:47.807 --> 00:53:52.607
my brain that I'm not aware of has passed the threshold and I'm going to make

00:53:52.607 --> 00:53:54.987
a movement whether I urge it or not.

00:53:55.247 --> 00:54:01.007
That's what it may be. So it's really quite a frightening conclusion for people

00:54:01.007 --> 00:54:03.787
who would like to be even free will because it shows that,

00:54:04.787 --> 00:54:07.967
these feelings that we have about will could just

00:54:07.967 --> 00:54:12.947
be completely wrong because that because you would you would have a feeling

00:54:12.947 --> 00:54:17.027
that i've willed them but it turns out you haven't something in your brain has

00:54:17.027 --> 00:54:20.547
happened and this has popped into consciousness and you've interpreted that

00:54:20.547 --> 00:54:24.607
as well if it's true for that event it could be true for all sorts of other

00:54:24.607 --> 00:54:25.827
decisions you make in your your life,

00:54:25.947 --> 00:54:31.427
that some bit of your brain made something happen. And then it was reported up to consciousness.

00:54:31.947 --> 00:54:37.367
Although I wouldn't say that it's, uh, you, you, you said that maybe that feeling was wrong.

00:54:37.747 --> 00:54:41.987
Uh, but I wouldn't say interpretation of it that we put on it is that we made a choice.

00:54:42.407 --> 00:54:44.867
Ah, I see. So mens rea doesn't hold that.

00:54:46.028 --> 00:54:51.308
In that model no not very well yeah so so that's an important point because

00:54:51.308 --> 00:54:55.128
in the trust of the feeling i was having like you you invited us to some big

00:54:55.128 --> 00:54:58.828
party um with lots of uh booze and,

00:54:59.388 --> 00:55:04.448
god knows what then entertainment but then you open a door and then it's an

00:55:04.448 --> 00:55:09.888
empty room and there's no no no drinks no snacks no nothing yeah decision making

00:55:09.888 --> 00:55:11.668
yeah so haven't you haven't you

00:55:11.688 --> 00:55:16.708
now actually throwing out the dancers with the bath water, um,

00:55:16.988 --> 00:55:19.068
because of the limitations of your methods?

00:55:19.768 --> 00:55:24.128
I mean, the, the goal here was not to prove that free will exists, right.

00:55:24.248 --> 00:55:28.188
But was, was to shed some light on how this all works. No, you shed some light

00:55:28.188 --> 00:55:29.808
on it by turning the light off.

00:55:30.908 --> 00:55:32.168
Well, then you turned the light

00:55:32.168 --> 00:55:35.068
on and the room was empty. The room was empty. I think, I don't know.

00:55:35.688 --> 00:55:40.688
When the risk is, he actually turned on the lights in another room.

00:55:41.668 --> 00:55:45.448
Right. We have shifted. We have shifted now somewhere else.

00:55:46.048 --> 00:55:49.668
No, I think this is where we thought the party was. You put the lights on.

00:55:49.828 --> 00:55:51.608
There is no party. There's no will.

00:55:51.848 --> 00:55:56.368
There's just these things causing stuff to happen, and you heard about it.

00:55:56.868 --> 00:56:02.668
Yeah. So, I mean, what do we do with that? But it was deeply dissatisfactory.

00:56:02.908 --> 00:56:05.428
Well, only if you wanted to believe in conscious will.

00:56:07.148 --> 00:56:11.188
Because you believed that there was a party to go to. That's right.

00:56:11.188 --> 00:56:15.308
I like parties, but the thing is there are things, there are things like,

00:56:15.328 --> 00:56:19.708
like mens rea and there are, there are costs of failure, right?

00:56:19.808 --> 00:56:27.828
So if we really believe with Wagner and his friends that, that conscious will

00:56:27.828 --> 00:56:30.088
is an illusion and, or not operational,

00:56:30.648 --> 00:56:35.148
there's a huge cost associated with that because we're dehumanizing ourselves

00:56:35.148 --> 00:56:40.648
and we're giving up the sense of responsibility for action. The cost of that is huge.

00:56:41.188 --> 00:56:43.548
And if you do that to protect yourself.

00:56:45.065 --> 00:56:50.985
Relatively primitive scientific models and methods, then I think that's a very naive move.

00:56:51.645 --> 00:56:58.205
Because I do believe, given the cost of failure, we ought to critically question the models we use.

00:56:58.325 --> 00:57:00.505
And we already know drift diffusion is great.

00:57:00.805 --> 00:57:06.265
We do really boneheaded tasks as a macaque monkey for months on end.

00:57:06.765 --> 00:57:08.565
Great. We get drift diffusion.

00:57:09.105 --> 00:57:13.105
But as soon as As I make the decision test, it's a little bit more complicated.

00:57:13.545 --> 00:57:16.665
Drift and fusion is not predictive of anything.

00:57:17.025 --> 00:57:20.765
If you give a complex brain a stupid test, you get a stupid code.

00:57:20.965 --> 00:57:24.705
And the stupid code can be deciphered by boneheaded neuroscientists.

00:57:24.845 --> 00:57:29.545
But the thing is, if you give a complex brain a complex task, it uses complex codes.

00:57:29.705 --> 00:57:32.285
And that's where the beauty is. And that's where the scientific challenges are.

00:57:32.605 --> 00:57:38.105
So what I feel now, we're a bit on the slippery slope. Oh, my methods are not really helping me.

00:57:38.425 --> 00:57:43.085
Let's just rephrase the problem. let's scrape away all the complexity so at

00:57:43.085 --> 00:57:47.045
least I can hold on to the method that has given me tenure.

00:57:47.165 --> 00:57:50.225
And it's not for you, okay, but the socialness of the field, right?

00:57:50.585 --> 00:57:55.045
Drift diffusion drives me mad indeed because I think it is distorting our understanding

00:57:55.045 --> 00:57:58.745
of decision making and certainly of voluntary decision making.

00:57:58.785 --> 00:58:01.585
I don't think you can just point to drift diffusion for this.

00:58:01.725 --> 00:58:08.165
So all you need is for that noise to be auto-associated, yeah?

00:58:08.225 --> 00:58:14.985
So that's all you need for this result to happen and the problem for your version is that you want to.

00:58:16.225 --> 00:58:23.965
Have some role for that experience of making the choice and this result shows

00:58:23.965 --> 00:58:28.445
that you can have an experience of making the choice without having made any

00:58:28.445 --> 00:58:34.005
choice in consciousness that matters so so that i mean there may be other situations

00:58:34.005 --> 00:58:35.485
in which you make a choice that matters,

00:58:35.525 --> 00:58:37.985
but in this situation,

00:58:38.365 --> 00:58:42.125
you have the qualia, you have the experience of making a choice,

00:58:42.977 --> 00:58:46.557
But no, we've shown that... I appreciate your intent to mediate here,

00:58:46.617 --> 00:58:48.057
whether to take offense to your summary.

00:58:50.377 --> 00:58:56.037
Because the point I was making earlier, Aaron is showing to us that pink noise

00:58:56.037 --> 00:58:59.937
doesn't do the job, unless it's some really special case of it,

00:59:00.077 --> 00:59:04.417
because he shows a much more S-shaped transition, right?

00:59:04.497 --> 00:59:06.837
Where you don't have sort of gradual accumulation of anything.

00:59:07.437 --> 00:59:10.557
There's accumulation of nothing and a very rapid switch. Well,

00:59:10.577 --> 00:59:12.617
that's when we squeeze the pink noise out of the picture.

00:59:12.977 --> 00:59:18.937
Exactly. Yeah. But that would be the real ground truth of the decision-making signal.

00:59:20.237 --> 00:59:26.477
Abrupt. Yes. Late and abrupt. So it looks very different than this sort of autocorrelation.

00:59:26.477 --> 00:59:29.137
That autocorrelation signals some background thing that happens has nothing

00:59:29.137 --> 00:59:30.317
to do with the decision-making process.

00:59:31.397 --> 00:59:33.317
Right. This is the consequence.

00:59:35.177 --> 00:59:39.877
It's, I mean, it is, it's part and parcel of the decision-making process,

00:59:39.977 --> 00:59:43.497
but it's just that But when we look at it through the lens of this event-locked

00:59:43.497 --> 00:59:46.157
averaging, we see it in that misleading way.

00:59:46.397 --> 00:59:48.597
But it's like a non-specific contributor.

00:59:49.177 --> 00:59:52.717
Yeah. It's not a decisive contributor decision, right? Right.

00:59:52.957 --> 01:00:00.977
So, but I think it would be, as scientists, we must also be willing to say, well, we don't know.

01:00:00.977 --> 01:00:06.817
We don't know, but it's more than decision-making, because for the free will case,

01:00:07.117 --> 01:00:12.737
there is a layer of the ability to reflect upon the decision,

01:00:12.877 --> 01:00:18.577
to own the urges to act, to not only act,

01:00:18.777 --> 01:00:24.257
but to also, and I want to act, I own this urge to act, and I experience that,

01:00:24.277 --> 01:00:28.157
and I can declare it, and I could have done something else.

01:00:28.657 --> 01:00:32.557
But there are many requirements of which the decision-making process is one

01:00:32.557 --> 01:00:36.257
of the requirements, but it's not the whole story. It's only one of the necessary conditions.

01:00:36.757 --> 01:00:40.337
And there I think we should be careful. So yes, studying decision-making can

01:00:40.337 --> 01:00:41.417
be fantastically useful.

01:00:41.977 --> 01:00:46.337
Drift diffusion models can be a great tool to do that, but they're not the whole story.

01:00:46.517 --> 01:00:49.237
So this is why I got a little bit excited, because I feel like,

01:00:49.257 --> 01:00:53.717
oh, we have to be careful not to collapse the complexity of voluntary action

01:00:53.717 --> 01:00:56.657
into the more reduced view of decision-making,

01:00:56.657 --> 01:00:59.857
which is only a necessary condition i mean i think what what we

01:00:59.857 --> 01:01:03.557
could agree is that in this task it's a very minimal task for

01:01:03.557 --> 01:01:09.217
any kind of volitional control and uh i think what aaron's done is produced

01:01:09.217 --> 01:01:13.957
the sort of uh the minimal model of that and occam's razor said well why should

01:01:13.957 --> 01:01:18.957
there be anything more if this random process that's also correlated is enough

01:01:18.957 --> 01:01:20.237
to explain the result then.

01:01:21.291 --> 01:01:24.571
Quite possibly all there is the disturbing thing for you

01:01:24.571 --> 01:01:27.511
is that the quality are associated with it

01:01:27.511 --> 01:01:32.091
are like the quality are associated with other decisions so i think the pressure

01:01:32.091 --> 01:01:36.511
is now on people who want to defend you know sort of some strong role for consciousness

01:01:36.511 --> 01:01:42.611
and decision making to find another paradigm where where you can demonstrate

01:01:42.611 --> 01:01:46.131
that because you mean where volition then comes

01:01:46.251 --> 01:01:48.431
in as a separate factor, or what?

01:01:48.651 --> 01:01:53.891
Yes. Yeah, well, I mean, or in this paradigm, to show that it's more than just

01:01:53.891 --> 01:01:58.251
this minimal model, this, I mean, that there might be something else that's

01:01:58.251 --> 01:02:00.771
happening in those last few hundred milliseconds.

01:02:01.251 --> 01:02:03.831
Right. Would you agree with the summary there? Yeah, and I think,

01:02:03.871 --> 01:02:09.131
I mean, I think one of the, again, one of the things that, that is to be recommended

01:02:09.131 --> 01:02:13.851
about the, about this model is its simplicity. It's a very simple model,

01:02:13.911 --> 01:02:15.351
and it accounts for the data.

01:02:15.591 --> 01:02:23.151
And that puts, I think that reverses the burden of proof, in a sense, now to the other side.

01:02:23.271 --> 01:02:32.051
Say well, why would millions of years of evolution result in a human who lives

01:02:32.051 --> 01:02:37.091
one full second behind their own self decisions?

01:02:37.091 --> 01:02:39.911
Oh, I have an answer to that, but that's not an interview. But the thing is,

01:02:39.951 --> 01:02:46.251
we already know, as you know, Libet, he understood the consequence of his observation

01:02:46.251 --> 01:02:48.451
or the implications of his interpretation.

01:02:48.811 --> 01:02:53.731
So he said, well, maybe the free will acts in holding back the action,

01:02:53.871 --> 01:02:56.691
in interrupting the action, stopping the action. Right, right.

01:02:57.191 --> 01:03:01.771
And actually, it turns out, if you start to do tasks where you have stop signals,

01:03:01.911 --> 01:03:05.791
so like countermanding, then drift-of-fusion models stop working.

01:03:06.131 --> 01:03:08.431
They're not predictive of anything. Okay.

01:03:09.111 --> 01:03:12.191
So you really have your counter-example.

01:03:12.911 --> 01:03:16.811
As soon as you start to voluntarily withhold action because you get the stop

01:03:16.811 --> 01:03:22.991
signal, then just integration to threshold is not explanatory in any way of

01:03:22.991 --> 01:03:24.851
the performance of a macaque monkey.

01:03:25.900 --> 01:03:34.200
But the other way to recover your notion of human and of choice is to embrace

01:03:34.200 --> 01:03:40.440
the unconscious as part of the self and a source of your decision-making. Oh, absolutely.

01:03:40.760 --> 01:03:45.160
And if you do that, then none of this is a worry because you just say,

01:03:45.200 --> 01:03:48.560
well, I am not just my conscious process.

01:03:48.640 --> 01:03:51.920
I'm the whole of what my brain is doing in the context of the body.

01:03:52.360 --> 01:03:57.320
And that collective system is making sensible decisions. And some of them are

01:03:57.320 --> 01:03:58.340
being reported to consciousness.

01:03:58.840 --> 01:04:02.680
It shouldn't bother me if most of them aren't as long as they're good decisions.

01:04:02.840 --> 01:04:10.540
As long as your conscious urge is a meaningful best guess on the part of your

01:04:10.540 --> 01:04:13.400
brain as to the decision that you made and the time that you made it.

01:04:13.540 --> 01:04:17.240
And if consciousness is able to monitor that and say, well, hang on a minute,

01:04:17.280 --> 01:04:18.400
we made a poor choice there.

01:04:18.820 --> 01:04:23.260
Maybe we can, maybe there's a role for consciousness If you guys could finally read my papers.

01:04:24.340 --> 01:04:27.300
I know. I'm telling you what was in your papers. Ah, thank you.

01:04:27.540 --> 01:04:32.980
You know that I see consciousness and also volition act towards the future.

01:04:33.580 --> 01:04:37.940
And all real-time control, as the ones you manipulate as experiments, is all subconscious.

01:04:39.060 --> 01:04:42.260
And consciousness is catching up because consciousness tries to reconfigure

01:04:42.260 --> 01:04:44.760
you for the future. So you will yourself into the future.

01:04:45.240 --> 01:04:47.940
And that's the big mistake people make. They think real-time control,

01:04:48.100 --> 01:04:50.160
because they never program robots to control anything.

01:04:50.620 --> 01:04:53.600
They think real-time control is conscious.

01:04:53.980 --> 01:04:58.580
It cannot be. It has to be automated quick, because that's your survival system.

01:04:58.820 --> 01:05:01.780
It has to act in real-time, and anything can happen to you.

01:05:01.960 --> 01:05:05.800
But what you then do, you reassess, and that's why the monitoring also comes in.

01:05:05.880 --> 01:05:09.160
You have to reassess, you revalue, you extract norms from your environment,

01:05:09.300 --> 01:05:11.960
and you reconfigure your real-time control, so in the future,

01:05:12.080 --> 01:05:13.940
you can will yourself to be better.

01:05:14.480 --> 01:05:19.520
So in that sense, indeed, this is I'm not scared by that we're converging on

01:05:19.520 --> 01:05:24.720
a common view that would be really scary so Aaron, so now that we all agree

01:05:24.720 --> 01:05:29.300
and are humming along and you promised to read all my papers.

01:05:30.400 --> 01:05:35.600
So you have been wading through this really difficult territory this minefield

01:05:35.600 --> 01:05:40.460
of volition and you're still standing which is amazing, that's a real accomplishment

01:05:40.460 --> 01:05:42.220
so you're a great scientist,

01:05:43.158 --> 01:05:47.538
So given that experience and these accomplishments, what is Aaron's law that

01:05:47.538 --> 01:05:49.178
we should follow to understand the brain?

01:05:50.058 --> 01:05:55.778
Well, I think what we should start looking at and what hasn't been the case

01:05:55.778 --> 01:06:05.638
in the past is look at instances where conscious states might actually play

01:06:05.638 --> 01:06:07.818
a role in an upcoming decision.

01:06:07.818 --> 01:06:11.398
Decision so there's i i i think

01:06:11.398 --> 01:06:14.278
that in literature there's been a maybe because it's

01:06:14.278 --> 01:06:17.378
easier to do maybe because it's sexier but there's been a bias

01:06:17.378 --> 01:06:24.258
toward finding examples of things that you can do without consciousness um and

01:06:24.258 --> 01:06:29.018
that that leaves open the possibility that there are some things that you do

01:06:29.018 --> 01:06:34.258
need consciousness for uh but we haven't been spending our time uh investigating

01:06:34.258 --> 01:06:36.758
So how does that allude to the law?

01:06:37.578 --> 01:06:41.858
Ah, I don't know about a law. Look, that's a little mug or something.

01:06:43.478 --> 01:06:49.178
It's like 10 words. So it's Aaron's law. Aaron's law is some actions,

01:06:49.378 --> 01:06:52.938
some volitional actions might actually be conscious,

01:06:53.338 --> 01:07:01.118
might actually have a conscious antecedent, let's say.

01:07:01.118 --> 01:07:04.598
Um so we've we've we've

01:07:04.598 --> 01:07:08.878
looked at a lot of examples of of tasks where uh

01:07:08.878 --> 01:07:12.038
you you consciousness seems

01:07:12.038 --> 01:07:15.238
not to be involved so aaron's law is some consciousness

01:07:15.238 --> 01:07:20.178
might play a role sometimes yeah sometimes maybe rarely but maybe sometimes

01:07:20.178 --> 01:07:26.658
right all right uh the other thing is so we were in in paris recently for a

01:07:26.658 --> 01:07:31.218
new machines conference which was fantastic and Tony likes Paris now,

01:07:31.338 --> 01:07:32.538
so he wants to go back to Paris,

01:07:33.338 --> 01:07:34.818
and he will come visit you four years from now.

01:07:35.038 --> 01:07:39.638
Okay. And he will come and check, he will come to the notebook and check whether

01:07:39.638 --> 01:07:46.978
you have actually falsified or confirmed the key hypothesis for your scientific program.

01:07:47.678 --> 01:07:52.698
So what's the key hypothesis you would like to see tested in that time frame?

01:07:54.038 --> 01:07:57.438
Four years. Four years. Prediction, yeah.

01:08:02.058 --> 01:08:05.218
Or maybe by then the channel tower will be filled up with concrete.

01:08:05.418 --> 01:08:09.318
Yeah. Two years. Maybe. Two years, yeah. Okay. Earlier.

01:08:12.875 --> 01:08:18.055
I think what I'd like to see done is a test of whether or not,

01:08:18.095 --> 01:08:25.295
to the extent that this can be done, whether or not one can draw a causal arrow

01:08:25.295 --> 01:08:27.195
between conscious states and actions.

01:08:28.735 --> 01:08:35.435
And if we could get a handle on how to come up with a paradigm to test that,

01:08:35.575 --> 01:08:39.475
I think that would add sort of counterweight to the kind of research that's

01:08:39.475 --> 01:08:41.595
been done over the past decades.

01:08:41.595 --> 01:08:44.075
What's the specific prediction then from that?

01:08:44.475 --> 01:08:48.515
Ah, I guess it's true. I didn't give you a prediction. That's all right.

01:08:48.715 --> 01:08:50.575
Yeah. So I've paid enough attention.

01:08:50.835 --> 01:08:55.835
Okay, so, yeah. So, okay, a big prediction is this.

01:08:58.075 --> 01:09:04.395
You can, what the model, one of the things that the model says is that the,

01:09:04.395 --> 01:09:10.555
let's say the character of the noise will in part determine the shape of the readiness potential.

01:09:11.775 --> 01:09:21.235
So if you, let's start with describing the power spectrum of pink noise has

01:09:21.235 --> 01:09:23.575
this one over F slope to it.

01:09:24.515 --> 01:09:30.455
So that the slope of the power spectrum gives you what we what you could call

01:09:30.455 --> 01:09:34.235
the character of the noise or the degree of autocorrelation in the noise.

01:09:36.075 --> 01:09:43.775
So one prediction that follows from that highly counter counterintuitive prediction,

01:09:43.875 --> 01:09:46.775
or at least a novel prediction,

01:09:46.935 --> 01:09:52.135
is that the character of the noise, the 1 over f exponent of the noise,

01:09:52.255 --> 01:09:56.155
should predict at an individual subject level the shape of the readiness potential.

01:09:58.355 --> 01:10:00.315
Isn't that the retrodiction? Don't

01:10:00.315 --> 01:10:04.795
you really know that, given the results you have? No. No, not at all.

01:10:06.215 --> 01:10:10.335
So, no, that's very much a prediction and a strong one.

01:10:11.635 --> 01:10:16.715
It's difficult to test, but it is testable. All right. Tony, you wrote it down?

01:10:18.795 --> 01:10:21.435
It's on record on tape. Yeah. Aaron

01:10:21.435 --> 01:10:24.055
Sugar, thank you very much for this conversation. Thank you for having me.

01:10:27.055 --> 01:10:32.875
The CSN podcast was produced by the Convergent Science Network of Biometrics

01:10:32.875 --> 01:10:39.255
and Biohybrid Systems, a project funded by the European 7th Research Framework Program.

01:10:40.855 --> 01:10:46.155
For more interviews, recorded lectures, or upcoming conferences in the field

01:10:46.155 --> 01:10:52.395
of biometrics and biohybrid systems, go to csnnetwork.eu.

01:10:52.720 --> 01:11:00.720
Music.