Numenta Interview With a Neuroscientist, 25 Jan 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Konrad Kording Part 1 - Interview With a Neuroscientist

Description

Konrad Kording is a professor at University of Pennsylvania, known for his contributions to the fields of motor control, neural data methods, and computational neuroscience. He runs the Kording Lab, or K-Lab, which focused on computational neuroscience early on and now focuses on causality in data science applications. His lab has made an impact across many fields overtime, including Bayesian brains, causal effects in human behavior, and uncertainty in the brain.

A

Howdy folks and welcome back to interview with the neuroscientist now this series has moved off of YouTube, so I just wanted to. Let you guys know who are watching this on YouTube that this is now a part of our podcast. We have a new podcast called Numenta on intelligence, and so, if you want to subscribe to it, search for new mental on intelligence, I have interviewed two neuroscientists so far on that podcast Alex, Vaughn and Blake Richards.

A

So if you want to hear those you'll have to go subscribe to that, podcast and I will continue to put interviews there. However, for this one I decided to try and record while I was skyping with dr. Conrad according and according turned out pretty well, so I thought why not put it out on YouTube at the same time as we release the podcast, so I'm gonna put it out in both places, so I hope you enjoy this interview with the neuroscientist Conrad Courtney, okay, yeah I. Think it's because I turned the recording on.

B

A

Talking with you so.

B

What exactly is the objective?

B

What's driving the.

A

Whole thing, so my drive is, is a couple of things: first, so I'm a community manager for Numenta and I'm trying to foster this community of. Like hobbyists scientists engineer it's a very diverse community of people. You know when you just open something up like this.

A

You get a lot of interesting perspectives, but my main focus is education and I really am trying to focus young people like with this series of interviews in general, I'd like to to try and engage people to be interested in how intelligence works and how the brain is implementing it, and things like that, because that's what got me into it right, so I'm, sort of focusing on the new generation of AI experts as they're growing and I really want them to learn how intelligence works, not just the Bayesian models. Necessarily, you know well.

B

So that's costs. Yes,.

A

And I'm really I'm, focusing on how my company is trying to understand the brain and the discoveries that we think we're making the theory that we the way we think it were. You know because it's theory, it's a it's theory, and we want to we're so we're not doing experiments, but we're always looking at the experimental data to make sure that the theory still works.

A

You know in the context of these new new experiments and stuff coming up and the reason I came to you is because you're you're sort of the type of neuroscientist now that the way you think about the brain really aligns well with the way we think about the brain. This.

B

Is great to hear I mean like a as you know, I come from a Bayesian background right now, I might be a little more connectionist than Bayesian reason.

A

B

You can't get the Bayesian out of thinking like me, I see everything through the Bayesian lens, even if I'm not doing Bayesian things anymore. That's.

A

A totally different, that's great, because that's a perspective that I can't present to my audience, because I didn't come from that perspective by trade I'm. A software engineer, I, didn't learn any of the maths involved. I just learned how to build things with something right.

B

A

Came out of this from an engineering standpoint, how do I build intelligence? That was my interest like what, if something is intelligent and we can figure out how to reverse engineer it and build it. So that's really. The direction we're coming from and I know and we're meeting in the middle I mean eventually we're all gonna meet in the middle and build something amazing. Like that's the vision, you're a third-rate. You.

B

Know you should have started the interview it starts to be found like people might find.

A

We may as well turn the sand. I mean it's.

B

A

Cool, it's so great to meet you by the way and I haven't like introduced you to the audience or anything. But since we're already recording the interviews probably started everyone. This is dr. Conrad, cording, I, don't have a good intro. Let me let you introduce yourself, I, think that I find that's better for people anyway, so I always mess it up. Well,.

B

I am a chef I'm good at introducing myself, so yeah I mean you're, a scientist I'm a famed experimentalist I started my PhD recording from primary basic optics of the cat, didn't walk all that. Well, so I ended thinking ma, but how we could use math and computation to understand the brain and I looked at the brain from many different perspectives over time.

B

I was basically for a long time and think about neural computing, my entire life in a way I'm interested in deep longing and how the brain might be like deep, longing and I might be there front and Beyond neuroscience I'm interested in what we can meaningfully say about the world using data that I am I'm bad at introducing myself. In that way, it's a it's I'm, just interested in whatever is interesting.

B

A

A good way to put it I mean that's this I'm the same way I just haven't taken like the PhD track, and a lot of people in our community are the same, we're very passionate about trying to understand how things work, but, but you know, there's there's, there's no major I could ask this a lot in our community.

A

What should I study you know if I want to understand like how intelligence works and, and this next sort of era of AI, what should I study, computational, neuroscience or straight-up neuroscience, or go straight with math and physics, and it's hard to say really, but.

B

They think the answer is yes, you should probably study all of them like at some level. If you don't understand psychology, it's hard for you to get what intelligence is about. If you don't study, math, it's very hard to make what you want to say concrete. If you don't study computation, you can't actually implement it in lots of problems. You only see it once you try to build it and and similarly methods from physics. They just help. You think about the brain in a broader way. I think it.

A

Took me a very long time to finally come to like a somewhat of a basic understanding of intelligence, and even that's very broad and abstract. You know please.

B

Please share it: I'm, sometimes very unclear, but mock's. Oh gosh,.

A

Don't put me on the spot, you're.

B

A

You so let me let me flip it on you, because you said you used to be a Bayesian. It's sort of, like maybe you've changed your view of it. One of the things are the the things I found. It was a theme in your writing or in in their papers.

A

That you produced is that if you structure your experiment in the right way, there's I mean that's amazing, how much data we can get now from experiments, and it's just getting more and more and and because it's so messy, sometimes you can structure your experiment that you will find what you're looking for sometimes, even though it may not be what you're looking for right. Maybe you could talk a bit about that yeah.

B

I mean like oftentimes in behavioral experiments, but you look as you're. Looking for some kind of an effect say is uncertainty irrelevant when you make decisions so I give you a task where uncertainty is really useful. Like you, knowing how uncertain you is really essential. So there's one thing you should do if your very answer: another thing that you should do, if you very sad about the situation and everything else will be the same.

B

So if you want only uncertainty matters and then yes, you used uncertainty and we've shown that in many experiments, but if I give you a situation where the reward matters and nothing else, you'll look further reward. If I give you a situation where, at some level, arousal matters or something like how much you're like engaged in a task. Well, then that will matter if I give you a task, we say color matters.

B

Well, then color matter, so at some level uncertainty that was like in the early 2000s like MIT to thoroughly click 2005, maybe up to 2010. There were all over some lots of labs. That said well, let's see if on some discipline, and they all found that uncertainty is important yeah. But if we had instead getting all excited about color, then we might have had brain theories that I centered around the idea of color.

B

So in the end like when these feels like really start going, the question is really what what we didn't know in India like we.

A

Now know that answer remembers.

B

Sure, if I ask you, how certain are you that, like you, will get up between 6:00 a.m. and 9:00 a.m. tomorrow, you'll get me a pretty precise answer to that. If I ask you, how submit is that I'll get up in an interval which you'll have Mon softly but sure you will be less precise and once I give you some information, you'll be better at that.

B

So so my view is very much based on like uncertainties like that centerpiece of the way intelligence works, and you can tell a story where everything you do is about uncertainty. Now you can tell another story, which is everything is about lying, there's, a story that you can tell just as well, where you could say. Well, if you make a mistake, if something is wrong, you'll change and next time, you'll be better. No, then you can equally in a way and explain a lot of intelligence. Nothing.

B

And certainty, Gazoo and suddenly says, along with something that I know a kind of star house of mine, which means that, if I'm very uncertain about something I will rely less on the thing, then, if I'm very certain about something now, a longing system will do the same thing, because it figures out that in the cases where it was very sudden, but not the things, it did worse than the cases where it was very sudden and used the things and in the uncertain fame cases, it was very uncertain about things in very much relied on it.

B

There was a mistake, and so therefore, where we are that continuum I'm not sure not like you could be, that the brain we're born is people who had their to deal with uncertainty, because uncertainty is so important, mm-hmm, in which case, if you wouldn't like and something is dug in too late, the plan with which our brain is made. Alternatively, we could be really good at longing and we would never know the difference. Would.

A

You say that uncertainty is sort of one dimension of an aspect of our brain. It applies to a lots of other things that are that are happening there. Yeah.

B

I think it is an aspect of reasoning. It's an important aspect of reasoning, not like things that your mom Sutton everybody less important in a way, and if you have a bunch of things that you're somewhat uncertain about each of them, you can typically combine them to become awesome. No right right.

A

Lay your friends, that's problem. Solving! It's.

B

Prom, something exactly in dealing with uncertainty is an unavoidable aspect of intelligence of problem. Solving right.

A

So let me say: let me eyes this a lot in my head, it's hard right when I think about ideas and objects or things you know discrete things that we think about. You can find uncertainty to it. I like to think about those things is like multi-dimensional attractors like there's some match in your brain. There's some certain neurons that fire, when you think about a specific thing and the uncertainty about that is sort of like how messy and noisy. Is that a try?

A

How well-defined is that thing in your neurons and the connections between them, but.

B

That is one way how the brain could deal with uncertainty. It's not the only way how you could deal with uncertainty. Let's say it's possible, it kind of. Let's say we have some estimation. Is that thing behind the guitar that I see? Now, let's say if my video was somewhat blocked, it could be that it's a key type could also just be that it's like a painting or something behind you yeah. No, so there's two very different ways how the brain could represent such a thing. It could either be if I'm uncertain.

B

If they say the image quality is low. I kind of like my brain activity is very messy, but, alternatively, my brain activity could be absolutely not messy at all where I could say. This could be a guitar exactly this guitar with prime 18.9 or this.

B

It's not messy at all, I could say: guitar, not guitar, and so at some level, this messiness of neural code, as as something that communicates to other parts of the brain that we uncertainty about uncertain about. It is only one of many codes. You could, for example, say that there's just a cell that says how long something is my visual system at the moment, in which case everything is and suddenly has no effect whatsoever, no messiness involved ever, but we have like one cell that basically says very.

A

Uncertain not very.

B

A

Interesting, so that so, if you get an ambiguous sensory input and you're trying to do object identification, you could once you match it with the best thing you could just snap. That's it you've made the decision right, you've, sort of applied your vision of the guitar. If you've decided that's a guitar, you make the decision. I'm gonna apply my idea of what a guitar is to that object in space right and it's no longer really you're certain about I mean your certainty at least has gone way up, because you've made that decision right.

B

And let me let me show you key areas in which case these two ways of thinking about uncertainty feel very different. So let's say one case where you see a guitar and it's very dark, so this small number photo. So you can't be sure if it's the guitar- and in that case you can say maybe the neural activity is very messy because kind of stuff comes in and we're unsure how to interpret it. In this little small number of photos, yeah.

A

B

That case, like this disorderly, seems like a very natural way of thinking but uncertainty. Okay, let me take you to another case where it seems like really weird. So, if I study movement a lot yeah so in movement when when they don't show you your hands, you don't see your hand for a while it turns us did you become uncertain of where you and this because your proprioception has this thing. That's called drift. So basically, if I'm like rotate your hand here and hold it long enough, you will kind of no longer.

A

Know exactly where I know exactly what this feels like, because I meditate and sometimes I'll, wake up a feeling, like my arms, are in a different place, knowing that they aren't really there.

B

A

They're somewhere different there's.

B

Also, the do you know the rubber hand illusion no.

A

It's that okay.

B

Rubber hand illusion is in me to take rubber. That's not yours, I put it next to your arm, then they hide your akshar okay. So there's like now a divider.

A

Think about it like this, a divider, you see the rubber hand.

B

You don't see your real hand and then what they do is they stimulate the rubber hand and your real hand with like just some mechanical device at the same time, right and after they've done it often enough. It feels as if this rubber hand is like totally your hands. Yeah.

A

That makes sense to me because you're now, you're sort of like projecting your experience that you've observed several times predicting that you're gonna feel this, and since you see it happening, even though you don't feel it happening, you're like well, that's good enough. That's good enough like, but.

B

Do you feel it happen, that's how they trick your brain into by stimulating it. At the same time, whenever.

A

B

That the rubber hand gets touched, your real hand gets touched and.

A

By doing that, awesome enough, they.

B

Convince your brain that the rubber hand is totally your arm and then, if someone gets up a knife and steps into the rubber hand, you will be scared to death right okay.

B

But what did what it shows is that that you have considerable insomnia, but where your hand us now, when you don't see it now, here's the interesting thing. If you want to know how much uncertainty and you have about where your hand, this it's not something fuzzy about the visual input or something it is about. You memorizing how long it's been since you last saw your hands: oh right, right! Okay! So in that case, how unsub in us, something is not something that comes from the visual stimulus is exactly the same adjustors.

B

It is something that comes from so.

A

Telling something little over time right this.

B

Right: it's something that you can choose to learn about something you need to integrate seriously.

A

Everyone case, based on what you're doing at any moment, in the context of your actions. That's.

B

Right so in one case, I'm suddenly something that's like in the image. In the other case, it's something that you lump over time. It is not in the image at all, but there's something that's instantaneous right. So in that second view, the idea that uncertainty is sort of something that's in the in the answer in the in the fuzziness of the neural representation. Kind of doesn't make much sense because, like limit system is, is.

A

Exactly the same, where.

B

Is that uncertainty at that part coming from some.

A

Other disconnection up here, yeah.

B

Now this is, this is weird yeah.

A

Yeah, it's uh I mean you can convince you it. This is sort of the problem with belief. You know you can convince yourself over time if you're given enough evidence, that's something completely wrong is true. You know, and you believe that until you get enough counter evidence that you can change your beliefs and if someone convinces you it experimentally, that that's that's your hand, that's your hand, that's your hand. I can see that I mean they're tricking. You, your Chi, they're, changing your belief structure in your brain for that small period of time.

A

B

You can, of course make it may see an argument for that know like if the rubber hand actually is your hand, it is very unsurprising that, whenever the rubber hand gets stimulated, you feel it all right, if the rubber hand, in your hand, would be different, how how improbable is it that they always get stimulated at the same time? So, in a way from a statistical perspective, your brain does the right thing it right.

B

It's very likely that it, the rubber arm acts, is yes, if you don't know that the experimentalist actually designs the stimulation to make trick you into believing it's the same right.

A

Yeah, that's. That is interesting, so we were we started off talking with about. You know, sometimes you're not able to find what you're looking for. Sometimes it's just looking for something you find a correlation that might not be the correct correlation because you're not looking at it from a more general aspect, so you I think you've written a bit about how to do these type of generalization studies. Is that a way to counter this in experiments anyway? Yes,.

B

In experiments I think in your sense, we really need to stop doing generalization studies much more. So what we typically do is we do one experiment and then, with one theory that goes with that experiment right, a generalization study, a do. One experiment figure out what the theory is for that, and then they do a very different expand and they see if my series still works. This is something that we almost never do in neuroscience and rarely in psychology these resulting theories from it. They are not put to the test.

B

Now, if you, for example, say is a brain area we recorded from it, we find that neurons have tuned in to orientation. Therefore, this must be an area that does orientation discrimination.

A

B

These know stuff in a lot of it's the same logic of what of a lot of what came afterwards. That statement without a generalization study means almost nothing for all that we care, but every feature of the world their wish it could change, could change the activity of those neurons. There was a recent carandini study that basically showed that locomotor activities is all over the visual cortex, so this is just to kind of without a generalization study, it's very hard to know what we have long in theory space. So.

A

How should neuroscientists go about that? Well, one of the things you do is you collect data from a lot of different places right you create your own studies, because everyone's got so much data, so you think that there's there's an area, that's right for other labs to try and do sort of these cross generalization studies using existing data yeah.

B

I think times times a pretty good fast data parasites. No five.

A

Years ago I'll tell you I'm.

B

Like a very much believed in in their being like an ecosystem for ideas, yeah.

A

B

Is the probability that an experimental lab can look at their data and figure out kind of statements that make sense about intelligence, not like, let's Lou, between like data and ideas? That is something that we have in the past, often delegated to purely experiment. Labs and intelligence is pretty complicated, at least as far as I'm concerned.

B

For the moment, I think we need like an ecosystem where people we can come up with ideas, find ways of formalizing and find ways of testing it, and that used to be very difficult, and it's not just on the experimental site. The same thing once there is site that lots of theorists are loathe to actually get their hands dirty and like show that their ideas are born out in data I I think we need those dredges well.

A

I'm totally on board with that cause, I think that there's so many good ideas out there, and so many people approaching this problem from different places. Making those bridges is super important so that we can all you know, build something and progress. This area together in the future I almost forgot to mention this is part one of a two-part series but I'm sure there's a link somewhere. You can get to the next one.