Listeners:

Top listeners:

00:00 00:00

Go to album

Paul Verschure on consciousness and distributed adaptive control CSN Podcasts
Edvard Moser on grid cells and entorhinal cortex CSN Podcasts
Giacomo Rizzolatti on mirror neurons and action understanding CSN Podcasts
Robert Axelrod on game theory and prisoner's dilemma CSN Podcasts
Adrian Owen on disorders of consciousness and vegetative state CSN Podcasts
Jonathan Whitlock on markerless motion capture and posterior parietal cortex CSN Podcasts
Luis Puelles on neuroanatomy and prosomeric model CSN Podcasts
Zoltan Molnar on subplate neurons and cortical development CSN Podcasts

Paul Verschure & Tony Prescott on synthetic psychology and robot models

Episode 9 15.03.2018

PLAY EPISODE

Season 2018

0 Followers 10 Episodes

Description

What would it take to build a true science of the mind , one that combines brain theory, robotics, and behavior into a unified framework? Paul Verschure and Tony Prescott reflect on a decade of interdisciplinary research at the intersection of neuroscience, psychology, and engineering, asking whether synthetic models can finally deliver the explanatory theories that biology alone has failed to produce. Subscribe for more from the Convergent Science Network podcast series. In this special episode, Verschure and Prescott turn the microphone on each other to discuss the intellectual foundations behind the BCBT summer school and the Living Machines conference. Starting from the famous Rosenbluth and Wiener argument that understanding complex biological systems requires building simplified physical models, they examine why robots offer something animal models cannot: complete access to every parameter, behavioral realism, and the ability to test sufficiency of a theory in real time. The conversation traces a lineage from cybernetics through Breitenberg’s synthetic psychology to their own Distributed Adaptive Control framework. Central to the discussion is the tension between top-down behavioral modeling and bottom-up neural circuit analysis. Verschure describes how abstract behavioral models and detailed hippocampal simulations have converged to unlock new features like vicarious trial and error and mental time travel in robotic systems. Prescott pushes back on the limits of sufficiency arguments, advocating for completeness and convergent validation across multiple levels of description. Both agree that neuroscience suffers from an excess of technology-driven data and a deficit of genuinely explanatory theory , a gap that synthetic psychology is uniquely positioned to fill. The episode also features a candid exchange with Christine Aicardi on responsible research and innovation within large-scale projects like the Human Brain Project, exploring the limits of collective reflection as an ethical framework and the structural challenges of implementing responsible governance in science. Part of the Convergent Science Network podcast series from the BCBT Summer School.

Tagged as:
Framework robot models synthetic psychology Theory

About the author

CSN Podcasts

Both the triumphs of humanity and its most evil deeds have resulted from collaboration. In a time where humanity is required to aspire to the former and minimize the latter, the question arises of how collaboration arises and why it fails. Surprisingly, this phenomenon, so central to who we are, is not well understood. Hence, a collaborative effort is required to understand collaboration in its full biological, psychological, sociological, cultural, and economic complexity and to translate this understanding into operational impact. This series of podcasts is one step toward achieving these complementary goals. The Collaboration Podcast presents interviews with people who are central orchestrators of collaboration in various domains including business, government, science, art, health, sustainability, and the military. The discussions were conducted by Prof. Dr. Paul F.M.J. Verschure and members of the Program Advisory Committee of the Ernst Strungmann Forum on Collaboration (https://www.esforum.de/forums/ESF32_Collaboration.html) during 2021 and had the goal to sketch a map of opportunities, challenges, and obstacles in human collaboration. The forum took place in May 2022, and now we would like to share this series of interviews with a broader audience. The full report of the Forum will be published in 2023 by MIT Press. The podcast was produced by the Convergent Science Network (https://www.convergentsciencenetwork.org/). Context: The stability of social systems depends critically on realizing sustainable methods of “collaboration,” yet how and by which means collaboration is achieved is not clearly understood; neither are the conditions or processes that lead to its breakdown or failure. Collaboration can be understood as cooperation between agents toward mutually constructed goals. Part of the reason for our lack of understanding is that the phenomenon of collaboration is, by nature, a highly multidisciplinary problem, and effective research into its complexities has been difficult to achieve across the broad range of scientific and technical disciplines involved. The need for a fundamental understanding of collaboration, however, has become increasingly important. Not only does humankind demand answers as it attempts to address critical challenges at multiple scales (e.g., climate change, migration, enhanced automation, social and economic inequality), but ever-increasing technological and economic means of interconnecting people and societies are disrupting long-established, familiar patterns of how we interact. Radical technological changes that are ongoing have the potential to reshape collaboration in ways that are currently hard to predict or influence (e.g., by altering configurations in interaction, information creation, and modes of communication). On one hand, such changes could disrupt hitherto stable forms of collaboration by affecting critical communication channels and traditional roles, as can be observed in the rapidly changing patterns in governance, commerce, and social interaction. Conversely, technology could lead to the emergence of novel, successful forms of collaboration that deviate from traditional “hierarchical” architectures. Evidence of this can be seen in areas as diverse as highly automated manufacturing plants, the open science movement, collaborative software repositories, user-centered services, and the sharing of economy-based modes of organization. Without a fundamental understanding of the mechanisms, processes, and boundary conditions of collaboration, it is not possible to evaluate or predict which of these possible scenarios are sustainable or even plausible. The Forum “How Collaboration Arises and Why it Fails” (May 8–13, 2022, Location: Frankfurt am Main, Germany) Chairs: Andreas Roepstorff and Paul Verschure Program Advisory Committee: Jenna Bednar, Julia R. Lupp, Bhavani R. Rao , Andreas Roepstorff, Ferdinand von Siemens, and Paul Verschure

Timestamp

00:00:03 - This is the Convergent Science Network podcast. Leading researchers in the domain
00:00:10 - of neuroscience, brain theory and technology are interviewed by Paul Verschoor and Tony Prescott.
00:00:21 - This is Paul Verschoor, here together with Tony Prescott. but this is the last
00:00:28 - podcast of the 10th edition of BCBT.
00:00:35 - And essentially it's up to Tony and me now to discuss our talks.
00:00:40 - But also I think to reflect a little bit of what we tried to achieve,
00:00:44 - all the preceding BCBT's and the current BCBT.
00:00:48 - Because in some sense we also started this whole thing, the summer school and
00:00:53 - the Living Machine Conference, that's now running for six years,
00:00:59 - with the idea to really build or carve out, if you want, a multidisciplinary domain,
00:01:05 - that combines the study of mind and brain with the appropriate technologies
00:01:10 - to really build, if you want, artificial mind and brain as a model.
00:01:16 - And this is definitely reminiscent of, let's say, the cybernetics revolution
00:01:20 - of the 40s and 50s. Right.
00:01:22 - And I'm definitely greatly inspired by that period.
00:01:27 - And also, Tony, you started your talk with this famous Rosenbluth and Wiener
00:01:31 - quote, like the best model of a cat is the cat and preferably the same cat.
00:01:37 - And you use it more often in your talk, so you find it apparently an important point of departure.
00:01:44 - So why do you find that a useful observation? Well, I think it's interesting
00:01:50 - because so many people misunderstand that quote, because it's meant ironically.
00:01:59 - And people use it as if it, occasionally people use it as if they think it's
00:02:05 - just meant to take that literally.
00:02:06 - And always study the cat if you
00:02:09 - want to understand the cat whereas what uh rosenbluth and
00:02:13 - vena were saying is look uh
00:02:16 - you can't understand the cat uh
00:02:19 - in its full detail um because you will end up with a description that's as complicated
00:02:26 - as the cat indeed you won't ever get there because it's difficult very difficult
00:02:30 - to measure and the whole point of of what they were saying is that actually
00:02:35 - if we want to understand any animal,
00:02:38 - such as a cat or a human, then we have to use models.
00:02:42 - And in biology, everyone accepts this.
00:02:46 - But what they use is one animal as a model of another.
00:02:50 - So typically in neuroscience now, we use mice as a model of the human.
00:02:56 - And that involves a whole lot of assumptions, which we don't often discuss,
00:03:01 - or they're not often pushed to the fore, perhaps the way they should be um but
00:03:06 - even more than that we we use um.
00:03:09 - A a mouse that might be uh anesthetized or
00:03:14 - uh you know it might be uh that we
00:03:17 - are uh holding it down pinning it down in
00:03:20 - some sort of head restraint and making it do some behavior and in
00:03:23 - each of these situations we are stepping further and further
00:03:26 - away from the thing we want to understand which might be the
00:03:29 - freely moving freely behaving animal or person so
00:03:33 - um i like that norbert vena a quote
00:03:36 - because the rest of that article it's a lovely
00:03:39 - short article which sets out the
00:03:42 - value of physical models and of course they were
00:03:44 - writing in the 1950s 1940s at the time when they didn't really have computer
00:03:50 - models but they could build physical devices to try and understand stuff and
00:03:55 - that what that's what he was advocating he was saying let's build some simplified
00:04:00 - physical models in order to understand these complex biological systems.
00:04:05 - And of course, now we can build these very sophisticated physical models, which we call robots.
00:04:11 - Right. So what is interesting about this, I was in your answer,
00:04:15 - in some sense now we can look at different kinds of models, right?
00:04:19 - Because using one animal species to try to understand another animal species.
00:04:25 - For that You might use the word model, but it's in a rather different meaning
00:04:30 - than if you talk about constructing an artifact that is a model, right?
00:04:36 - So I think we should also keep that apart because in terms of accessibility.
00:04:41 - You basically say, look, I have more experimental control over this animal model
00:04:46 - in order to make inferences about, let's say, the human brain.
00:04:51 - But if you talk about an artifact, you really have to put that together from the bottom up.
00:04:56 - So it's it's a level of control is is
00:04:59 - of course an order of magnitude larger than
00:05:02 - what you would have with an animal model yeah i mean so with the
00:05:05 - um the physical uh model let's say
00:05:08 - it's a robot you have access to absolutely everything
00:05:12 - that you're interested in because you've built it you also pretty much know
00:05:16 - how it works although what you can never be quite sure until you've built it
00:05:19 - what it's going to do uh and uh that's the great thing about it and of course
00:05:24 - the The reason that people are sometimes skeptical about us using robots as models is,
00:05:30 - of course, all the biological substrates of the animal aren't present in the robots.
00:05:35 - We're trying to approximate them with, you know, silicon and plastic and metal.
00:05:39 - And that's a hard job to do. And so….
00:05:43 - Whether people accept that this is a good strategy depends a little bit on whether
00:05:49 - they agree with the sort of position, I think, that you and I share that these
00:05:54 - physical artifacts are, in some interesting sense,
00:05:58 - devices that can obey the same principles as humans and animals in the way they operate. Right.
00:06:05 - But there are two things here. On the one hand, you automatically then fall
00:06:10 - into this question like, okay, what then makes a good model?
00:06:15 - So we have discussed abstraction. Abstraction is required.
00:06:18 - But then there already we see an issue that models are being – given the computational
00:06:24 - technologies we have, we can start adding a lot of detail.
00:06:27 - So we are slipping away from this original Rosenblatt and Wiener idea of abstraction
00:06:35 - and construction because now we say, no,
00:06:38 - we can throw in as much detail as we want and end up with something like blue
00:06:42 - brain that is as complicated as the original neural system we measured from
00:06:46 - and as incomprehensible, right?
00:06:48 - So this issue of abstraction, I think, is an important one to really protect
00:06:52 - when we speak about models. You have to be clear about your abstractions.
00:06:56 - And then the second thing, and I don't think it's sufficiently appreciated,
00:07:00 - is that robots give us access to behavior, and behavior is a further constraint we can impose on models.
00:07:08 - Like, with animal models, what you see happening is that to keep things controllable,
00:07:14 - because that's the key constraint of any experiment,
00:07:17 - you push animals into a corner of their behavioral space where you might never
00:07:23 - find them in the real world.
00:07:24 - And this is something Lea was pointing to, right, with her animals that live
00:07:28 - now in these pens in the real world. Right.
00:07:32 - You might have constrained or distorted their behavioral output so much that
00:07:40 - actually you don't really understand what these animals are doing.
00:07:43 - And I think behavior is one of the key constraints we have, and we lost it largely
00:07:47 - over many decades, also in the reaction against, if you want,
00:07:52 - behaviorism, which is strong adherence to behavior.
00:07:54 - But the robot allows us, again, to bring in behavior as a constraint on these
00:07:58 - abstract constructs that we call models.
00:08:02 - Yeah, I think, and this is sort of ethological realism that you can begin to
00:08:07 - bring back with robot models is going to really help.
00:08:13 - But I think there's also a little bit of a problem in this field that you also
00:08:20 - need to work with these models.
00:08:22 - You have to have an understanding of how to do controlled experiments still.
00:08:26 - And so we find a lot of people chucking together
00:08:29 - physical devices that resemble some animal
00:08:33 - and it does some behavior that resembles the animal behavior but
00:08:36 - it's not necessarily helping us to answer any question so
00:08:40 - uh the methodology for how we
00:08:42 - go about this what um breitenberg
00:08:46 - who is another great you know uh inspiration for
00:08:50 - me and i think for us both what what he called uh synthetic
00:08:53 - psychology um it's good
00:08:57 - to try and thrash out what is the best methodology or
00:09:00 - what are appropriate methodologies for doing synthetic psychology
00:09:03 - well right and i think that's a bit what we've
00:09:06 - been trying to grapple with at bcbt and
00:09:10 - living machines is to look across the range of approaches people are taking
00:09:15 - and ask questions like how do biologists and engineers need to work together
00:09:21 - in order to progress this field how what is an appropriate way to design an
00:09:26 - experiment that involves a robot as opposed to an animal?
00:09:30 - And I think that your talk today was a nice example of when you were talking
00:09:35 - about vicarious trial and error and linking the literature from Tolman through
00:09:41 - to early studies on how rats look both ways in a maze before deciding which way to turn.
00:09:47 - Recent data from David Reddish showing that there's really activity in the hippocampus
00:09:53 - that shows the rats really thinking about turning left before it goes right,
00:09:57 - and then showing in a robot model that has embedded within it a model of that
00:10:02 - neural system that we can capture those properties.
00:10:05 - And I think within the 10 years we've been doing BCBT, I think our understanding
00:10:10 - of that methodology has progressed, actually.
00:10:13 - And it's not straightforward to see how to do this when you first get going
00:10:17 - with it. Right. No, absolutely.
00:10:20 - So I think this, uh, so, so what I see that methodologically what helped me a lot, um, is that,
00:10:27 - We always have been advancing our models along two lines or two levels of abstraction.
00:10:33 - So one, the distributed adaptive control theory, which is very strongly anchored
00:10:39 - in behavior and robotics, which helps me to think about behavior and underlying,
00:10:44 - if you want, psychological processes of memory, attention,
00:10:47 - decision-making, action selection, and so on, without immediately constraining
00:10:52 - myself by substrate, like how can neurons do this?
00:10:56 - What I always did try to do at that level is to say, well, let's at least not violate the obvious.
00:11:02 - Let's not violate things we know, such as neurons do not broadcast global information
00:11:08 - to each other, for instance.
00:11:10 - So you try to impose these kinds of constraints. You try to keep any kind of
00:11:15 - rules of learning based on local information and so on.
00:11:19 - But then, so that helped us to really build these integrative behavioral models,
00:11:25 - which means you have to really think about real-time, real-world, embodied control.
00:11:31 - And then in parallel, we would run these more detailed models where we say,
00:11:34 - okay, here we make a prediction about, let's say, conjunctive representations
00:11:38 - playing a role in optimal decision-making, right?
00:11:43 - One of the early predictions we had was that sensory motor states are like the
00:11:48 - primitive representation elements for optimal decision-making in a Bayesian sense.
00:11:53 - And that then became a driving hypothesis to look at the brain.
00:11:57 - And with that, we went to the hippocampus. And using very detailed and also
00:12:01 - anatomically and physiologically sophisticated models, working with Cesar Renacosta and John Lisman,
00:12:08 - we really could interpret a lot
00:12:12 - of properties of this hippocampal system that looked initially
00:12:15 - very puzzling but in that case we didn't worry
00:12:17 - too much yet about behavioral consequence but once
00:12:20 - we nailed that model and we had the basic principles in place then we could
00:12:24 - bring it together again in a robot model because now I could replace a bunch
00:12:28 - of the let's say memory systems that I had defined more algorithmically with
00:12:33 - a much more constrained trained model of hippocampal processing that then suddenly
00:12:36 - gave me new features that I had never thought of before.
00:12:39 - Because initially in DEC, I was always thinking about,
00:12:44 - short-term memory in terms of a sequential representation of sensory-motor states
00:12:49 - that I could then link to goals for policy generation.
00:12:53 - But then by mapping that back to the hippocampus, we learned that actually hippocampus
00:12:58 - is doing mind travel, right?
00:13:00 - It's exploiting these sequences in vicarious trial and error to actually make
00:13:04 - predictions and then have an internal simulation of what the world might look
00:13:08 - like when you go in one direction as opposed to the other.
00:13:11 - And that's now a whole new feature that we unlocked in the context of this overall
00:13:15 - framework of DAC and optimal control or robust control in the context of behavior.
00:13:23 - So there's a real, very constructive synergy now between these lines of modeling
00:13:26 - that I think it took, of course, some time to build that momentum.
00:13:31 - But I think now we're really gathering the fruits of that. I mean,
00:13:36 - if we say that the core problem, and I don't think there's one problem,
00:13:40 - there's many problems we're interested in, but one core problem is the question of architecture.
00:13:45 - What is the control architecture of the brain? And obviously,
00:13:48 - that's the core question behind a lot of your work.
00:13:53 - Then there's there's a top-down way of approaching that
00:13:56 - which is you look at behavior you look at what systems
00:14:00 - in principle could generate that behavior and you
00:14:02 - try and build those systems and demonstrate that they're sufficient and then
00:14:08 - there's the more bottom-up approach you look at the circuits that you find in
00:14:12 - in the biological systems that have that behavior and you see how how those
00:14:15 - circuits could have properties that could instantiate those principles.
00:14:22 - I mean, so would you agree there's this combination of top-down,
00:14:26 - bottom-up that you're trying to do sort of in parallel lines,
00:14:30 - and they're trying to feed in to answer these central questions about the architecture.
00:14:35 - So there are the different levels of description here,
00:14:39 - but in a way where we want to push for the high-level description,
00:14:43 - which is going to be then the most powerful set of principles
00:14:46 - yeah no look absolutely you're right
00:14:49 - and what's really important to also not give
00:14:52 - any kind of exclusive status to either of the two you have to do both to to
00:14:57 - to find to identify the constraints um so yeah i see this really as as a combined
00:15:03 - top-down bottom-up approach however i think it's important to observe that But
00:15:08 - to make that work constructively,
00:15:11 - you must commit yourself to clear empirical benchmarks.
00:15:15 - Otherwise, we're in fantasy land and then we very quickly wave our hands and
00:15:19 - we speak of biological inspiration and so on.
00:15:21 - But scientifically, that's not going to help us. So it's really important to
00:15:25 - anchor both of these, the top-down and the bottom-up view and the models that
00:15:30 - come out of it, to very clear empirical benchmarks, which must be grounded in
00:15:34 - the anatomy, the physiology, and the behavior.
00:15:37 - That we know brains and minds generate.
00:15:42 - When we talk about that, though, with modeling, there's different goals that you can have.
00:15:49 - And it's not always prediction that you're going to put first.
00:15:52 - It's more things like sufficiency. Is the model sufficient to generate the behavior
00:15:58 - that we're interested in?
00:16:00 - And you might also have predictions. But my experience is that people who have
00:16:05 - built robot models have occasionally come up with some interesting predictions
00:16:10 - that biologists might not have had.
00:16:12 - But that doesn't happen as often as maybe it could or should.
00:16:18 - But actually they have another contribution to make which they can test the
00:16:21 - sufficiency of the model.
00:16:23 - They can say, I instantiate this theory. Either it does or doesn't work or it
00:16:27 - needs to be improved in this way.
00:16:30 - And I think another thing that it does that people overlook is that it also
00:16:33 - helps the biologists focus on,
00:16:35 - what are the important questions, because biologists, as any natural scientist,
00:16:42 - tend to be attracted by phenomena that they can observe, and then they think,
00:16:47 - oh, well, I can study this phenomena.
00:16:48 - And they don't always pick, and because you can't, the ones that are going to
00:16:55 - be important in terms of understanding function.
00:16:56 - Whereas there's an engineering approach which says, well, in order to solve
00:17:01 - this problem, and we probably have to have this kind of mechanism.
00:17:05 - Can you see that in the brain? So it forces a different kind of question into the biology.
00:17:11 - Well, I know you are a nice guy, but the sufficiency argument,
00:17:17 - I'm not so convinced by that, you know, because it's very weak,
00:17:20 - right? We need testability.
00:17:23 - And sufficiency means something like, well, I haven't rejected it yet,
00:17:27 - but I might in the future.
00:17:28 - And I think what we should not forget get is behind any modular assumptions.
00:17:34 - Assumptions are testable predictions or that from these
00:17:37 - assumptions you can derive testable predictions and i
00:17:40 - think people are often not paying enough attention to that
00:17:43 - they often run away from their predictions because probably
00:17:46 - they know if they get declared the model collapses immediately because
00:17:49 - the the assumptions were just too strong and so
00:17:53 - so that's one thing about sufficient behind is a sufficient model
00:17:56 - are testable assumptions that that
00:17:59 - that we should you'd look for okay well i i mean
00:18:02 - i i think sufficiency is part of it completeness is
00:18:05 - another you know sort of the more of detail of
00:18:09 - the system you can capture the better you're going i think that's
00:18:11 - where your principle of convergent validation comes in it's a it's a question
00:18:16 - of completeness can i account for the biology at multiple levels no no so i
00:18:20 - think if you can take that that that together with the sufficiency then you
00:18:25 - start to uh really be able to say well well,
00:18:30 - anything that can fit all these criteria is going to be a good model.
00:18:34 - And yeah, great if you can do prediction too.
00:18:37 - But I think to some extent, neuroscience, for instance, has been too captured
00:18:45 - by this hypothesis testing idea.
00:18:48 - So to get in a top journal, you have to say that you're proposing a new theory
00:18:54 - of X, you've got this strong hypothesis,
00:18:56 - you tested it, the data you know uh
00:18:58 - came out the right way and you published
00:19:01 - your paper and that encourages people not to
00:19:04 - build on what's gone before so much but to say that oh
00:19:07 - i've done something new uh it it encouraged an
00:19:10 - element of cherry picking which has resulted in
00:19:13 - i think in recent years people saying well we need to go back uh
00:19:17 - to the beginning and measure the brain completely without having these a priori
00:19:21 - theories about how it works so i don't know if hypothesis testing interesting
00:19:25 - the way it's been done in neuroscience can just be imported into synthetic psychology
00:19:30 - so but i'm i really disagree with you now i mean sorry now you're going too far because.
00:19:37 - First i believe that um first convergence validation is it's more about how
00:19:44 - do you deal with the intrinsic indeterminacy of a model right so so models have certain parameters,
00:19:51 - and you use these parameters to fit the curve in the end, right?
00:19:56 - It's like it's a curve-fitting exercise in some way.
00:19:59 - And now you can add as many parameters as you have data points.
00:20:04 - So now you have a problem of overfitting.
00:20:07 - And to counteract that, you want to add more constraints from different levels of description.
00:20:13 - So the conversion validation is a way to think about this indeterminacy of models.
00:20:18 - And we saw examples of that also in BCBT that people would explain a sort of response curve.
00:20:25 - And suddenly we have a new magic parameter in a model to do that.
00:20:28 - And it's nice, great, it's a way to think about that specific phenomenon,
00:20:33 - but as a model, it's not completely satisfactory, you know.
00:20:36 - And then if you now map it to neuroscience, I don't feel at all that neuroscience
00:20:41 - is sort of suffering from an excess of hypothesis.
00:20:45 - To the contrary, I think it's suffering from an excess of technologies that,
00:20:51 - you know, people get enslaved or entrained by the technologies they have available.
00:20:55 - And every technology will unlock another set of potential correlations in the
00:21:00 - universe, and we're going to chase them all down.
00:21:03 - So, actually, I feel that we're really hypothesis-starved and data-rich.
00:21:09 - Well, I think that what you mean by hypothesis or what I mean by hypothesis
00:21:13 - is something a bit different.
00:21:15 - So, what I'm thinking of in terms of a hypothesis is it can be some quite relatively straightforward,
00:21:24 - say, a way of saying, well, people in the past have said this system is wired
00:21:30 - up in this way, and here we have data that shows something different. Oh, like that.
00:21:33 - Okay. Yeah, so what we're lacking in neuroscience is truly explanatory theories
00:21:41 - of systems. Absolutely.
00:21:43 - And yes, explanatory theories should give rise to predictions.
00:21:47 - But I think they should explain first. So, I mean, I've...
00:21:53 - My reading of philosophy of science, I've been kind of, I was,
00:21:57 - you know, skilled in the 1980s when Popper was everything.
00:22:03 - I always forget how old you are. Yeah, exactly.
00:22:06 - But I think the sort of the, all the emphasis on Popperian falsification,
00:22:13 - which is good experimental design,
00:22:17 - has detracted from the need to have strong explanatory theories.
00:22:23 - And I read David Deutsch's Fabric of Reality,
00:22:26 - and he's a physicist, but the first chapter of that book really drives home
00:22:33 - the difference between science as explanation and science as a predictive tool.
00:22:39 - And these are two quite different things. Science as a predictive tool is great,
00:22:43 - but that doesn't necessarily mean you're explaining anything.
00:22:46 - And to me, the first goal of science is to explain. Sure.
00:22:49 - Look, I completely agree with you. So, actually, I did my philosophy of science in German.
00:22:57 - So, you can imagine what the content was like, rather normative.
00:23:02 - But, of course, it included Popper, but not only.
00:23:06 - But, indeed, it's limiting. Popper is definitely limiting, and Deutsch is right.
00:23:11 - And the criteria for any theory is, in my opinion, threefold.
00:23:14 - It is to explain and predict and control.
00:23:18 - Right? And these are, for me, the criteria of a theory. That's why DAC is phrased
00:23:22 - the way it is, because we try to explain adaptive behavior, different forms.
00:23:27 - We make testable predictions, and then we control real-world systems,
00:23:30 - or we control recovery in patients.
00:23:33 - For me, those are the three criteria. And what has helped me a lot to make sense
00:23:38 - of these methodological requirements are these ideas of Bas van Fraassen,
00:23:43 - a philosopher of science who wrote his book, The Scientific Image,
00:23:47 - who is basically advancing the idea that theories...
00:23:52 - Are actually empirically just need
00:23:55 - to be empirically adequate so we just have to accept that
00:23:58 - they're the best possible description we have
00:24:01 - today of a set of phenomena but they
00:24:04 - are contingent they're evolving they're dynamic and at
00:24:07 - some point they might be rejected and that helped me
00:24:10 - a lot to escape from this very stringent normative popperian
00:24:14 - framework where you always are so stuck with
00:24:17 - with falsification because it it is
00:24:20 - definitely limiting and and not helping us forward especially if you're in a
00:24:24 - domain where you have to cut across many levels of description in particular
00:24:28 - mind and brain yeah well i think yeah i think that's absolutely right and the
00:24:33 - problem with with popper and indeed when i was studying this and i think it
00:24:37 - was 1980 this was already
00:24:38 - evident was that if your theory
00:24:42 - is wrong if it's been falsified there's there's
00:24:46 - no way in that framework to say where that's a better theory than any other
00:24:49 - but in fact we we live every day with with theories that are falsified and as
00:24:53 - you i think you said today you know all models are wrong so uh we've got to
00:24:57 - live with these wrong theories wrong models and we've got to try and improve
00:25:01 - them and i think i i think you're right explanation prediction control are three very
00:25:07 - good drivers for how we can make better theories.
00:25:10 - Yeah. So, but I, I do think that we, we, we can't just look at the biological
00:25:14 - sciences as a model for how to do this, uh, new kind of synthetic science.
00:25:20 - I do think that there's a.
00:25:23 - A gap in science for these
00:25:26 - explanatory theories and uh that's
00:25:30 - what we're we're trying to fill and maybe yeah we can we can make predictions
00:25:34 - off that and you made some predictions based on which you described today in
00:25:38 - which uh the the other hard part of this is of course persuading a biologist
00:25:42 - then to test your prediction right yeah sure uh but actually we we have succeeded
00:25:48 - I've succeeded in doing that once in a while, right?
00:25:50 - Including Edward Moser has been testing some of our predictions, which is great.
00:25:54 - But also there, you should see these predictions as a form of dialogue with
00:25:58 - the empirical sciences.
00:26:00 - So it's not like, oh, now I have this normative statement and you empiricists go test it, right?
00:26:05 - It's part of the dialogue that you have to try to establish,
00:26:07 - which is not always easy because the empirical scientists or the biologists
00:26:13 - are not necessarily trained to be very receptive to that. It's not a problem we have.
00:26:18 - But the biggest problem I see is that, actually, I started also with Tolman.
00:26:25 - And actually, what's interesting is that since Tolman, and especially Hull,
00:26:29 - Clark Hull, so we talk early 50s now, there have been no more comprehensive
00:26:34 - integrative theories in psychology.
00:26:37 - And in my opinion, whether you like it or not, neuroscience is largely dealing
00:26:42 - with the mind. All these functional properties of brains, in the end,
00:26:46 - traditionally were in the area of psychology.
00:26:50 - So this is what we tried to explain, but we threw it away.
00:26:54 - And now we're very worried about what neuron X does to neuron Y.
00:26:58 - And of course, since we can measure thousands of them at the same time,
00:27:01 - now it's sort of huge population we're trying to analyze.
00:27:03 - And then we might talk about forms of signal transduction and whatever. Right.
00:27:08 - But we'll have sometimes lost sight of the actual questions we're dealing with,
00:27:12 - which is what's memory, what's attention.
00:27:14 - As you also saw yesterday in Francesca's talk on hippocampal system, I think she did great.
00:27:19 - And she also clearly showed that, right? How we have to get back to these larger
00:27:24 - questions in this case about cognitive development, for instance.
00:27:27 - And I think this is where the synthetic psychology can really help to link mechanism
00:27:31 - to again function or in other words, brain and body to mind.
00:27:37 - I think, I mean, I would agree, but I would say Piaget was also somebody who
00:27:42 - was looking for these large scale theories and, you know, and up until certainly
00:27:46 - the 70s was developing them.
00:27:48 - And I think you can look at Piaget and you can see almost where psychology has
00:27:54 - gone right, but also gone wrong.
00:27:56 - Because in you know taking apart all
00:27:59 - of Piaget's experimental work and he
00:28:02 - was a good experimentalist but he uh over
00:28:05 - interpreted his results I think is what we can say uh and uh people have thrown
00:28:10 - away his whole framework on the basis of of that you know whereas uh and I think
00:28:15 - now development psychology is is fairly theory free or or they're into different
00:28:20 - camps you know which are quite far apart.
00:28:23 - And it's difficult to see how we can pull some of these areas of experimental psychology.
00:28:30 - I know development cognitive neuroscience is kind of a field,
00:28:32 - but it's not as big as it should be, certainly in the psychology community.
00:28:37 - And there is this kind of gulf between the neuroscientists, the developmentalists,
00:28:43 - the cognitive science who are trying to bridge that gap, but not necessarily
00:28:47 - succeeding. Well, look, you're right. But in that sense, I think what happened.
00:28:52 - So, for me, it's really a transitory figure because what's interesting about
00:28:57 - Tolman and Hull, they had this physics metaphor in mind.
00:29:00 - So, they really wanted to build a still logical, positivistic model of theories
00:29:05 - of the mind, which didn't really work out for them.
00:29:09 - But they definitely had the right intuitions. They're really on the right track.
00:29:13 - Piaget already starts to get dissolved in a universe of different experimental
00:29:19 - manipulations and interpretations and resonances with interpretations.
00:29:23 - So Piaget, it's very difficult to sort of condense that into one comprehensive
00:29:29 - theory that covers the whole of psychology.
00:29:31 - Well, he has this theory of equilibration, which has adaptation and assimilation as its two sides.
00:29:39 - And you can really link that. I mean, he didn't talk about dynamical systems
00:29:43 - as such, but essentially what he was describing was how dynamical systems can self-organize.
00:29:50 - And he was writing just before the second wave of connectionism,
00:29:56 - and you can really see how his ideas have been validated to a large degree by
00:30:02 - what people have been doing with those kinds of learning models.
00:30:06 - Um uh it hasn't yet really fed back
00:30:09 - into psychology because i know when i talk to developmentalists they
00:30:12 - all sort of uh get very very upset
00:30:16 - and annoyed when you say oh piaget was was great
00:30:19 - because they think he's ancient history i mean he is a historical figure
00:30:22 - and his experimental results are
00:30:25 - now you know for the museum but i think there
00:30:28 - was a core set of ideas there he was
00:30:31 - really battling cognitivism as well i mean he was he was
00:30:34 - at a time when cognitivism was very strong he was he was
00:30:38 - one of the figures it was exactly they were trying to push aside sure he
00:30:40 - was on the lone voices yeah it's of reason yeah and so i agree with that but
00:30:45 - what i want to say i'm not disagreeing about the the key role that piaget should
00:30:50 - be playing in our current thinking about about the mind but as compared to hull
00:30:56 - and tolman piaget's outlook was not.
00:31:01 - Covering the whole of psychology, he was clearly focusing on adaptation,
00:31:06 - development, and change.
00:31:07 - But for instance, if you look at Tolman, for instance, he really wants to go
00:31:11 - the whole way from genetics,
00:31:14 - the endocrine system, motivation, to cognition, moral thinking,
00:31:20 - and consciousness, right? The whole enchilada.
00:31:24 - And Piaget already reduced it a little bit.
00:31:27 - That's not a criticism of Piaget but but that's it it is for me it was signifying
00:31:32 - the direction psychology was taking it sort of becoming more fragmented and
00:31:36 - also reduce this ambition and I feel that the synthetic psychology we're now
00:31:41 - advancing using that term from Breitenberg also is a real opportunity,
00:31:47 - to put psychology back on the map of science to say no this is what it's really
00:31:52 - about if you want to understand the brain these are the questions we got to
00:31:55 - worry about yeah I mean I mean,
00:31:56 - I think one of the reasons maybe why we're harking back to these figures from the 50s,
00:32:01 - 60s, and 70s is that it has become much harder in science to be the sort of
00:32:07 - generalist that these figures were in some way and to be able to go across different
00:32:14 - disciplines of mind and understand them.
00:32:18 - And so there's a need to build sort of interdisciplinary expertise,
00:32:23 - which is going to be largely teams.
00:32:27 - And there needs to be an understanding that that should be the strategy.
00:32:32 - And I think part of the issue at the moment is it's still hard to do that.
00:32:38 - There's a lot of talk about interdisciplinarity and how to support that,
00:32:41 - but still some of the aspects of the world of science militate against really
00:32:50 - interdisciplinary work.
00:32:51 - Well, I agree. This is a very important point because I think interdisciplinarity
00:32:56 - definitely is high risk.
00:33:00 - And in the face of scientific disciplines, they get more specialized,
00:33:05 - they get more technology-driven, that are also very much influenced by their
00:33:09 - own local cliques and networks and tribes, if you want. It's very tribal.
00:33:15 - On top of that, we have incentive systems that are driving people into very ruthless competition.
00:33:24 - Of course, this leads to that all the forces are stacked against what we need.
00:33:28 - So the forces are stacked against the multidisciplinarity and also the freedom
00:33:32 - to try risky hypothesis, right?
00:33:35 - So certainly in an environment where right now much more value is placed on
00:33:39 - what's called innovation or immediate impact as opposed to, let's say,
00:33:44 - fundamental advances. So yeah, you're absolutely right.
00:33:48 - And I think that the way through that for us has been that in the synthetic
00:33:53 - psychology, we actually are building artifacts and we are also having impact in robotics.
00:33:59 - We can solve certain problems or we can contribute to solving problems,
00:34:03 - not only in the psychology or the application of robots, but also really in
00:34:08 - the control of robots that gives us a bit more traction and maybe a bit more
00:34:12 - also if you want protection from this kind of criticism.
00:34:17 - And of course, also in our case, that we were successful in mapping the theory
00:34:22 - of DEC to the most effective neurorehabilitation method for stroke recovery.
00:34:28 - Of course, has helped us a lot.
00:34:30 - Because basically, we have changed, if you want, the criteria on which we have
00:34:35 - to judge theories in neuroscience and psychology by saying, well,
00:34:39 - it's not only about convincing your peers, it's about actually having measurable
00:34:43 - impact in the real world.
00:34:44 - Well, if you claim to know how the brain works, please go fix a patient somewhere.
00:34:48 - And if we would just apply this criterion consistently, I think we would reduce
00:34:55 - a lot of noise in the field.
00:34:57 - And get things recalibrated in a maybe a bit more constructive way.
00:35:02 - Yeah, I think that's right. And it's actually, with hindsight,
00:35:07 - it's a bit surprising to me that there isn't more emphasis within the field
00:35:13 - of cognitive science on trying to discover new therapies. I think it's coming back now.
00:35:20 - I certainly hope so. But it didn't feel that way 10, 20 years ago.
00:35:25 - And to some extent i think cognitive science
00:35:28 - has has tried to keep its focus on understanding
00:35:32 - mind and understanding the uh unimpaired
00:35:36 - mind as a first step is a good way to go uh
00:35:39 - it's also seeded some of the potential uh to do uh applications and impact to
00:35:45 - disciplines like ai unnecessarily i think because actually a lot of the best
00:35:50 - ideas in ai come through psychology and cognitive science and then get sort of rediscovered as AI.
00:35:58 - But then I think the other way around, the cognitive science hasn't done itself any favors,
00:36:04 - by disconnecting a little bit from psychiatry and all these other disciplines
00:36:09 - that are concerned with brain disease and impairments.
00:36:14 - Yeah, but you know what plays a role there? I think that given the incentive
00:36:17 - structures we try to survive in, people have become very risk-averse.
00:36:22 - And as we earlier agreed, all models are wrong.
00:36:27 - But as long as you can resonate with your peers and support each other's models,
00:36:32 - everybody can feel comforted and supported.
00:36:38 - But as soon as you map to the real world and you say, okay, here's my model
00:36:41 - of schizophrenia and now I'm going to bring it to the clinic, that's high risk.
00:36:45 - Because then if it doesn't work, it's really very obvious.
00:36:49 - And I think, so that's not a problem. People are very risk averse.
00:36:52 - That's why they don't want to make that step. So we have to really rethink the
00:36:55 - incentive structures in order to open that up.
00:36:58 - But in a way, I think what we're also saying is that it's not helpful to have
00:37:04 - this gap between so-called pure science on one hand and so-called applied science on the other.
00:37:10 - The best basic or pure science also has implications in the real world that
00:37:16 - people have overlooked sometimes.
00:37:19 - And I think there's also, again, within sort of academic publishing,
00:37:25 - there's sort of a preference for these sort of big discoveries in pure science
00:37:31 - so that you can make your career by just advancing knowledge without having
00:37:35 - to have innovation impact.
00:37:37 - So I'm kind of in favor of having more emphasis on the innovation side as a
00:37:44 - way of driving thinking within science, provided it's done sensitively. Right.
00:37:49 - This is really, these are good points, and I agree with you.
00:37:53 - And sometimes we can go back to the cyberneticians again, right?
00:37:57 - Because many of them had actually real-world concerns. Yeah.
00:38:00 - Like Wiener working on control systems, also...
00:38:06 - Ashby was a psychiatrist, right? So they were worrying about real problems.
00:38:11 - And for them, as far as I understand reading about them and reading their work,
00:38:16 - there was no divide between these applied concerns and the principles they took from it.
00:38:22 - They never complained about it or saw any thresholds or obstacles between that.
00:38:28 - And now this has been all reconceptualized in a way.
00:38:33 - Actually, also Vannevar Bush, right? Shortly after the Second World War,
00:38:37 - science, the endless frontiers, setting up the National Science Foundation of
00:38:41 - the States, was very much prosperity depends on science.
00:38:44 - There's a natural linkage between the human condition and science.
00:38:48 - And that was never really questioned. And now it seems we have to have this
00:38:53 - divergence like, oh, we have all these people concerned about the so-called
00:38:56 - basic questions. And then we have other people who do the sort of application
00:39:00 - of these and the two shall never meet.
00:39:03 - And I think that's a massive mistake. take. I don't think that that needs to
00:39:06 - be the case. And I also see it in our own work.
00:39:09 - Let's take again the example of neurorehabilitation. Every intervention is like
00:39:14 - an experiment and every patient provides new and valuable information on the basic theories we have.
00:39:20 - Like we have all sorts of ideas about how error might modulate learning in stroke patients.
00:39:27 - These are hypotheses and they're being tested every day in the clinic with real
00:39:30 - patients and we get the information back in real time, and it's advancing our theories.
00:39:35 - And I think this tight coupling between basic and applied science,
00:39:39 - which I then call Vico's loop, after one of my heroes, John Battista Vico,
00:39:43 - of the fact and the truth are reversible.
00:39:47 - We can find these loops, we can find these synergies between application and
00:39:50 - basic science if you look for it.
00:39:53 - And it's not happening enough. And another thing which is happening now,
00:39:58 - which certainly wasn't as evident a decade ago, is people actually looking at
00:40:06 - the technologies that are coming out of robotics,
00:40:09 - AI, cognitive science, and now saying, well, do we want these technologies?
00:40:13 - And so there's much more of a focus on, okay, where is this going to go?
00:40:19 - You know, who's going to benefit from this? and there's a growing realization
00:40:25 - that some of the technologies that have developed in the past 20 years have
00:40:29 - increased prosperity, but for the few rather than for the many.
00:40:33 - And we want to do something to counter against that.
00:40:38 - So, I mean, if we want, so I think it, you know, a lot of our science is publicly
00:40:43 - funded, but also I think as scientists, we want to feel that the contribution
00:40:47 - we're going to make is going to have a positive impact.
00:40:49 - So how do we build that into, uh
00:40:52 - what we're trying to do here how do we ensure that our
00:40:56 - goals and well not just our goals
00:40:59 - of the work that we're doing we're going to focus it
00:41:02 - towards positive impacts beneficial impacts right well that's really i think
00:41:07 - the key question we got to answer collectively and i i think that um we have
00:41:12 - to pick our problems carefully but maybe also rethink a little bit how How have
00:41:17 - we really structured the scientific enterprise?
00:41:20 - I think the main criterion right
00:41:23 - now for our science is how much you get appreciated by your colleagues.
00:41:29 - And that's very strange, right? Because these tribes also developed their own
00:41:34 - biases and expectations that might be completely besides reality.
00:41:38 - And if you also look at impact, for instance, you can look at many domains,
00:41:44 - cardiology, cancer, brain disease, education.
00:41:50 - Education, I don't think that after many decades of following this sort of peer-based,
00:41:58 - validation of a field has led to now a massive progress.
00:42:03 - I mean, of course, we cannot say that's been zero progress, but it has not been fantastic.
00:42:09 - Like in the case of neurorehabilitation, we looked at this over the period 1975
00:42:13 - till now. Now, we analyzed dozens of meta-analysis
00:42:18 - of the impact of stroke rehabilitation, and it stayed the same.
00:42:23 - So your chances of recovery 40 years ago are the same as to now.
00:42:29 - Well, that should give us some pause because in the same period of time,
00:42:32 - billions of euros have been spent on brain research and associated fields. Zero impact.
00:42:41 - So what's wrong here? And I think what's wrong here is really the model in which
00:42:45 - we have been pursuing this science, which was sort of divide and conquer,
00:42:49 - run after the technologies, go for greater levels of detail,
00:42:54 - or essentially just measure whatever the tools allow us to measure without even
00:42:58 - posing questions or even driving hypotheses.
00:43:01 - And I think by sacrificing psychology as our anchor point of questions,
00:43:06 - this has been the consequence.
00:43:08 - And I think the way back is still long and we're not doing very well.
00:43:12 - Uh, so that sounds a bit pessimistic. No, because we're doing something about
00:43:17 - it. We're in the clinic. We already treated 800 people and that's the beginning.
00:43:20 - There's still about 59 million to go.
00:43:23 - But in terms of, you know, getting, uh, science back on track towards solving
00:43:28 - some of these problems facing, uh, humanity,
00:43:32 - uh, what is, what are the quick steps that we could make, you know,
00:43:38 - sort of collectively, or at least we could agitate as a community to do more of.
00:43:41 - Okay, yeah, sure. Now, look, you're right.
00:43:44 - And that's also interesting. You see, both with Tolman and Hull,
00:43:49 - I talked about earlier, they really had this clear outlook.
00:43:53 - They also, of course, lived through the Second World War and they also saw it
00:43:56 - as their responsibility as psychologists to make recommendations and have impact at that level.
00:44:01 - Like, how do we advance the human condition?
00:44:04 - And I really think that should be our concern today. So that also means we have
00:44:08 - to rethink our science. Like, how are we going to really deal with the human condition?
00:44:12 - Is the way in which we have organized the scientific enterprise actually helping
00:44:17 - us to advance human condition?
00:44:20 - Take as an example, economy. Economy, in the end, is strongly dependent on human behavior.
00:44:26 - So how can we advance economy without linking it very, very closely to psychology
00:44:30 - and psychology to neuroscience, right? So I think there are bridges we have
00:44:37 - to build to make progress there.
00:44:40 - And we also have to, I think, be willing to pose a larger question.
00:44:47 - Because if we want to change the human condition positively,
00:44:50 - which we better do sooner than later because we're facing some serious challenges,
00:44:54 - and global warming is only one of them.
00:44:56 - But let's say income inequality, which is a very much social psychological phenomenon,
00:45:02 - I think is a massive stress on our society we have to deal with.
00:45:06 - But that means we have to understand things like greed and hoarding and tribal
00:45:11 - behaviors and forces that drive inequality.
00:45:15 - Do we have anything in our hands today to do that? No. Then you can say,
00:45:20 - well, look, you know, we have to improve education.
00:45:24 - But the most recent meta-analysis on the main factors that drive education in
00:45:30 - terms of its impact in terms of learning outcomes are not very conclusive.
00:45:36 - So there we still don't really know what we're doing and I think,
00:45:39 - It is really urgent that we rethink very carefully how we organize the scientific
00:45:44 - enterprise, how we link it to the real world, and also how we advance more integrative
00:45:49 - paradigms to bring these disciplines together.
00:45:53 - Because I think to understand the human condition in the end means we have to
00:45:56 - go from almost, let's say, the physics of bodies and brains to the sociology
00:46:02 - and culture that they can give rise to.
00:46:05 - And right now, we don't have these integrative paradigms. I think also,
00:46:08 - I mean, economics is vitally important for, you know, sort of fulfilling basic human needs,
00:46:15 - but also there are other needs and there are, you know, sort of epidemics of
00:46:21 - social diseases or, I mean, loneliness is not a disease,
00:46:26 - but it's academic portions in a world that has never had higher population.
00:46:32 - And higher levels of social networking, I can add to that. Well, that's true.
00:46:37 - Yeah. So if we can advance theories that can help address that.
00:46:42 - So I think, and partly this is about what collectively we are trying to optimize.
00:46:48 - And so I think the beginning of this has to be some theory about the human life and what it's for.
00:46:56 - You know i mean it's not for anything i don't think as materialistically
00:46:59 - would say but at the same time you know we have
00:47:02 - uh this ability to be aware of ourselves and
00:47:05 - to decide you know uh what we want our life
00:47:08 - to be for uh to set our own goals um and
00:47:12 - so there's uh one
00:47:15 - of the goals i think of our science is to try and help answer
00:47:18 - that question because what personally i wouldn't
00:47:21 - want to do is to spend my life basing my my goals and and my daily efforts on
00:47:27 - uh things which turn out to be illusory and i think that is a risk you know
00:47:32 - that there are lots of false prophets out there giving you reasons to live which
00:47:36 - aren't aren't good ones well this is this is a.
00:47:41 - This is the crux of the whole story, right? Because basically what you're saying
00:47:44 - is that we must go back to the basic question of eudaimonia.
00:47:51 - And that was the reason also why the Greeks never built a rocket to fly to colonize
00:47:57 - Mars, because their main question was, what is the virtuous life?
00:48:02 - And that really means like, what's the good life, right? What is virtue?
00:48:05 - What are the norms we should apply to that? And we have drifted away from those questions a lot.
00:48:11 - We always have felt like, well, there are forces shaping our society that we
00:48:17 - don't need to touch from that scientific perspective, but maybe we should.
00:48:20 - Because in our also definitely more and more secular society,
00:48:23 - where else can we get our norms from, if not from a deep understanding of who
00:48:28 - we are and what our limitations are?
00:48:29 - And I think this is an important responsibility now for psychology and neuroscience
00:48:33 - to pursue, but there are very few people who do that. And if they do it,
00:48:39 - they do it after retirement because it is just a very risky proposition right now.
00:48:43 - Yeah. I mean, I think it was not this Pope, but the last one that said that
00:48:46 - some of these questions are for religion, not for science.
00:48:50 - But you can sort of see what he's
00:48:52 - getting at in that maybe the scientific outlook as it is, is too narrow.
00:48:59 - Arrow um on the other hand uh
00:49:02 - i don't think in in principle that we
00:49:05 - should be excluded from taking a more scientific approach to these questions
00:49:10 - of of how we uh what we should go for in life but perhaps we also have to stretch
00:49:16 - what we're doing as science as scientists to encompass more of the humanities
00:49:20 - as an outlook absolutely what i find interesting there is that.
00:49:26 - And it's very funny that we live in this sort of bubble of happy beliefs about who we are as humans.
00:49:33 - For some reason, we have great difficulties with recognizing also the kind of
00:49:39 - destructive animals we are and definitely can be.
00:49:45 - And we often see that now in our society as sort of anomalies that we can sort
00:49:49 - of, we don't need to worry about too much.
00:49:52 - Criminals, we can lock up their way and for the rest, everything is fine.
00:49:56 - But understanding human condition to me also means especially the destructive
00:50:00 - forces that humans can mobilize because only then can we, if you want,
00:50:05 - defend the future of humanity against those destructive forces.
00:50:09 - And I think I would see that as an important objective of our research.
00:50:13 - Also, for that reason, this is one of the motivations why we are very much involved
00:50:19 - in the study of the Holocaust and the commemoration of the Holocaust and Nazi crimes,
00:50:24 - because I feel this is a source of highly relevant information about the limitations of humanity.
00:50:32 - I mean, however ugly and terrible this is, and it really is terrible,
00:50:38 - we have to understand these boundaries on what humans can accomplish for good
00:50:45 - and bad, because only then can we find countermeasures to protect ourselves from ourselves. Yeah.
00:50:52 - I mean, this is one of the things that has stood out for me in this BCBT is
00:50:56 - actually how many times, sometimes in the talks,
00:51:00 - sometimes in interviews and very often over dinner that the conversations have
00:51:05 - come around to these bigger, if you like, questions about,
00:51:10 - you know, what are we going to do about human society and where it's going?
00:51:17 - How we can take account of the fact that we are, I think, as John Doyle put
00:51:25 - it, the sort of apes with guns, and that's a very dangerous situation.
00:51:29 - And as you say, we have to understand our own limitations better.
00:51:34 - We also are now not just apes with guns, but apes with internets and AIs,
00:51:41 - and there's all the potential that that can have, and robots,
00:51:44 - of course, to both improve our condition, but fundamentally change it in a way.
00:51:51 - Because some of these devices we can use to almost transform what we are.
00:51:56 - Sure, absolutely. And, you know, we're getting to the point,
00:52:00 - and I think this is one of the exciting but also frightening things about this
00:52:04 - area of living machines where we can interface our brains.
00:52:07 - We already do, of course. I mean, a screen is a kind of interface,
00:52:10 - but we are able more and more to connect to these technologies in a very direct
00:52:16 - and intuitive way, which is very exciting.
00:52:20 - But that the implications of that i think for the human mind i think we also
00:52:25 - touched on this you know how technology changes the minds of children yes and
00:52:29 - the minds of our children are going to be different from ours when we were growing
00:52:33 - up so uh and then i think so what we want to do probably.
00:52:37 - More in the future with our school and our uh our
00:52:41 - conference is is to see how we can
00:52:44 - bring all these things together because they do they do seem to be strands that converge
00:52:48 - in a way as you which suggests why we call it convergent science absolutely now
00:52:51 - look i fully agree and to also
00:52:54 - finish up this this last point we were discussing one one
00:52:58 - massive weakness that that the human mind has is that we always adjust our set
00:53:04 - points very quickly you know so um we see disasters happening around us it might
00:53:10 - be mad dictators with nuclear arms or or mad dictators dictators,
00:53:15 - controlling the most powerful nation in the world, it might be hurricanes,
00:53:20 - it might be global warming, the Anthropocene is upon us, it might be self-driving cars, right?
00:53:26 - There's a long list of things that not that long ago we saw as real threats
00:53:30 - and we were really worried about it.
00:53:32 - And very quickly we changed our set point. We are like...
00:53:38 - Cognitive homeostats, you know, so we very quickly zoom into this new set point
00:53:43 - that sits at this new average now of disastrous challenges and it doesn't worry us anymore.
00:53:52 - So I think this is a weakness of the human mind that we are often driven to
00:53:58 - action by homeostatic and also emotional systems that very quickly readjust to these new norms.
00:54:06 - And I think we have to really develop a metacognition here, a critical self-reflection
00:54:11 - that tells us, no, we cannot give in to certain outrages of our human norms.
00:54:17 - And we must insist that we're going to act against them with the best of our science.
00:54:23 - And there, the program, I think, is also partially spelled out already at the
00:54:27 - level of the United Nations where the sustainable development goals have been defined.
00:54:33 - If you look at them in detail, then I think here they're a bit redundant,
00:54:38 - but there's some obvious things like equal opportunity, food security,
00:54:43 - right for education, and so on.
00:54:46 - So that agenda is defined, but what we're completely lacking today is a comprehensive
00:54:50 - science, technology, and also socioeconomic agenda to bring these things about.
00:54:58 - And I really feel that the living machines community and the BCBT community
00:55:04 - that we're trying to grow should start to focus itself more,
00:55:09 - occupy themselves more with those challenges because this is our future,
00:55:13 - this is the future of our children.
00:55:17 - And, of course, we don't want to compromise their future too much.
00:55:19 - So, yes, we have to really rethink to place the school and living machines and
00:55:26 - also the conversion science network on a plateau where we can really start to
00:55:32 - be also agents of change,
00:55:35 - if you want, or positive change by helping us to build new frameworks, look upon ourselves,
00:55:40 - and how we can change the reality for the better.
00:55:43 - I mean, I think you're right that we need to be very much aware of what we might
00:55:49 - be losing as we are gaining new things.
00:55:52 - We also, I think, have to be, and it's the flexibility that you're talking about
00:55:59 - to sort of adapt to culture is almost what's got us here as a species.
00:56:03 - And you know and maybe we are a bit complacent
00:56:06 - but we have in the last hundred years done amazing things
00:56:09 - i think we've reduced absolute poverty in the world from
00:56:12 - 80 percent of the population down to below 20 and
00:56:15 - it's forecasted to go down to 10 so there's there are achievements that we can
00:56:19 - point to that things are going in the right direction so i'm i'm not absolutely
00:56:23 - pessimistic about you know the fact that we can't solve these challenges and
00:56:27 - in the past we've always sort of have relied on technology and science as one
00:56:31 - of the paths, not the only one, to resolve challenges.
00:56:34 - And I think that's part of the agenda is to say that, you know,
00:56:37 - we can push science and technology in the right way and we can deal with these problems.
00:56:44 - I don't want to sound too pessimistic because actually intrinsically I'm not.
00:56:47 - I'm an optimist. That's why we engage with many challenging questions.
00:56:51 - But we should be careful not to fool ourselves, right? To claim like,
00:56:55 - okay, humans have done all these great things.
00:56:59 - And I think that's also very doubtful. It's also a story we like to tell ourselves.
00:57:04 - Like, in the end, we were great. We put a man on the moon.
00:57:06 - But if you look, for instance, at the domain of health where we are active,
00:57:11 - the amazing thing is that the
00:57:13 - life expectancy has improved over the last decades dramatically. Right.
00:57:18 - So right now it would be people who are born now will be like over 80.
00:57:23 - And it used to be the 60s, right? So for us, it will be the 70s somewhere.
00:57:27 - So yes, dramatic improvements.
00:57:29 - But if you look at the change in healthy life expectancy, it has stagnated in
00:57:36 - the same period, right? That's very interesting.
00:57:40 - So we can keep people alive longer, but that means it will be miserable longer.
00:57:44 - All right? So we should also… Well, they'd be paying for healthcare for longer.
00:57:48 - Good there might be something there yeah no but but you see that that's
00:57:51 - very tricky so when you say okay poverty has been reduced but what
00:57:55 - does it really mean so have we have built
00:57:57 - maybe a poor middle class that is living every day under all sorts of stress
00:58:02 - because they have to pay off their debts of their credit cards and their their
00:58:05 - banks and their as we have seen also in the the crash that happened in our economies
00:58:11 - about 10 years ago right so so i think we should become more critical about
00:58:16 - these stories we tell ourselves.
00:58:17 - Like also this popular story you hear about, oh, humans have become less violent
00:58:21 - because less people are being killed.
00:58:24 - But maybe that's exactly the same point as with improved life expectancy.
00:58:28 - Yeah, we live longer, but we're miserable.
00:58:30 - So, okay, we're not being killed anymore, but maybe now we're just slowly,
00:58:35 - we're made to die slowly because of cardiovascular disease that's induced by
00:58:40 - stress, but doesn't count anymore as aggression, right?
00:58:42 - So I think we should be more subtle in how we interpret these kinds of so-called
00:58:47 - accomplishments of humanity.
00:58:48 - Yeah, I think we maybe reduce one problem and we create another one somewhere else by doing that.
00:58:55 - And I think that's sort of the way it goes.
00:58:57 - But to some extent, these problems we're talking about are a product of our
00:59:03 - own misconception about ourselves.
00:59:06 - So for me, the fundamental goal or a fundamental goal for our field is.
00:59:11 - To understand what we are and perhaps correct some of those misconceptions.
00:59:16 - So some of those misconceptions, I think you can trace their roots in European
00:59:22 - philosophy and thought and to people that have.
00:59:27 - You know, taken the idea of the soul and then translated it into the modern
00:59:32 - notion of consciousness and this idea that you are a consciousness,
00:59:35 - which is somehow in your body but
00:59:38 - not necessarily of it uh and you know
00:59:41 - that the most recent version of this is the notion that you
00:59:44 - could somehow take that and upload it into a machine
00:59:47 - or a robot and have eternal life and i think this this um
00:59:51 - sort of very western idea of the self is
00:59:54 - at the root of some of these own problems uh our
00:59:58 - own unhappiness for instance uh because i think
01:00:01 - our happiness is linked to all these other physical things not just
01:00:04 - in a sort of the consciousness is floating inside our.
01:00:07 - Heads and then uh a lot of
01:00:10 - these other problems that we have in the world are are down
01:00:12 - to this kind of individualism that we have really sort
01:00:15 - of put on a pedestal in the west exactly so that that's
01:00:19 - a really interesting point you know because in the western
01:00:22 - cultural tradition there is this this myth
01:00:25 - of the the the lone or the individual genius who
01:00:29 - sort of gives rise to change right so we
01:00:32 - think about also in art for instance you would think about the
01:00:34 - great composers mozart beethoven and so on who now
01:00:37 - are sort of dictating their will on all future
01:00:40 - generations of musicians because there's a fixed score from which that piece
01:00:45 - is now forever produced and we seem to apply that model to or to ourselves also
01:00:51 - when we are We have outstanding individuals like the captains of the technology industry.
01:00:59 - They have suddenly forgotten that they might have gotten in that position more
01:01:03 - by sheer luck and contingency and not necessarily by individual genius.
01:01:09 - And so it's really interesting to see that these are the characters that are
01:01:13 - now suddenly pontifying about how they want to upload their mind or how they want to live forever.
01:01:18 - But think about, imagine we would have the mind of Nero on a hard disk somewhere.
01:01:25 - Who would care about turning it on, right?
01:01:28 - Maybe we should have some norms or some ethical rules that would prevent us
01:01:34 - from poisoning the cultural and psychological environment of future generations
01:01:38 - with the complete nonsense that the current self-declared prophets of the techno
01:01:44 - religion seem to have of themselves.
01:01:47 - Yeah, I think that that is a risk and we should, you know, sort of one of the
01:01:51 - things that we should be talking about, actually one of the more near-term risks.
01:01:56 - But I think what we can do in a positive way, I think, is to,
01:02:02 - help people including those people understand themselves
01:02:05 - better because they're suffering from a delusion uh and
01:02:08 - that delusion is probably making them unhappy uh and also you know wishing for
01:02:14 - this silicon afterlife or whatever um and money bro yeah and i think the so
01:02:20 - the the interest that we have in human subjectivity which i know you're doing a lot of work on.
01:02:29 - Should allow us the goal should be to have a
01:02:32 - new understanding of the human self or that helps
01:02:36 - us be more content with what we are because my own
01:02:40 - sort of ideas in this i guess i look to
01:02:43 - sort of more to the western religion eastern religion such as uh
01:02:47 - zen buddhism which i think have had this idea for a
01:02:50 - longer time but they've never you know they got
01:02:53 - us to a certain point with it and i think we can take this
01:02:56 - idea and actually well say why these things that they discovered
01:02:59 - uh might be true um through our understanding of
01:03:02 - the sure but buddha buddha is a bit in some
01:03:05 - sense also a bit frustrating right because hey look unhappiness results from
01:03:11 - wanting things so that means i can be content and avoid unhappiness by not wanting
01:03:18 - things so So now the whole religious exercise is focusing on not wanting things.
01:03:24 - Okay, it's a method. You can try to do that.
01:03:27 - But it also means you're denying part of our humanity, which is we intrinsically want things.
01:03:33 - Yeah. So there might be other ways to deal with that that might be,
01:03:36 - let's say, more liberating and creative than saying, oh, let's not want. Yeah. Right?
01:03:41 - In some sense, the Judeo-Christian tradition is something similar,
01:03:45 - right? We should also not want outside of a very well-defined framework and
01:03:51 - then in some sense use religious meditation to stay within that want.
01:03:55 - And if we exceed the framework, then there are all sorts of rituals to get back into it.
01:04:00 - But maybe we also have to accept the fact that we are religious.
01:04:04 - Wanting things that we do pursue these
01:04:08 - wants often in irrational ways maybe it's
01:04:11 - also through that acceptance that we can find a better
01:04:14 - way to to deal with it because if you look now if
01:04:18 - you look at at the way humans now go off
01:04:21 - the rails in our society for instance through
01:04:24 - addiction right which is a massive problem especially in the
01:04:27 - states um i think this
01:04:30 - tells us something think very deeply about our society not
01:04:33 - being able to deal with people's wants in in
01:04:37 - a sort of a well-managed way and there
01:04:41 - you see for instance here in Europe or also in Holland where I'm coming
01:04:43 - from the attitude towards let's say drugs of
01:04:46 - abuse has been very different and less suppressive and
01:04:50 - the problems of want in that domain of
01:04:53 - addiction are there but it's often more towards alcohol and.
01:04:56 - Less towards the kind of opiates that you see creating havoc
01:05:00 - in the states right so so yes we can then say oh
01:05:03 - let's all become buddhists but okay the west
01:05:06 - is also very expensive to go to these sort of buddhist monasteries to
01:05:09 - meditate and look at the wall and so on but maybe there
01:05:12 - are also more scientific informed methods we can
01:05:15 - apply here to manage our wants through
01:05:18 - let's say other forms of metacognition i mean i think you're
01:05:21 - right and i think uh you know i was um a
01:05:25 - big reader of herman hess when i was young and
01:05:28 - uh his book the glass speed game is uh i
01:05:31 - always read that as a metaphor for cognitive science it
01:05:34 - was these groups of group of people that studied music
01:05:38 - and maths and all these esoteric things
01:05:40 - as a kind of path to enlightenment uh and it
01:05:43 - was one of the things that inspired me to be a cognitive scientist and i think you
01:05:46 - know buddhism is a bit like that there are the monks in in their
01:05:49 - monastery of which there are very few and they meditate all day and then
01:05:52 - there's everybody else uh and not everybody can
01:05:55 - be a monk clearly and not everybody can or wants
01:05:58 - to be a cognitive scientist so what we can do is
01:06:01 - uh you know sort of with the insights that we
01:06:03 - gain we can hope to reshape society and one
01:06:06 - of the things that stood out for me in this school was uh the talk that we had
01:06:13 - on decision making and agency from patrick haggard and you know patrick was
01:06:20 - you know a little bit hazy about uh you know.
01:06:25 - Getting rid of the homunculus, I mean, he wanted to do that,
01:06:27 - but he didn't want to entirely vanish it.
01:06:29 - He still wanted something which was influencing our decision process.
01:06:34 - But I mean, I think for me, perhaps for you, whatever's influencing that decision
01:06:39 - process, there's not going to be a subject in there.
01:06:41 - Ultimately, there's going to be other brain processes. processes,
01:06:43 - maybe some of those processes we are going to label as self-processes,
01:06:48 - and that's getting close to our theory of self.
01:06:51 - But I think what Patrick was saying, and I would agree as well,
01:06:54 - is this is going to change our whole way of thinking.
01:06:57 - If this scientific view percolates into society,
01:07:01 - it's going to change our way of thinking, for example, about how we deal with
01:07:06 - people that abuse drugs,
01:07:07 - the whole notion that we We punish people because they do crimes,
01:07:12 - may have to go out the window, and we may have to globally adopt a more Scandinavian
01:07:18 - view of prisoners' rehabilitation.
01:07:22 - And there might be a path to do that by bringing some of these scientific ideas out to the public.
01:07:28 - Sure. No, I completely agree with that. You're right. Right.
01:07:31 - And what's interesting is I showed it in our own models, in this case of foraging,
01:07:36 - that we have this hidden assumption about that the mind is operating as a single integrated entity.
01:07:45 - But that's an assumption, right? And that's also what Patrick,
01:07:48 - in some sense, was alluding to.
01:07:50 - And also others like Mike Kazanica has been talking about this,
01:07:54 - that you have these sort of dual process ideas about the mind, where,
01:08:00 - or as Mike Kazanica indeed also showed in the split brain patients he worked
01:08:05 - on from Sperry when he was a student,
01:08:07 - but we don't, our metacognitive system that leads to our ability to declare
01:08:14 - and experience and have theories about ourselves,
01:08:17 - might actually be rather disconnected from the subconscious processes that drive
01:08:24 - our behaviors, as for instance, core behavior system that Bjorn Merker would
01:08:28 - talk about, that I mentioned also this morning.
01:08:30 - These are very primitive systems that drive our wants, right?
01:08:35 - And they're automatic, they're strongly genetically defined,
01:08:40 - they're operating really outside of the window of consciousness.
01:08:43 - We have no idea what they do, but we see ourselves doing things,
01:08:46 - right? So I think this hidden assumption of this continuity and transparency
01:08:51 - of operation in the mind is breaking us up here.
01:08:55 - And if we are able to see more also the internal contradictions that exist in
01:09:00 - minds, I think it might help us a lot by devising better interventions.
01:09:06 - So there is a path to increasing contentment through this, but what does it
01:09:11 - involve then if my midbrain is leading me down paths which may be immediately
01:09:18 - rewarding but ultimately bad for me? What's the strategy?
01:09:21 - Well, at least have the metacognition to recognize that.
01:09:24 - I think very often there is no metacognition about it.
01:09:28 - If people don't even know that they
01:09:31 - can be driven by these forces that they
01:09:35 - don't have direct cognitive access to then stuff happens
01:09:37 - to them and they would say like well yeah i just it happened to me i did it
01:09:42 - right but maybe what we have to understand is that through our consciousness
01:09:47 - we can will ourselves into the future we can will ourselves towards being one
01:09:53 - person or another other by biasing our actions against.
01:09:56 - Well, now you're getting a bit mystical because this is, where's this homunculus
01:10:00 - that can will come back from? No, it's a dual process theory, right?
01:10:03 - So I'm saying that, that your executive systems combined with medical systems
01:10:09 - can allow you to build a theory about yourself.
01:10:12 - And we all have this. And these theories can be more or less elaborate because
01:10:15 - they say, well, I like running and cycling and I'm not much into chess nowadays.
01:10:20 - Right? So I have a theory about myself. But these theories can be expanded to
01:10:24 - how it would behave under certain conditions where people might show destructive behavior, right?
01:10:30 - And if we just understand that we are able to have an insight in these factors,
01:10:37 - it's in some sense a very psychoanalytic view on how we operate.
01:10:40 - But yes, I think this can at least be beneficial.
01:10:43 - Yeah. And so by acknowledging its dual process or multiprocess,
01:10:48 - and one process might not have direct access to the other.
01:10:53 - We have to learn to develop theories of ourselves. Right now,
01:10:56 - we don't train people to do that.
01:10:59 - But it sounds a little bit like monks learning Zen Buddhism,
01:11:03 - but we're now training them to think metacognitively.
01:11:07 - Exactly. To say, oh, I have wands, and I have wands that might damage others
01:11:11 - or that might damage myself on the long run.
01:11:14 - So now I can rescale these wands because I have a theory. In the case of the
01:11:20 - Buddhist monks, you have to say, I'm going to meditate the hell out of my wands
01:11:24 - by staring at this wall for a long time.
01:11:27 - I have to reduce the wands that have to disappear.
01:11:29 - While an alternative might be that you say, look, I have to understand myself
01:11:33 - why I have these wands and how I can regulate them without damaging others and myself.
01:11:38 - I'm not necessarily claiming this is a route to success, but I'm just giving
01:11:42 - a hypothetical scenario of how a more rational approach could be deployed to this.
01:11:47 - Yeah yeah but now the one thing i was
01:11:49 - wondering maybe one thing that that you could
01:11:52 - do to this i mean we discuss often how how are
01:11:55 - we going to coexist with technology right right would
01:11:58 - you see technology as as assisting you
01:12:00 - in developing these kinds of math and cognitive theories of of your own behavior
01:12:05 - um you mean as a scientist or or personally well both um well yeah i think we
01:12:13 - we export a lot of our cognition into these devices that we have and they're hugely powerful.
01:12:19 - I would, I mean, I would personally hope that with the advances in AI we are
01:12:24 - now making that machines will take on some of the load of trying to develop theories,
01:12:30 - you know, and I think we can, there perhaps hasn't been enough focus on.
01:12:37 - Sort of automating scientific theory development. I know it's a, there was a,
01:12:43 - a machine a few years ago that rediscovered Newton's laws. So it's maybe several
01:12:47 - hundred years behind where it needs to be to help us now.
01:12:51 - But I do think that we already gain a lot from having all these tools and machines
01:12:57 - and also databases that support our science.
01:13:02 - And I think the role of the scientist is changing as well,
01:13:06 - because we no longer need to remember everything
01:13:10 - thing we've read and have it in our heads we just can access it so
01:13:12 - for somebody like me who's kind of big picture it it's
01:13:15 - great you know i can i i don't have
01:13:18 - to remember the details but i can go and find them when i need them so
01:13:22 - uh i do think that our cognitive capacity to do science is expanded by these
01:13:27 - tools and maybe there is some exciting develops in the future where ais will
01:13:31 - be helping us identify new paths to take but but are you then do you see that develop
01:13:39 - along these kinds of paranoid boss rom scenarios like ai's
01:13:43 - will have a zillion different ways to cheat us out of reality and take over
01:13:47 - or or do you think that that our coexistence these ai's will take a different
01:13:54 - form um i mean i'm kind of quite optimistic about that i mean i uh.
01:14:01 - I think superintelligence will happen.
01:14:04 - I mean, I think there already is AI superintelligence in lots of domains.
01:14:08 - You know, sort of chess playing is one where AIs are much better than us.
01:14:12 - Already, stock trading is another where they do 90% of the stock trades.
01:14:18 - Um so uh there are
01:14:20 - a few domains of human behavior where we're still ahead
01:14:24 - of ais and will be for some time a scientific
01:14:27 - discovery i think is is one that probably be
01:14:30 - okay for a while but uh i do think that in
01:14:33 - these different areas ais are going to advance uh
01:14:37 - i the notion super intelligence is
01:14:39 - i think you you also agree this is is is it's
01:14:43 - not a good idea because i think there are multiple uh kinds of
01:14:46 - intelligence and of course they all correlate so uh
01:14:49 - iq is a good measure across the board
01:14:52 - but it doesn't mean that all of the different parts of your
01:14:54 - intelligence are the same and operate in the same way uh so
01:14:57 - and one of the goals i guess of what
01:15:00 - we're trying to do is to work out what those systems and
01:15:03 - components are um and then build
01:15:06 - a theory of mind based on those i think eventually we will
01:15:09 - be able to build machines that can reason about
01:15:12 - themselves which is one of the defining aspects of
01:15:16 - ourselves um my own view and i think
01:15:19 - if you watch a lot of science fiction movies as i do.
01:15:21 - Uh the the bad ais and the bad robots are
01:15:25 - always the ones that don't understand themselves and
01:15:28 - they have some some mission uh perhaps one
01:15:31 - that's been programmed in by some misguided human uh and
01:15:35 - i think humanity will still be its own biggest uh
01:15:38 - risk even in these coming
01:15:41 - days of ai that people will misuse ai but
01:15:45 - i'm more optimistic that if we build ai's that can
01:15:48 - reason about themselves as ai's they will actually be
01:15:51 - useful and helpful and not be motivated to
01:15:55 - take over and destroy everything else but i do think it's
01:15:57 - right that we that we're thinking and talking about this so of course
01:16:00 - but do you feel that these kinds of alarmist messages
01:16:04 - going around the internet right now
01:16:07 - the social networks and all these declarations being signed
01:16:10 - by scientists like oh stop ai we cannot weaponize ai um we have to this will
01:16:17 - be the most dangerous things we invent right so do you think we should we should
01:16:21 - be jumping on that train as well and sign all these declarations well i did
01:16:26 - sign uh one or two of them so um.
01:16:31 - I think we do have a responsibility as scientists to think about even sort of
01:16:36 - these low likelihood, but very high risk scenarios.
01:16:42 - So, you know, sort of a super intelligence, which would be very anti-human is,
01:16:48 - I think, theoretically possible.
01:16:50 - And we have to think about what are the paths that could lead to that and try
01:16:55 - and make sure that those can't happen.
01:16:58 - So I think it's right that we're already thinking about that now.
01:17:01 - Um, and, but there is the, there are quite a few people who are stoking paranoia,
01:17:08 - which I don't think is necessarily helping.
01:17:11 - Um, and there's also a risk that we, you know, throw out the good with the bad,
01:17:16 - you know, that we, that we decide not to have, uh, AIs and robots that are really
01:17:20 - going to improve the human condition.
01:17:22 - Yeah, but don't you think that it's a bit of a propaganda exercise?
01:17:26 - Because first, we know what goes on in the field, and we know that there's still
01:17:31 - many obstacles before we reach general AI.
01:17:34 - So for scientists to sign these declarations, to me, is also testifying to an
01:17:41 - overstatement of the capabilities of the field.
01:17:43 - Like it's also a little bit hyping a
01:17:46 - field that actually doesn't have the capabilities that it claims to have.
01:17:50 - For instance, take the DARPA robot challenge from 2015,
01:17:55 - where the benchmark was essentially Fukushima nuclear power plant disaster and
01:18:01 - have robots, humanoid robots that can operate in these environments. It was one big disaster.
01:18:06 - And these robots are still remote controlled. In the meantime,
01:18:09 - we have these huge propagandistic statements about, oh, AI is going to take over.
01:18:14 - Well, it's still far removed from that. And then secondly, what worries me is
01:18:18 - that this is driven very strongly by people with business interests.
01:18:22 - And if, for instance, if I'm making money by selling cars that have autopilots…,
01:18:29 - Then indeed, just for my marketing purposes, I would like to give my customers
01:18:34 - the impression or the illusion that I care about their well-being and I care
01:18:38 - about ethical standards.
01:18:40 - So by sort of fanning the flames of, oh, let's be concerned about AI.
01:18:45 - I'm creating trust with my customer so they buy my cars in which they can kill
01:18:50 - themselves if their autopilot is facing an unpredictable situation,
01:18:54 - which also has happened.
01:18:55 - So there's also a hoax behind it, which we should be very careful with.
01:19:01 - And don't forget, as you also said yourself, the biggest threat to humans are humans.
01:19:07 - And before the AI will take over, we might already have killed ourselves.
01:19:12 - So I think AI should be much more worried about the real problems we're facing.
01:19:16 - And I can tell you, if you go to applied domains where we need more science,
01:19:21 - education, healthcare, and so on, we are still far removed from having effective systems.
01:19:27 - And I wish that we would put more energy in that direction as opposed to running
01:19:32 - around like headless chickens who lost their intelligence to be worried about
01:19:37 - artificial intelligence.
01:19:38 - Because on top of that, there's this very strange and naive belief that there
01:19:43 - will be this discrete moment in time where it's all going to happen.
01:19:46 - And again, we always tune our standards, our baseline to the reality we find ourselves in.
01:19:55 - So we seem to forget that we are, as you said, already co-evolving with our technology.
01:19:59 - So it is a very gradual process where we change as the technology changes.
01:20:04 - So this discrete point of transition, like Skynet woke up and now it's conscious,
01:20:08 - because that's essentially the scenario they have in mind. It's just Terminator, you know.
01:20:13 - So these guys don't even have any imagination.
01:20:17 - It's just not what we see happening around us. Yeah, I mean, I think you're right.
01:20:21 - I think there are people who are using this as a smokescreen,
01:20:25 - you know, look over there while I do this, and hopefully you won't notice.
01:20:29 - I think there's another bunch of people riding this anti-AI bandwagon, and they are,
01:20:36 - I think, people who are developing this older version or defending this older
01:20:40 - version of humanism that see actually AI robotics as a threat to their whole
01:20:46 - conception of the human condition.
01:20:48 - Because if we were to build a robot that could walk and talk and describe itself
01:20:52 - as having internal states, then that would really challenge a lot of preconceptions
01:20:58 - that, uh, people have about themselves.
01:21:01 - Sure. Um, so I think, I think people find that threatening and I think,
01:21:05 - uh, you know, robotics in particular has, uh.
01:21:09 - You know, it's almost taken over from the zombie or the psychopath as,
01:21:16 - you know, the sort of the bad guy in the movie.
01:21:19 - And there's a reason for that, actually, because robots are much better psychopaths
01:21:23 - than a human psychopath.
01:21:24 - Robots are even more devoid of emotion or compassion, potentially.
01:21:30 - So it fits with our cultural imagination to see these things as dangerous.
01:21:35 - So it's an easy story for the media to sell.
01:21:38 - Of course. I mean, for all of these reasons, yes, it's in the headlines a lot,
01:21:41 - perhaps far more than it needs to be.
01:21:44 - And what would be better, I guess, is if we can have, and I think it's partly
01:21:49 - the fault of the community not to do more to communicate about the science and the research.
01:21:55 - I mean, so like you say, sometimes people look behind the curtain and,
01:22:00 - you know, they see the Wizard of Oz there controlling his DARPA robot and the
01:22:05 - DARPA robot falls over and it becomes evident that maybe things aren't as advanced as they might be.
01:22:11 - On the other hand, if you look at the previous DARPA road challenge,
01:22:14 - the initial one, the cars got a couple of miles and then crashed.
01:22:19 - But then a few years later, they
01:22:21 - were driving hundreds of miles across the Arizona desert by themselves.
01:22:25 - So you can make rapid progress. So it's not an exponential by any means,
01:22:31 - but there are some sort of sharp lifts which take you up to another level.
01:22:36 - And perhaps we have just had one. I mean, I think people are extrapolating from
01:22:40 - the recent advances in machine learning to say that we're up on the exponential.
01:22:45 - I think we're not. We've jumped up a level.
01:22:49 - And we're now maybe flatline again for a while. And then there might be another sharp jump. Okay.
01:22:54 - But don't forget with the self-driving challenge where the Stanford robot Stanley
01:22:59 - won, the team that won was the team with the most planners of GPS waypoints for the car to reach.
01:23:07 - And the only sort of adaptive element that made the difference between Stanley
01:23:13 - and Carnegie Mellon was that Stanley had a little reinforcement learning system
01:23:18 - that would accelerate when there were no obstacles on the road.
01:23:20 - But the whole planning was done by humans prior to the start.
01:23:25 - So again, much less impressive than you think it is, right? So we should be
01:23:29 - very careful how we interpret those results.
01:23:31 - But look, so you're in this human brain project as well, which,
01:23:36 - okay, you are an exception, but it's largely a waste of money, in my opinion.
01:23:40 - And we should do something about that because we cannot afford in the face of
01:23:45 - limited resources to throw it away like that.
01:23:48 - But imagine I give you 10 million, let's give you euros because the pound has no value anymore. So...
01:23:57 - If I give you 10 million euros for a 10-year project, what would you dedicate that money to?
01:24:04 - In one word, what's the key concept or question you would pursue with that?
01:24:10 - So this is just a normal-sized project. What about a flagship?
01:24:14 - A billion flagships. I'll give you a billion. No problem. Okay.
01:24:17 - I'll give you free coffee as well. I think if I had a billion euros,
01:24:21 - I would be setting out to build human AI systems that can really think about
01:24:30 - the challenges that the world faces.
01:24:32 - Um i think if you look at the
01:24:35 - ipcc and the way
01:24:39 - that that the international panel for climate
01:24:42 - change and the way that groups of scientists have
01:24:46 - and people also looking at the economic and
01:24:49 - social impacts of climate change these people have come along
01:24:52 - and they've worked with very sophisticated computer models and
01:24:55 - over a period of time they've refined these models and
01:24:58 - they've refined their thinking to the point that we can now with pretty
01:25:02 - high confidence say that the result of human
01:25:05 - activity generating co2 is going to cause
01:25:08 - levels of temperature rise which are going to make life bad for us and as a
01:25:13 - result of that i mean you say you mentioned the power to explain predict and
01:25:18 - control i think that's a really nice example of it because the first of all
01:25:23 - they they had to build systems that could explain data.
01:25:26 - And then they had to build systems that could predict what happened in the future.
01:25:30 - They are now being able to show that they can predict, you know,
01:25:34 - from predictions they made 10 years ago, they can point back and say,
01:25:37 - this is, uh, no, it's at least as bad as we said it was going to be possibly worse.
01:25:42 - Uh, and then they've also got a program of control, which they are proposing.
01:25:47 - It's, it's not fully fledged.
01:25:49 - Uh, and if you look at the paris agreement it looks
01:25:52 - like the whole world with one or two notable exceptions are
01:25:56 - starting to get behind this and so that was uh
01:26:00 - i wouldn't say it's a science driven thing but it
01:26:02 - was driven by a community that really cared about
01:26:06 - the science of the climate and really thought that if we did this in a global
01:26:10 - way we could make it in a difference and i think uh with a billion euros we
01:26:16 - could do the same for some of the other big problems that we face And I think
01:26:20 - the first one we might want to look at is wealth inequality.
01:26:24 - So you would want to build these models to try and understand human social behavior,
01:26:29 - human economy, you know, as we said, bring the psychology and the economics together. Right.
01:26:35 - And bring forth some proposals about how we can reorganize.
01:26:40 - World markets you know sort of maybe rethink the structure
01:26:43 - of capitalism in a way that won't favor us
01:26:46 - going to this extreme because the other alternative is that
01:26:49 - that we're going to end up as a society collapsing in
01:26:52 - some way uh because you know the
01:26:55 - extremes of wealth inequality you can only see it some cities uh
01:26:59 - some people living in gated communities uh with
01:27:02 - more wealth than they can imagine what to do with you know booking trips to
01:27:05 - the moon and things and then uh you know over
01:27:08 - the world people living in extreme poverty so that ought to
01:27:11 - be a priority and i think uh what the
01:27:14 - climate change model has showed is that
01:27:18 - you can really use computers for good to predict
01:27:20 - and then you know with using these
01:27:24 - computers under control this is this is so i
01:27:27 - completely agree with you but and what you also see here is that
01:27:30 - why does the climate change project actually
01:27:34 - work because it gave the tools to
01:27:37 - humanity to change its metacognition about its
01:27:39 - own state yes right so this is
01:27:42 - very powerful and therefore indeed i agree with you the computer models are
01:27:46 - not necessarily immediately there to interfere with that reality but they should
01:27:50 - help us as humans to develop a metacognitive state to understand our condition
01:27:54 - so that we can change it for the better indeed that's one of the things i think
01:27:59 - that i got from nick bostrom's book about.
01:28:02 - Super intelligence because he talks about different kinds
01:28:05 - of super intelligence that you can build and one
01:28:08 - of them is the oracle and the oracle you just go
01:28:11 - along to and like the oracle of ancient greece you can
01:28:13 - ask any question gives you the answer uh and so
01:28:17 - uh essentially the climate change scientists have
01:28:20 - built an oracle they say well what's the temperature going to be like in 10
01:28:23 - years it gives you an answer um you know within some bounds uh and then uh he
01:28:29 - also talks about well we could build genies and genies would be oracle but they'd
01:28:33 - also have power to change stuff and you know genies are potentially much more
01:28:37 - dangerous so if we build these ai oracles.
01:28:41 - Then we still have the control as humanity decide what's the best advice you
01:28:47 - can give us in this situation the oracle was said if you did this then there
01:28:51 - might be less wealth inequality and then we can then choose or not choose to
01:28:55 - do that right now i think over time perhaps we probably will do what the AI says,
01:29:00 - because it will turn out that those predictions are good. Well,
01:29:03 - we will shape it. We will also shape it. Exactly.
01:29:05 - And it will be a human machine system. So there will never, I don't think ever
01:29:09 - in these climate change systems are they just running an algorithm.
01:29:13 - It's people interpreting data, putting it into the algorithm,
01:29:17 - tweaking the algorithm, working out, improving it all the time.
01:29:20 - So it's a human machine system which understands an important aspect of our world.
01:29:26 - So we should build more of those. Sure. Actually, in 2005, as one of our art
01:29:31 - projects, we presented the Synthetic Oracle.
01:29:33 - And the whole idea was indeed that you engage with, let's say,
01:29:36 - an interactive sound and light composition that depends on your actions to create,
01:29:42 - let's say, a very implicit kind of immersive experience that helps you to meditate on your state.
01:29:49 - Okay. And actually, it was very
01:29:51 - effective. Of course, it didn't tell you anything about climate change.
01:29:53 - But I think this – so I think we have a principle here, right?
01:29:57 - It's all about also changing our metacognition so we can understand our own
01:30:00 - situation, that we can step in and make a difference, as was done with the Paris Climate Accord.
01:30:07 - But now we have to scale it up because our problems are much broader than that.
01:30:11 - So, okay, Tony, look, I'll get back to you once I have the billion,
01:30:15 - but I think we have a good plan for the future.
01:30:20 - I forgot to sort of try to rip some holes in your theories about the brain,
01:30:26 - but I'll do that next time.
01:30:27 - Well, thank you very much for this conversation. Thank you.
01:30:33 - The CSN Podcast was produced by the Convergent Science Network of Biometrics
01:30:39 - and Biohybrid Systems, a project funded by the European Sevens Research Framework Program.
01:30:47 - For more interviews, recorded lectures, or upcoming conferences in the field
01:30:52 - of biometrics and biohybrid systems, go to csnnetwork.eu.
01:31:00 - Music.

CSN Podcasts

Paul Verschure & Tony Prescott on synthetic psychology and robot models

Season 2018

Description

CSN Podcasts

Episodes

Ton Coolen on immune networks and neural networks

Neil Burgess on boundary vector cells and place cells

Giacomo Rizzolatti on mirror neurons and action understanding

Francesca Cacucci on hippocampus development and grid cells

Bechir Jarraya & Lynn Uhrig on anesthesia and consciousness

Timestamp

Be the first to leave a comment

Leave a commentCancel

Legal

Follow us on

Paul Verschure & Tony Prescott on synthetic psychology and robot models

Description

Episodes

Timestamp

Be the first to leave a comment

Leave a commentCancel

Tags

Legal

Follow us on

Login to enjoy full advantages

Please login or subscribe to continue.

Go Premium!

Enjoy the full advantage of the premium access.

Stop following

Cancel subscription