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

José Halloy on collective behavior and bio-hybrid robots

Episode 2 15.03.2018

PLAY EPISODE

Season 2018

0 Followers 10 Episodes

Description

Can you infiltrate a cockroach colony with robots and steer its collective decisions? Physicist José Halloy explains how simple mathematical models from statistical physics capture the self-organizing behavior of animal groups , and how biomimetic robots that smell like cockroaches can be used to test and manipulate these models from the inside. Subscribe for more from the Convergent Science Network podcast series. José Halloy joins Paul Verschure and Tony Prescott at the BCBT summer school to describe his work on collective behavior in animal-robot hybrid societies. Drawing on dynamical systems theory, Halloy and colleagues have shown that cockroach aggregation under shelters can be modeled with a small set of differential equations capturing positive feedback from social attraction and negative feedback from environmental saturation. The key insight is that even populations of identical individuals with no hierarchy can produce consensus decisions through these simple nonlinear mechanisms , a principle found at every level of biological organization from gene regulation to neural circuits to social groups. The discussion focuses on a landmark experiment in which small robots, coated with cockroach pheromones to pass olfactory recognition, were introduced into cockroach colonies. Despite having no resemblance in shape and running on a finite state machine rather than a neural controller, the robots were accepted as group members and could influence collective shelter choice. By programming the robots to prefer a different shelter, the researchers demonstrated that a minimal number of artificial agents can shift the consensus of the entire group , a nonlinear effect predicted by the mathematical model. The conversation explores the limits of this approach: why it works for cockroaches (which rely primarily on olfactory recognition) but is far harder with fish or vertebrates (which are more multimodal), what the framework reveals about the relationship between individual cognition and collective behavior, and whether the dynamical systems approach from physics can scale to more complex species. Halloy argues that while these models capture specific mechanisms in specific experiments rather than the full complexity of an animal, the methodology of positive and negative feedback networks producing emergent behavior is a universal lesson across biological scales. Key topics include collegial decision-making without hierarchy, the insect Turing test, why robots need not replicate neural architecture to reproduce behavior, the role of internal states like hunger and fear, and what ant colony optimization in computer science owes to biological models. Part of the Convergent Science Network podcast series from the BCBT Summer School.

Tagged as:
bio-hybrid robots Cockroach collective behavior Negative Feedback Olfactory Recognition Robots

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.
00:00:08 - Leading researchers in the domain of neuroscience, brain theory and technology
00:00:13 - are interviewed by Paul Verschure and Tony Prescott.
00:00:20 - This is Paul Verschure with the Convergent Science Network podcast together
00:00:24 - with my colleague Tony Prescott.
00:00:27 - And we have here as our guest José Aloy. Hello, José.
00:00:32 - José was a speaker as well at our BCBT Summer School 2016.
00:00:40 - So, José, you came from physics and now actually what you're studying is collective
00:00:45 - behavior of animals and also in hybrid sort of animal-robot societies.
00:00:51 - So, how did you go from physics into this domain of the ethology or neuroethology
00:00:56 - of behavior? Well, it's because in physics, people are interested in collective systems anyway.
00:01:04 - Statistical physics deals with collective systems. Then you have dynamical systems
00:01:09 - that can deal with collective systems.
00:01:11 - And so it came to the mind of people in that field that you can use the tools
00:01:17 - of physics to build up mathematical models to describe experiments done with animals.
00:01:24 - And for instance, I started to work on that with Jean-Louis de Nouveau in Brussels,
00:01:29 - and he was very successful in showing
00:01:33 - that he was capable of building
00:01:36 - experiments that led to mathematical models based on differential equation or
00:01:41 - stochastic equation that gave extremely good results to describe what uns societies
00:01:47 - and other collective behavior in animals were happening.
00:01:53 - So from there, Jean-Louis had the idea that because all those results had a
00:02:02 - high impact in the new field of collective robotics.
00:02:07 - People in collective robotics were trying to mimic also, building also a collection
00:02:11 - of agents that together can perform a task.
00:02:15 - They were getting inspiration from biology and it came, a very simple idea.
00:02:20 - If we have animals on one side and you have a robot, biomimetic robot on the
00:02:26 - other side mimicking their behavior, if you succeed in mixing them,
00:02:31 - you get a new group, a bio-hybrid group of robots and animals.
00:02:35 - And because you're able to tune the behavior of the robot, you're capable of
00:02:40 - tuning the behavior of the whole group.
00:02:42 - Because essentially what the mathematical models were showing is that those
00:02:47 - collective behaviors can be described by emergent properties of the system.
00:02:52 - Which means that there is no clear hierarchy. Even if the hierarchy is completely
00:02:56 - flat, every individual is equal, you still get emergent behavior.
00:03:01 - And then if you inject a few individuals in the system, you can tune the whole system.
00:03:08 - And it's a field that is now quite common in physics.
00:03:13 - More generally, it's the field of active matters, where now physics is trying
00:03:18 - to build from matter systems that reproduce some of the collective behavior
00:03:24 - observed in higher animals, like ants, bees,
00:03:27 - or birds, or fish, at the level of matter.
00:03:30 - So there has always been an interplay between statistical physics and animal collective behavior.
00:03:36 - Right. But in some sense, you're
00:03:39 - also then proposing that you can build robots or use robots as a probe.
00:03:44 - Into an animal population to understand what the principles are by which such a population operates.
00:03:49 - Yeah, it's a tool also. So you can test if your model works when it is embodied.
00:03:56 - Because of course you can show that simulations reproduce results that are similar to the animals.
00:04:03 - But then it's even better if you can prove that when you embody that model in
00:04:07 - a completely different agent, which is a robot, that has nothing to do with the animal, Still,
00:04:15 - it works, which proves that your model is capturing something true from collective behavior.
00:04:21 - And then, of course, if you try to mix those robots inside the animal group,
00:04:26 - you're testing hypotheses.
00:04:28 - And you're testing modulating parameters. It corresponds to building an experiment
00:04:33 - where you're able to modulate a parameter of the collective,
00:04:37 - which is not very easy to do experimentally.
00:04:40 - You can imagine other systems. For instance, people have been trying to teach
00:04:44 - animals something, to inform them.
00:04:47 - Then you re-inject them in a group of naive animals.
00:04:52 - And because they already know, have learned the solution, they are influencing the whole group.
00:04:57 - Or you can even imagine doing genetic mutations and some of the individuals
00:05:01 - have different capabilities, etc.
00:05:04 - But learning on some animals is very difficult or impossible.
00:05:08 - And genetic mutation, the link between genetic mutation and higher level behavior is very weak.
00:05:15 - I mean, it's not very easy to do, especially with animals like ants, bees or fish.
00:05:20 - It's complicated. So the robot might be an interesting tool,
00:05:23 - a kind of microscope to probe collective behavior at the animal level.
00:05:30 - So you're proposing that we can model the behavior of a group of animals using
00:05:35 - a small set of differential equations,
00:05:38 - and that's quite a strong claim because we're obviously going to ignore most
00:05:46 - of the richness of those animals and what they're like and how they're made up,
00:05:51 - and we're going to look at some very high-level property which we can describe mathematically.
00:05:55 - Why do physicists think that this is going to succeed?
00:05:59 - Okay, so it has been proved that it works on some experimental physical cases,
00:06:04 - like for ants, for bees, for fish, for birds.
00:06:08 - Those models have been capturing the essential mechanism of the group,
00:06:15 - how it works. But of course, and it's the philosophy of a physical approach
00:06:21 - of systems, you want a minimal model.
00:06:23 - You want the most simple model capable of capturing what you're observing in an experiment, right?
00:06:30 - But of course, it doesn't mean that these mathematical models are a model of the animal.
00:06:37 - They are not. They're just capturing a specific mechanism in a specific experiment.
00:06:44 - And so there are strong limitations to that.
00:06:47 - I mean, for the last 25 or 30 years, there have been a lot of success in animal
00:06:52 - collective behavior, but still, when you look at the global result,
00:06:57 - there are not so many models that give interesting results.
00:07:01 - But still, the idea that you can have simple mechanism, simple mechanism that
00:07:08 - produce interesting collective behavior without taking into account the whole
00:07:13 - complexity of the animal.
00:07:15 - Is a message. You see what I mean? So you described an experiment with cockroaches
00:07:21 - where you have two containers and you put the cockroaches, I think,
00:07:27 - randomly into the containers and then you see how they,
00:07:30 - choose to go into one bin or another and you use differential equations to show
00:07:35 - how that system might evolve over time.
00:07:39 - And I guess one of the things that can be confusing about this.
00:07:43 - Is your model a description of the system,
00:07:47 - or is it not capturing the causality in the system in the way that,
00:07:51 - say, a mechanistic model of individual cockroaches and how they make decisions
00:07:56 - would capture causality?
00:07:58 - I think it's both. It gives you a description of the system,
00:08:03 - and it gives you also the cause of the solution that emerged in the system.
00:08:08 - It gives you the mechanism that produce this emergent behavior at the population level.
00:08:13 - But still, you can go down at the individual level and find yet other causes
00:08:21 - that lead to this solution that emerge in the system.
00:08:26 - And what those models have shown, if you think in terms of dynamical system.
00:08:32 - You have a network of feedback regulation in the system based on positive feedback and negative feedback.
00:08:39 - And nonlinear effects, right? And if you mix that, you get emergence of interesting properties.
00:08:46 - But that thing, you find it at all levels of living systems.
00:08:51 - You find it at the genetic level, you find it at the metabolic level,
00:08:55 - you find it at the physiological level, and you find it at the population level.
00:09:00 - Those abstract concepts of network of regulatory feedback and nonlinear effect
00:09:06 - leading to the emergence of interesting patterns has been a lesson of the domain.
00:09:13 - But it seems to me that there are different aspects to this, right?
00:09:17 - I think on the one hand, you're saying, I don't need to model the animal as such.
00:09:22 - I just want to have a model that allows me to do a meaningful perturbation of
00:09:27 - that animal group or individual.
00:09:30 - And that then also brings this whole issue, okay, but what would be then a sufficient approximation.
00:09:37 - Of a meaningful intervention in such a group. And also in that context,
00:09:41 - you actually spent quite some time discussing the Turing test, right?
00:09:44 - Because in some sense, in the cockroach experiment, as Tony mentioned,
00:09:48 - you have this problem of imitation.
00:09:50 - You want to sufficiently imitate an individual cockroach to have a meaningful
00:09:55 - perturbation of the cockroach group, right?
00:09:58 - So would you, if you take the cockroach experiment as an example,
00:10:01 - would you really see that as an insect Turing test you're performing or it's different?
00:10:07 - Well, it's somehow, I think if
00:10:10 - you reduce the Turing test to a social election, I mean, it's like that.
00:10:17 - You're interacting with an artificial system and at a certain point you agree
00:10:22 - to interact with that system because it's becoming interesting enough for you to interact with.
00:10:28 - So animals are not at the symbolic level, they don't exchange,
00:10:31 - they don't, cockroaches or many, nearly all animals, maybe primate do that,
00:10:37 - but all animals don't really communicate symbolically, like they don't have
00:10:41 - the language, the same kind of language humans have.
00:10:45 - So their interaction are based on other modalities.
00:10:51 - And then, yes, if you prove that your robot is accepted by the group,
00:10:57 - And if the animals take into account that robot as a member of the group,
00:11:03 - whatever that means, I'm not saying that the cockroaches believe or think or
00:11:07 - whatever they do that the robot is a cockroach.
00:11:12 - We cannot answer that question, but they do take into account the presence of
00:11:17 - the robot as another member of the group. And because they do that.
00:11:22 - The robot is capable of influencing the whole group decision-making process.
00:11:27 - But what is interesting in the cockroach experiment, it's because we designed
00:11:31 - a biomimetic, a very simple, but yet biomimetic at the behavioral level.
00:11:37 - It was biomimetic at the behavioral level, not the shape, not the material and stuff like that.
00:11:42 - A social robot that was programmed to behave like a cockroach in the same experiment,
00:11:48 - the robot were taking also into account what the cockroaches were doing.
00:11:51 - And so they were also influenced by the cockroaches.
00:11:55 - And sometimes the cockroaches were driving the robots as a group to a place
00:12:00 - they wouldn't have chosen alone because they were programmed to prefer,
00:12:04 - to have another preference, for instance.
00:12:07 - You say that the robot is biomimetic in terms of its behavior. Yeah.
00:12:14 - But the control system that drives the robot, I think, is a finite state machine.
00:12:19 - Yeah. And that's not meant to emulate a cockroach nervous system or anything like this.
00:12:26 - So how do you get a cockroach-like robot that has behavior like a cockroach?
00:12:33 - Yeah, that's exactly the point.
00:12:35 - In fact, you don't have to copy all the levels of the living system.
00:12:40 - In fact, it's based on the idea that also has been discussed in physics.
00:12:47 - In the famous paper by Paul Anderson, Moore is different, that he published.
00:12:52 - He was a physicist of solid-state system.
00:12:56 - And he knows that you have emergence of properties at a certain level.
00:13:00 - Of course, they depend on the lower level. The properties of a solid depends on the atoms.
00:13:08 - It's obvious. But you can have a model, a level of description of the solid
00:13:12 - that is good enough to explain what the solid is about.
00:13:17 - So it means you don't need the nervous system of the cockroach necessarily.
00:13:22 - You don't necessarily need the nervous system of the cockroach to get the behavior.
00:13:27 - It can be based on another lower mechanism, like the finite state machine.
00:13:32 - Machine, but that doesn't exclude the fact that you would like to have also
00:13:37 - a biomimetic mechanism at the neural level. That's another question also.
00:13:42 - It's allowed, I mean. You could do that. You could add an extra layer of mimetism
00:13:47 - that you want to understand if you want to pile.
00:13:53 - Neural models, neural net models, biological relevant neural net models with
00:13:59 - individual behavior, with social behavior. That's another question.
00:14:02 - But you don't necessarily need to include all levels to get the result,
00:14:07 - because each level has its own properties somehow. You see what I mean? Yeah.
00:14:13 - So you're tweaking the robot control system to give it cockroach-like behavior,
00:14:18 - which will then generate results as though there was another cockroach in the group.
00:14:25 - I mean, aren't you worried there are a lot of degrees of freedom for you to tweak? And so...
00:14:32 - Yes, there were many, in fact. And what was interesting is that the model.
00:14:38 - This little set, very simple set of differential equations, were predicting
00:14:42 - many different possibilities to control the group.
00:14:45 - And we only tested experiment in one way of controlling the group.
00:14:50 - But we know from simulation and from solving the equation there are many other ways.
00:14:55 - For instance, you can build social robots that are social among themselves,
00:14:59 - but they do not take care of what the cockroaches are doing.
00:15:02 - You get a different set of solutions, so a different set of modulation.
00:15:06 - You can build robots that are completely non-social. They
00:15:10 - do not take care either of the other robots or the
00:15:13 - cockroaches you get different results and so
00:15:17 - on so you can or we chose to to
00:15:19 - to show the the case where the robot were the
00:15:23 - same behaving in the same way as the cockroaches but the model shows that you
00:15:28 - have many other different uh social capability social capability social interaction
00:15:33 - leading to different solution and of course the intensity of the relay of the
00:15:38 - interaction is also you can also modulate it.
00:15:41 - So you can add a parameter in the model that modulates the probability to respond
00:15:46 - to the presence of the other.
00:15:48 - In the case of cockroach, let's say this probability is one,
00:15:51 - because a cockroach considers another cockroach as a cockroach, so it's one.
00:15:55 - But you can, for the robot, you can reduce it. We have put one also in the experiment.
00:16:00 - We did. But in the simulation, you can put less than one.
00:16:03 - And then you can see what level of interaction, what intensity of interaction
00:16:07 - you need to still get an effect.
00:16:10 - It doesn't need necessarily to be one. Then you can also modulate the number
00:16:15 - of robots you're adding to the system.
00:16:18 - Our idea was to show that you need only a minimal number of robots to influence
00:16:25 - the system because it's a non-linear effect, which is interesting.
00:16:29 - But, of course, you can increase the number of robots in the system and you
00:16:36 - will get different kind of solution or modulation.
00:16:41 - But now, it was the case that you had to make the robot smell like a cockroach
00:16:46 - to be accepted, while in some sense, the shape didn't really matter, right?
00:16:52 - So, what then would be the minimal robot that you think would be plausible for
00:16:58 - a cockroach colony? So that's exactly the point.
00:17:04 - In fact, it's the crucial question, is how to make the animal respond and accept the robot.
00:17:12 - And it's very difficult because animals are multimodal. They use all their senses
00:17:17 - to perceive their environment and the others.
00:17:19 - And we were lucky enough, and it's well known, that insects,
00:17:22 - and in particular cockroaches, they
00:17:25 - base their recognition on tactile and olfactory cues more than vision.
00:17:31 - So the shape does not matter. What matters for them is that you smell like a
00:17:37 - member of the group, and that's enough.
00:17:39 - And of course, any mobile robot that is capable of moving around in the system
00:17:45 - and detecting the presence of shelter in terms of a different intensity of light
00:17:51 - below or outside of the shelter would be enough.
00:17:55 - Any shape. And of course, you need a size that is compatible with the physics.
00:18:00 - I mean, you're not going to put a robot 10 times larger than the cockroaches
00:18:04 - because it's a completely different physical world.
00:18:06 - But the shape in that case doesn't matter. But then we've tried with chicken,
00:18:10 - we're trying with fish, and it's a completely different story.
00:18:14 - Because those vertebrates are a bit more tricky. They are much more multimodal.
00:18:18 - So they do care about vision. They do care about shape.
00:18:22 - They also care about smell. They also care about behavior.
00:18:26 - So that's why I think, for the moment, trying to interact with fish is pretty difficult.
00:18:32 - There are a few groups that have been trying to do that.
00:18:35 - And it's much more difficult to get results with the cockroaches.
00:18:40 - Because I think fish are a much more multimodal animal, and they do take care
00:18:47 - of many different inputs from the sensory system.
00:18:51 - That's why shape matters, colors matters, speed matters, movements matters,
00:18:56 - much more than with the cockroaches.
00:18:58 - So basically, I think you may say we were lucky, but I don't think.
00:19:04 - I think Jean-Louis was clever enough to choose those insects because he has
00:19:11 - the intuition that it's going to work because it's an olfactory recognition essentially.
00:19:18 - But then in the cockroach experiment where you looked basically at the ability
00:19:23 - of groups of cockroaches or a hybrid group cockroach with robots,
00:19:27 - how they would disperse in an environment or aggregate in an environment depending on conditions.
00:19:33 - So they would aggregate under these shelters, right?
00:19:35 - And the notion that was tested there was called collegial decision making, right?
00:19:43 - So what does it really mean, collegial decision making in this context?
00:19:46 - There is a consensus emerging in the system. And this consensus emerges from
00:19:51 - individuals that are considered as perfect clone of each other.
00:19:55 - What the model is saying that even if you have a population of exactly identical
00:20:02 - individuals, you get the collective decision mechanism and the clever way of doing groups.
00:20:10 - I'm not saying that the cockroach population we're using were clonal individuals.
00:20:15 - No, there is a lot of inter-individual differences.
00:20:20 - But what the model shows that in completely total absence sense of any hierarchy,
00:20:26 - any difference from the individuals, yet you get a collective system producing
00:20:32 - this consensus on where to gather, which is not such a trivial task.
00:20:38 - Imagine you take a population of about 100 humans without specific structure,
00:20:43 - hierarchy structure in the group, naive, let's say, individuals,
00:20:48 - 100 of them, and then you ask them, okay, your job is to split.
00:20:52 - 50-50% between two rooms. Then you look at the process of how that decision is going to emerge.
00:21:00 - And it will take some time. We take people arguing, people discussing,
00:21:04 - people starting to count, going from one room to the other to check how many
00:21:09 - are you, how many are there on the other side. You can imagine that experiment.
00:21:13 - And it's going to be a complicated mechanism to produce this 50-50% land spread.
00:21:22 - Cockroaches don't do that at all. It's a very simplistic mechanism.
00:21:26 - They just move around. They know they are in a room. Interesting.
00:21:30 - They look around if they have enough bodies and the body, sorry.
00:21:34 - And then if you have this threshold function on the probability to get to go out, that makes a trick.
00:21:41 - So it means in terms of cognitive capabilities, it's very simple, fairly simple.
00:21:46 - What those models are showing, it goes back to your previous question,
00:21:51 - Tony, what those mathematical models are showing is even if you have high cognitive capabilities,
00:21:58 - you don't necessarily need to use them all the time, and in particular,
00:22:02 - to produce that kind of collective behavior.
00:22:06 - For instance, that's the same kind of approach that people who are modeling crowd movement.
00:22:11 - Of course, crowd movement with human beings. Human beings keep their high-level
00:22:16 - capabilities, cognitive capabilities, all the time.
00:22:19 - But when they are moving in a crowd, they're not necessarily using them.
00:22:24 - They are using much simpler mechanisms, and yet you get a structure in the crowd
00:22:30 - that emerges from the people moving.
00:22:34 - So in your collegial decision-making, you have sort of two opposing forces that
00:22:39 - guide the behavior, right?
00:22:40 - On the one hand, they want to seek shelter, and on the other hand, they want to aggregate.
00:22:45 - But that means if aggregation leads to a lack of shelter, then they will switch
00:22:50 - and look for another shelter.
00:22:51 - Yeah. This is essentially what happens, right? Yeah, of course.
00:22:54 - If the shelter is saturated, they cannot enter and they have to look for another place.
00:23:00 - But that means they would always aggregate. gate. This is a key driver.
00:23:04 - Those species of cockroaches are really a gregarious animal.
00:23:10 - They really want to be as a group, to shelter as a group.
00:23:13 - And it's been proved that they feel better and they grow better if they live
00:23:18 - in a social environment as a group than when they are isolated.
00:23:23 - And it's been shown even by biologists that if you tickle them with a plume,
00:23:29 - their physiology is better.
00:23:30 - So there's really animals that feel better as a group. So for them,
00:23:34 - it's important to aggregate.
00:23:38 - But then to describe those experiments, so we have this bifurcation model, if you want, right?
00:23:43 - There's sort of a moment of sort of scurrying around. There's exploration, if you want.
00:23:50 - And then very quickly, the population falls into this distribution of seeking
00:23:55 - shelter below the different shelters that they're offered.
00:23:58 - And then you interpreted that by saying that there's an interaction with positive
00:24:02 - and negative feedback. That there's positive feedback,
00:24:05 - generated by the animal, the internal control of the animal,
00:24:08 - and there's negative feedback coming from the environment.
00:24:11 - So how should I interpret that relative to this experiment exactly?
00:24:15 - Yeah, that's what we find. That's what also has been shown by those mathematical models.
00:24:21 - And it boils down to dynamical system theory that shows that,
00:24:26 - for instance, if a choice can be described by multiple stable steady state existing
00:24:32 - as a solution of the system,
00:24:34 - and those multiple steady states require the presence of a positive feedback in the system.
00:24:42 - But of course, a positive feedback is like a snowball effect.
00:24:46 - It's an amplification of what's going on. Then you need a limiting mechanism,
00:24:52 - otherwise the system is going to explode, to blow up, because these amplification
00:24:57 - mechanisms keep going on, going on.
00:24:59 - So you always find in physical system, in biological system, a limitation.
00:25:06 - But then what is interesting in the models, all the models I know,
00:25:10 - is that usually the positive feedback is implemented somehow at the individual level.
00:25:17 - It's very often a mimetic effect that drives the positive feedback.
00:25:24 - And the negative feedback is given by an environment constraint.
00:25:30 - Not enough room anymore.
00:25:34 - Not enough room in the shelter, then it kills the positive feedback.
00:25:38 - Nobody can enter anymore.
00:25:40 - So that is interesting. And you find that in all, again, at all level of living systems.
00:25:46 - You can describe a metabolic system as a network of positive and negative regulation.
00:25:51 - A genetic system as a network of positive and negative regulation.
00:25:55 - Neural nets, there are those with a positive and a negative feedback, inhibitor, activator.
00:26:02 - So you find that this theory of dynamical system gives you that you can build
00:26:09 - artificial system and that you can see that natural system behave,
00:26:13 - the mechanism under explaining their behavior is this network of positive and negative feedback.
00:26:21 - This dynamical systems approach that you have, so you took it from statistical
00:26:27 - physics, you brought it into animal behavior, specifically in crop roaches.
00:26:32 - And then from there, you want to try and generalize it to other species.
00:26:36 - Is that going to be an easy step? Or as you were saying before,
00:26:39 - other animals are more complicated.
00:26:42 - And does that mean that the approach is not going to scale up?
00:26:45 - So, in fact, there are a few examples, after all, that have been described like that.
00:26:53 - The classical examples are ants, or the cockroaches we've been discussing,
00:26:59 - some models with the bees, many models of schooling and schooling in fish.
00:27:05 - And then you have bird flocks, because those are very large groups of very large number of individuals.
00:27:15 - Where a physical model, a simple model inspired by the physics method,
00:27:21 - does something interesting.
00:27:23 - Then, de Nebourg again has shown that for some specific behavior in primates,
00:27:31 - you can still model that, that way, from some specific behavior.
00:27:37 - That doesn't mean you're describing the whole animal. It's just you're describing
00:27:41 - a mechanism that they use in some specific decision-making process,
00:27:47 - like who chooses the direction to go.
00:27:50 - So you have a group of primates sitting commonly there in their park,
00:27:55 - and then suddenly you trigger the group to start a movement towards some place.
00:28:01 - But then who is deciding where to go?
00:28:04 - And again, if you have a mechanism that is not based on a strict hierarchy,
00:28:09 - the boss is saying, everybody to do right.
00:28:13 - In cabbage and monkeys, it's not what is happening. So it's,
00:28:17 - again, some kind of self-organizing system that is driving the system.
00:28:23 - But it doesn't mean that all collective behavior or any other individual behavior
00:28:30 - could be described that way.
00:28:33 - So I'm not pretending that you can describe any kind of behavior or any kind
00:28:37 - of colleague behavior that way.
00:28:39 - In fact, it's very tedious to build those models because since I'm a student and getting older,
00:28:47 - I've seen people doing that and still there are not that many examples because
00:28:54 - of course animal behavior and collective behavior is much more complicated.
00:29:01 - Than, let's say, gas temperature emergence or something like that.
00:29:06 - People have taken these kinds of models and they've applied them also to phenomena of human perception.
00:29:12 - So, for example, the NECA cube, where you see this three-dimensional or this
00:29:16 - two-dimensional shape, you interpret it in one three-dimensional way and then flip into another.
00:29:22 - And that, I think, is a similar approach. So, in a way, is this technique sort of.
00:29:29 - Too sort of general, that it fits all these different kinds of systems,
00:29:34 - and so it tells us a little bit about them,
00:29:36 - but can it really give us useful insights into these systems and how we might,
00:29:43 - say, sort of interface with animal colonies more effectively or understand their
00:29:49 - behavior more effectively?
00:29:50 - Yeah, that's the research question, the ongoing research question,
00:29:53 - is what can you describe by this approach?
00:29:56 - And again, this approach works at very different level, but let's stay at the collective behavior.
00:30:01 - And then if you understand that, and if the model works for specific cases,
00:30:08 - then what we've learned also is that if you build artificial systems,
00:30:13 - they reproduce the same kind of behavior.
00:30:15 - For instance, what was surprising is that when Jean-Ryder Nabour did his AMT
00:30:20 - model for path selection,
00:30:22 - collective path selection by ants, people in
00:30:26 - computer science and robotics were inspired by that
00:30:29 - and then certainly we saw the the emergence of
00:30:31 - this un-colony optimization field people would
00:30:35 - that took up that model and then they started to use
00:30:37 - it as a heuristic for either network handling or optimization which was completely
00:30:44 - unexpected and they got apparently they use a whole field of that and they get
00:30:48 - interesting result then of course we've shown repetitively that you if you build
00:30:53 - robots that they can and reproduce those kind of behavior.
00:30:57 - But we are yet at a quite limited thing because we don't have a full description
00:31:02 - of the individual, if that's your question.
00:31:05 - It's never a full description of the individual.
00:31:08 - It's a description of a mechanism that takes place in a very specific case.
00:31:13 - So you may say, okay, it's not very useful because it's very narrow as a way
00:31:21 - to work. That's true. It's quite narrow.
00:31:23 - Not so many things explained.
00:31:25 - But that doesn't mean that more broadly, a dynamical system is not useful.
00:31:32 - Because again, if you think about the dynamical system as this network of positive
00:31:37 - feedback and negative feedback and non-linear effects, you find them to describe metabolic,
00:31:45 - genetic, neural nets.
00:31:47 - So the methodology of modeling is interesting.
00:31:51 - But I'm not claiming that it's a way to fully describe an individual in all
00:31:58 - occasions during all his life.
00:32:00 - But there's a problem here, right? Because I could also argue natural language
00:32:04 - works really well to describe cockroach behavior, so it's a valid method. it.
00:32:09 - But I think the question is, how does it help us to gain additional insight
00:32:14 - in the generation of behavior?
00:32:16 - So it's also a little bit, what's the leverage that it gives us, right?
00:32:19 - And there, I think there's an interesting issue because in some sense,
00:32:23 - the way you conceptualize behavior is in complete operational terms,
00:32:27 - also as the agent being under full control of its environment.
00:32:31 - Everything is in a direct control of external stimuli.
00:32:34 - And this, of course, was the same premise of which behaviorism built
00:32:38 - its science and also created its
00:32:41 - own doom because internal factors play
00:32:45 - a decisive role and you have to account for those as well so of course raises
00:32:50 - on the question for the for instance your case with the cockroach to what extent
00:32:53 - those internal states of that cockroach matter let's say hunger reproduction
00:32:59 - fear right so to what extent do these
00:33:03 - internal factors play a role, like motivational states,
00:33:06 - and then if so, how well can you capture those in that framework?
00:33:11 - First, they do matter, of course. And second, they are not captured by the model
00:33:17 - because the model is, again, the experiments were done in very specific conditions.
00:33:25 - So that we don't have to take care of those internal states.
00:33:29 - For instance, the cockroaches were well-fed or they were starving.
00:33:33 - So we are sure they start to explore, to look for food.
00:33:36 - There is no food in the system, so they're going to explore the system and stuff
00:33:40 - like that. They were kept in the dark
00:33:42 - so we're sure that all in the same physiologically sensitivity to light.
00:33:48 - And then you put them in an environment that has light.
00:33:51 - So the model only capture, and that's the limitation, with this specific experiment,
00:33:58 - given rather specific initial condition in terms of internal state of the animal,
00:34:04 - during a short lapse of time.
00:34:06 - So you may say, okay, but that's pretty narrow. And it is, but still we are
00:34:11 - learning, you know, it's a scientific process.
00:34:14 - It's very narrow, but still we are learning things.
00:34:17 - But then I could still argue, I could have then confirmed that model by running
00:34:22 - a pure cockroach-based experiment without inserting robots, right?
00:34:27 - Without inserting the robot probes. Yeah.
00:34:30 - And it has been done. Yeah. Yeah, so what's then the added advantage of also
00:34:36 - using the robot to test that very operational theory of the behavior?
00:34:41 - So there were multiple goals in that experiment.
00:34:45 - One of the goals was for roboticists, because roboticists from the EPFL were involved.
00:34:51 - We didn't build the robots, it's their job. So for them, for roboticists,
00:34:55 - it's interesting to build,
00:34:56 - in collective robotics, It's interesting for them to build biomimetic robots
00:35:01 - that are capable of performing a task and having a solution that looks clever, looks intelligent.
00:35:08 - So the motivation was, how do I build robots that, as a collective,
00:35:14 - can do interesting things?
00:35:16 - Okay, let's go biomimetic. That's their motivation.
00:35:19 - Our motivation was, okay, we can do experiments with cockroaches,
00:35:24 - but we can never really tune certain parameters,
00:35:29 - the number of informed individuals or the number of the type of interaction.
00:35:35 - If we do that, like a probe exactly that you inject in the system,
00:35:39 - you have another degree of freedom to perform experiments.
00:35:43 - And that's also a motivation. Then there is also another motivation that is
00:35:50 - hanging there, but we don't have any results result yet, but all we keep thinking about that is, okay,
00:35:57 - if you have this bio-hybrid system made of artificial agents and of natural
00:36:02 - agents, can it do something more?
00:36:06 - Uh than the the putting them together
00:36:09 - can you get something extra there that but
00:36:13 - due to the difficulties the experimental difficulties uh we don't have any results
00:36:18 - that show that you can do something more but at least you can show that you
00:36:22 - can connect uh living an artificial system that artificial system is interesting
00:36:28 - and that you can use them to tune some of the parameters.
00:36:31 - That's already an interesting result.
00:36:34 - So you, in this experiment, I think you already had an idea of what the dynamical
00:36:39 - system looked like, and you could sort of say what the equations might be that
00:36:43 - would govern their behavior.
00:36:45 - But if somebody is out there collecting a data set about animals cooperating
00:36:49 - or animals and robots interacting, how do they go about building a dynamical
00:36:55 - system description of that? Yeah, that's a completely different question.
00:36:59 - In fact, it's a completely, the methodology is completely different.
00:37:02 - So for instance, in the case of the cockroach again,
00:37:06 - and in the case of the ants, you have prior knowledge about those species,
00:37:12 - they were chosen because you have prior knowledge, prior biological knowledge,
00:37:16 - and, and, and Donnebourg invented an experimental methodology,
00:37:21 - more than a modeling method methodology, because the animal system,
00:37:24 - he hasn't invented them.
00:37:25 - I mean, even, so it's an old stuff, but the experimental methodology of binary chunks.
00:37:33 - With prior knowledge of specific species may lead to the fact that,
00:37:37 - okay, you understand if there is a social mechanism that produces,
00:37:41 - that makes this choice emerge. So it's a combination.
00:37:45 - No, for the fish experiments, we don't really have a prior model.
00:37:50 - So we have to build the robot, the fish experiment, all at the same time and
00:37:55 - to try to model the whole system at the same time.
00:37:59 - And it's much more difficult, in fact, because one of the drawbacks of this
00:38:04 - experimental field is you're combining two difficult tasks to perform collective behavior with animals,
00:38:12 - tedious, hard, difficult tasks, with the task of building robots from scratch.
00:38:18 - Tedious, difficult, and to program them and to make them biomimetic and stuff like that.
00:38:24 - So in a project, when you start to do that, it's hard work to get.
00:38:28 - So now you cannot just Science is not about looking around what's going on,
00:38:34 - collecting a lot of data, and magically out of the data you will have the insight that it's working.
00:38:40 - No, that's not the scientific method. At least that's not the scientific method I use.
00:38:44 - You have a predefined question.
00:38:49 - You gather data that you think is going to answer that predefined question.
00:38:54 - Or you build an experiment to gather that data, and out of that you prove that
00:38:59 - the model is working or not.
00:39:00 - So you have to design your experiment or the data acquisition in a way that
00:39:06 - captures, at least you guess, an educated guess that you capture something interesting out of there.
00:39:13 - So if you go to the classical example that Jean-Louis is always laughing about,
00:39:18 - okay, you want data, you go there, you have grass,
00:39:22 - you count the number of grass, little pieces of grass, and you have a lot of
00:39:25 - bunch of data. Are you going to learn something out of that?
00:39:29 - Not sure because it's not the amount of data
00:39:32 - gathering amount of data without having a hypothesis before
00:39:35 - so again it's a it's more an
00:39:38 - experiment experimental method than a
00:39:41 - mathematical model um methodology but there's a problem but there's a challenge
00:39:45 - here right because already the cockroach is a pretty complex organism okay and
00:39:51 - it can engage in many kinds of of complex behaviors it can navigate it has pretty
00:39:55 - good sensory capabilities um So in some sense,
00:39:59 - I could argue, but in your experiment, you push this high degree of freedom
00:40:03 - system in a very low degree of freedom task.
00:40:06 - And now you can show that with a low degree of freedom model,
00:40:09 - you can account for the behavior.
00:40:10 - But maybe you are actually, if you want, indeed counting sort of virtual grassroots
00:40:17 - that are fairly irrelevant towards really understanding what the cockroach is capable of.
00:40:23 - For instance, in the cockroach case, they're also outstanding in terms of their
00:40:27 - chemical sensing capabilities.
00:40:29 - And that is fully ignored in
00:40:31 - the current model and also in the robots that you have to work with them.
00:40:35 - So aren't you worried that there's a risk that your model is actually giving
00:40:42 - you such a low-dimensional description of this animal that it is almost meaningless
00:40:46 - given the complexity of its behavior in its actual ecological niche?
00:40:50 - Yeah, there are two answers to that. First of all, the model has implicit assumption.
00:40:55 - The implicit assumption do take into account the fabulous capabilities of those animals.
00:41:02 - What the model is taking into account is that they are very good at olfaction
00:41:06 - because that's the way they recognize each other.
00:41:10 - But you don't have to make a model of that olfaction necessarily because at
00:41:15 - that level, you can take it for granted. it, okay?
00:41:18 - So, but again, there is the trap of reductionism because it's a reductionist
00:41:24 - method, scientific method.
00:41:26 - And if you reduce too much your system, you may not get interesting results.
00:41:31 - And it's also the danger as.
00:41:34 - Any experiment to build an artifact, an artifact in the experiment,
00:41:39 - a bias experiment that is not relevant for the real stuff, the real system.
00:41:46 - It's always a danger, but it's the danger in every experiment with an experimental
00:41:50 - reductionist approach.
00:41:51 - You can either reduce too much and get not interesting result,
00:41:55 - or you can completely bias your system.
00:41:58 - But that you have all the time I mean, with that scientific method,
00:42:02 - and you have to be careful about that, of course, not to induce some.
00:42:06 - Then we have yet another interesting question, is that we don't have exactly
00:42:11 - the same mindset as the biologists.
00:42:15 - For instance, many biologists in behavioral science would say,
00:42:19 - okay, you have to really take care of what the animals are doing in their wild environment, right?
00:42:27 - In the natural environment. So for them, it's very important to study the animals
00:42:33 - in their real natural conditions, out there in the wild.
00:42:36 - And they are right, of course. We must do that.
00:42:39 - And then each time you bring an animal inside the lab, you're somehow starting
00:42:45 - to build an artifact in terms of experiment.
00:42:48 - Because they are not anymore in their natural environment. And so you may be
00:42:55 - studying something that is irrelevant for their natural environment.
00:43:00 - But I have the mindset of doing more than that.
00:43:04 - Yet, let's assume that it's really an artifact.
00:43:08 - They will never do that in their natural environment because such condition
00:43:12 - doesn't exist. What we are doing is testing the hardware.
00:43:16 - So if I can prove that the cockroach in a completely non-natural setup are capable
00:43:23 - of performing a clever choice,
00:43:26 - that means that this hardware is capable of doing it. And for me, it's interesting.
00:43:32 - For some biologists, they will argue that's irrelevant because they don't need
00:43:37 - to do that in their natural environment.
00:43:39 - It's always this discussion and it's going to be always there.
00:43:44 - But from my point of view, even if you're in a non-natural condition,
00:43:48 - if you prove that the hardware, which means the living system you're studying,
00:43:52 - is capable of doing that, you've learned something out of that.
00:43:56 - From my point of view as a psychologist, I guess what I'm hoping for out of
00:44:02 - this kind of research is some sort of general laws of social behavior that are
00:44:06 - going to apply across species and are going to generalize maybe to some interesting
00:44:11 - situations like people at football matches, things like this.
00:44:16 - And there does seem to be the potential for that with this kind of work.
00:44:21 - But I guess what you're also saying is that to actually demonstrate this concretely
00:44:28 - in terms of describing a set of equations that captures what's going on is very
00:44:34 - difficult and requires a huge amount of data.
00:44:36 - Well, Tony, what's also interesting there, in some sense, with your chicken
00:44:40 - experiment, or the experiment with the chicks,
00:44:43 - in some ways you showed the opposite Because you, in some sense,
00:44:47 - then invalidated the principles that were out there in literature for many decades,
00:44:52 - which is that imprinting works invariably for all chicks, right?
00:44:56 - They hatch, they see a mother-like figure, and they get imprinted,
00:45:00 - all of them invariably, on that object.
00:45:05 - And actually, your experiments that you were describing where you had chicks
00:45:08 - following robots, you saw a much higher variability.
00:45:12 - So, does it make you then more pessimistic about this ambition that Tony is
00:45:17 - sketching about identifying these sort of species independent principles?
00:45:22 - Well, it's a complicated question in fact.
00:45:27 - What you're looking for is somehow the ground of the and stuff like that.
00:45:35 - And it comes down, we may some do a cheap philosophy of science here,
00:45:39 - it comes down to this universal laws that we find in physics.
00:45:44 - Every theoretical physics at least is going to say, okay, Schrodinger equations
00:45:48 - and gravitation, so it's law of the universe. It's a bold statement.
00:45:55 - And we always dream of finding such simple laws because at the end of the day,
00:46:00 - you can write all of them on one single sheet of paper.
00:46:03 - Of course, to use them, it's more complicated, but at least you can write them
00:46:07 - on one single, a whole physics can be written in one single sheet of A4 paper, right?
00:46:13 - And then you are, wow, and that describes so many things and potentially including
00:46:17 - the structure of the universe.
00:46:19 - But if you work in the field of complex systems, and I'm from the statistical
00:46:23 - physics complex system field of studies, then you know that's not true.
00:46:30 - In fact, when you have a complex systems, you don't have a single model that
00:46:35 - captures everything of it.
00:46:36 - You need several models, a kind of kaleidoscope of models to answer different questions.
00:46:45 - Just to understand how this natural complex system works.
00:46:50 - And each time you ask a different question, maybe you need different kind of models.
00:46:54 - And then you say, okay, well, but still you're learning a lot of that complex system.
00:46:59 - But then it becomes a problem if you have a synthetic approach system.
00:47:04 - Because that's what we're really discussing, in fact.
00:47:07 - Let's imagine I want to synthesize something that is similar as a living system.
00:47:14 - I want an artificial cell, or I want an artificial cockroach,
00:47:18 - which is a completely different question.
00:47:20 - Then with this kaleidoscope, you don't know exactly how to use it.
00:47:26 - Because you're saying you will not have a single model of your system,
00:47:30 - then what do you do? Because you don't have the recipe to build it.
00:47:34 - And you are still, I think, I personally still don't know how to synthesize
00:47:39 - from scratch a complete system.
00:47:42 - Them so when we when you do the the cockroach
00:47:45 - robot at the end of the day that robot is not really interesting
00:47:48 - it's doing nothing it's behaving like a cockroach
00:47:51 - in a specific environment if you do if you
00:47:54 - you you look at in collective robotics the ant
00:47:57 - uh robots that they all follow each other and they do a trail a kind of pheromone
00:48:02 - trail whatever mechanism they use to do that great they do it but so what what's
00:48:07 - the use of those robots well they're a minimal description of what the cockroach
00:48:11 - must be doing in that task. Exactly. So that's useful.
00:48:14 - Yeah, but you're not synthesizing the robot that has been built is not a cockroach.
00:48:20 - It's a tiny subset of the cockroach. Now, if you want to build a robot that
00:48:25 - is much closer to what cockroaches are capable of doing, like you were mentioning,
00:48:31 - Paul, they have a lot of capabilities, internal state and so on.
00:48:34 - It's a completely different question and you are nowhere.
00:48:36 - We still have a lot of research to do. Well, just to kind of defend your research
00:48:41 - a minute, I think that what we're finding out with these kinds of experiments
00:48:46 - is how little complexity needs to be in the animal because the environment,
00:48:51 - and that includes the other individuals in it,
00:48:53 - will bring out all of these interesting emergent effects.
00:48:59 - And then that can apply also to us.
00:49:03 - We can say, well, this behavior that I thought I was doing was very sophisticated
00:49:06 - turns out to be some fairly simple control mechanism in my brain that's responding
00:49:11 - to the environment in this way.
00:49:14 - So it's a sort of classic Simon argument that the complexity is not in the ant.
00:49:19 - It's in the ant and in the environment interaction.
00:49:21 - Yeah, that I agree. That's one of the major lessons is that you don't need a
00:49:25 - very high complexity to produce what you are observing.
00:49:29 - But yet we have to think again about that. Well, it's a challenge we discussed earlier, right?
00:49:34 - You do constrain the degrees of freedom of the organism a lot.
00:49:39 - And then you can say, okay, it's a simple control system. But you already know,
00:49:42 - as you saw with the chicks and with the fish, suddenly life gets way more complicated
00:49:47 - and these simple rules already don't hold anymore, right?
00:49:50 - So I found it interesting that
00:49:51 - with the chicks, you actually observed this massive variability, right?
00:49:57 - That seems to be really systematic, right? So in that sense,
00:50:01 - maybe the notion of simple rules should also be critically analyzed because
00:50:08 - maybe a simple rule also is a simple rule that should imply the ability to generate
00:50:12 - huge variability across a population.
00:50:15 - And we're not really used to that, right? We want to think about gravitational
00:50:18 - forces and that's sort of pretty deterministic in that sense.
00:50:22 - So, but what you also did in the case, in the experiment with the chicks and
00:50:26 - also with the fish, is you start to automate much more of how you process this data, right?
00:50:32 - So you get sort of to an etonome, how would you call that? Atomics,
00:50:37 - it's not me. Atomics, okay.
00:50:39 - So that means you start to now capture massive amounts of data from behavior
00:50:44 - in an automated fashion, right?
00:50:46 - And in the case of the chicks, that really paid off because this allowed you
00:50:50 - to observe something most people had ignored, which is this variability, right?
00:50:54 - That you had sort of a larger portion of the animals indeed showing the imprinting,
00:51:00 - a smaller part actually not showing it, and an even smaller part actively avoiding
00:51:05 - this mother object, yeah?
00:51:09 - So do you see this sort of massive scanning of behavior in this automated fashion,
00:51:15 - session, including then inserting robots that now automatically get optimized
00:51:20 - to realize certain perturbations as the future of neuro-ethology?
00:51:24 - Yeah, that's one of the dreams, in fact.
00:51:27 - And still it's a lot of work. It's to try to automate as much as we can the
00:51:32 - theoretical and the experimental method.
00:51:37 - So to automate the experiments and to automate the data analysis and to automate the model generation.
00:51:44 - And it's a challenge. It's completely open. But it's sure if we make progress
00:51:51 - along that line, then we will be capable of maybe starting to do experiment,
00:51:58 - model production, embodiment of the model.
00:52:02 - Embodiment implies coding the controls of those robots.
00:52:07 - And all that, at the moment, take years.
00:52:11 - But, so take years, I mean, for simple, simple between quotes,
00:52:17 - but yet simple experiments like we have done.
00:52:20 - So we have to compress that time. So we, at the end of the day,
00:52:25 - the only thing that you cannot compress is the physical time of performing the experiments,
00:52:30 - right, because if you have to do experiments of one hour with some fish,
00:52:34 - you cannot compress that time. It's imposed.
00:52:37 - But then if you can analyze the data, produce some kind of model,
00:52:42 - produce some kind of controllers of the robot in the same lapse of time,
00:52:47 - you're saving a lot of time.
00:52:49 - And there are interesting theoretical questions out there. Can we automate model generation?
00:52:57 - It's again a holy grail of modeling. To what extent can we automate data analysis?
00:53:07 - It's again a holy grail in the system, the unsupervised machine learning that
00:53:13 - everybody's dreaming about.
00:53:14 - Because that's again a holy grail. So there are many fundamental questions behind that approach.
00:53:21 - Yeah, but then in your last set of experiments with the zebrafish,
00:53:25 - where also Also, the robots you insert have become much more sophisticated,
00:53:29 - and also, if you want the data analysis is more advanced and so on.
00:53:33 - However, you were not able to really systematically control or influence the
00:53:39 - aggregation behavior of these zebrafish.
00:53:44 - So what's the difference now going from the cockroach to the zebrafish?
00:53:47 - Why is it not working so easily? Why is the zebrafish so much more complicated
00:53:50 - to control? They have a completely different structure.
00:53:55 - They are completely different animals. Of course, they have vertebrates. They are not, etc.
00:53:59 - They live in a completely different environment. And they have a completely
00:54:02 - social and simple life, everyday life that's completely different.
00:54:09 - For instance, the cockroaches, they settle.
00:54:12 - After a while, they settle. They want to explore their environment.
00:54:16 - If they find a shelter, they settle there. The zebrafish are moving animals.
00:54:21 - They never stop moving during the experiment. They never settle.
00:54:25 - Okay, they stop from time to time, but they don't settle there after a while.
00:54:29 - We've been trying to give them some kind of shelter to see if they will select one.
00:54:36 - No, they don't do that. They oscillate between one to the other.
00:54:40 - So those animals are on the move all the time. But do they need that for their
00:54:44 - oxygen supply or for the gills?
00:54:46 - They need that because that's the way, exactly, that's the natural way of life
00:54:50 - in those ponds there in Nepal or India.
00:54:53 - They're always on the move looking for food, avoiding predators, for instance.
00:54:59 - And then they're in water, you have streams in water and stuff like that.
00:55:04 - So these are moving animals.
00:55:06 - They're all the time on the move. And so the question is, not only do you have
00:55:11 - to be accepted, you have to be part of the group, you have to be moving in the
00:55:14 - same time as the group, and then you have to try to influence,
00:55:18 - to modulate that on-the-move movement, perpetual movement during those experiments.
00:55:26 - So it's more challenging in fact. Right.
00:55:30 - So do you see this generalized to humans anyway? I'm not thinking too much about humans for the moment.
00:55:39 - Because I'm not well trained to understand the human behavior,
00:55:44 - but the methods, the scientific methods could be used in some cases.
00:55:51 - For instance, crowd movement, it's clear that many people are doing that kind
00:55:55 - of similar method and try to model crowd movement.
00:56:00 - Because crowd movement doesn't involve high cognitive capabilities.
00:56:05 - But I don't have a good knowledge about human behavior.
00:56:10 - And then again, I think this methodology can be also interesting,
00:56:15 - but for certain different kind of questions, let's say.
00:56:20 - But this is for you the outlook, so that your main thrust now is towards these
00:56:27 - more automated neuroethology experimental systems that autonomously do.
00:56:32 - I have two questions, let's say, if you want my roadmap of research until I'm retired.
00:56:39 - I have this set of methodological questions.
00:56:43 - How can we automate the process of data analysis and model generation and generating
00:56:51 - artificial systems that are capturing part of what we observe in the living system?
00:56:56 - I'm not the only one, it's a very broad question that is addressed by many people.
00:57:03 - And then there is another one which I find quite interesting,
00:57:07 - in fact, you know why, for reasons of sustainability again, living systems are
00:57:14 - going to be a very, very essential.
00:57:18 - They always, they've been always essential for humans, but we're back to think
00:57:23 - again about how essential they are.
00:57:26 - And in fact, all that research is about interacting and modulating the living
00:57:34 - system you're interacting with, right?
00:57:36 - And again, you find that at all level of living system.
00:57:40 - There are people who are interacting with cells and they want to modulate what the cells is doing.
00:57:45 - Like me, we want to modulate social behavior. You want to modulate metabolism.
00:57:50 - You want to modulate many things in living systems.
00:57:53 - And then you don't need, and that's one of the lessons of those models,
00:57:58 - Tony, if we go back to your question.
00:57:59 - You don't need necessarily to build a completely artificial living system to
00:58:05 - modulate the natural living system.
00:58:08 - You see what I mean? you can capture part of the living system,
00:58:13 - build an artificial one, a machine, let's say, that is going to interact with
00:58:19 - the living system and drive it to a state that you find interesting.
00:58:23 - It's very common. Breweries have been doing that for hundreds,
00:58:27 - thousands of years, I think.
00:58:29 - So the brewer, what is it doing? He's putting a living system in a tank and
00:58:34 - then it's driving this tank, controlling the temperature, the sugar level.
00:58:38 - To produce, to drive the yeast, to produce either wine or beer or something else.
00:58:44 - So that's a way to control a collective system, which is called yeast, right?
00:58:50 - And at the end of the day, you get the interesting product.
00:58:53 - In fact, we have to start to generalize that way of thinking,
00:58:57 - to have a system that drives a living system, But in an automated,
00:59:04 - in an autonomous way, in an automated,
00:59:06 - but also in an autonomous way, because of course, if you take brewery as one
00:59:11 - of the beautiful biotechnology that is centuries old,
00:59:16 - it's the brewer that is driving, keeping care of his tank fermentation and driving the system.
00:59:23 - Now, imagine you have a robot that is driving the tank and doing the beer. Can we generate a robot?
00:59:31 - Yeah, there's some automation for that kind of fermentation stuff.
00:59:36 - But can we find other systems where you want to drive living systems and you
00:59:45 - want it to be driven by another autonomous, an artificial autonomous system, a bio-rhebride system.
00:59:50 - And why do you want to do that? Because you want to explore the capabilities
00:59:55 - of the living system to produce interesting products for you.
00:59:59 - Beer and wine are extremely interesting products for us, but you can also dose-producing
01:00:05 - drugs or producing other kinds of materials that you want to extract from the living system.
01:00:12 - We're approaching a time in our history which is going to be a real step change
01:00:18 - when we're going to have robots, not just in factories, but in society and particularly in our streets.
01:00:27 - And I'm thinking perhaps most immediately we're going to have lots of vehicles
01:00:32 - on the roads which are actually robots interacting with other vehicles which are driven by people.
01:00:39 - So I'm wondering if these kinds of approaches that you're having is going to
01:00:45 - generalize up to, say, looking at the impact of driverless cars on how people will drive their cars.
01:00:53 - Because people are speculating, you know, once there are driverless cars on
01:00:56 - the road, other people will change their behavior because they will know that
01:01:00 - the driverless cars behave in predictable ways.
01:01:02 - But I think that there is going to need to be more of a science of this,
01:01:06 - because right now it is just speculation.
01:01:09 - Yeah, absolutely. I wouldn't dare by the methods I'm doing to apply immediately
01:01:14 - to bear, but that's a good question.
01:01:15 - When you introduce some autonomous system in contact with a living system, what's going to emerge?
01:01:21 - And we have no idea how to do that for the moment.
01:01:24 - We have still to work about that. How do you drive even a yeast cell? We have no idea.
01:01:32 - You say, how do you control, to what extent you can control its mechanism or
01:01:37 - its genetic regulation?
01:01:38 - It's not cleaner. But what your research shows is that we ought to be wary of
01:01:43 - the fact that there might be small perturbations you can make to the system,
01:01:46 - which could have very large-scale effects.
01:01:48 - That's selection from complexity and dynamical system.
01:01:51 - If you have nonlinear system, small change in the initial condition can lead
01:01:57 - to a big change in the system,
01:02:01 - the third lesson is a small change in the parameters of the system can lead to a bifurcation,
01:02:07 - which means a bifurcation means solutions that appear or disappear.
01:02:12 - And it's one of the major concerns if you think about ecosystem or the biosphere or the climate,
01:02:20 - it's that because those systems are most probably non-linear,
01:02:25 - if we start changing some of their parameters,
01:02:29 - like the quantity of carbon dioxide we're injecting in the atmosphere,
01:02:34 - we may have a strong non-linear effect and we cannot exclude that it can be a dramatic change,
01:02:41 - in the properties of the biosphere.
01:02:44 - We've seen collapse of some ecosystems that are robust to a certain extent,
01:02:51 - and then suddenly they collapse.
01:02:53 - So we've seen examples of that.
01:02:56 - And I'm sure that we don't have good ideas of how to deal with that or to control that.
01:03:01 - It's still a knowledge that is not well established.
01:03:04 - For instance, we have people that have been managing eutrophization of a lake, right?
01:03:10 - Suddenly you inject a fertilizer there you have a boom of algae and then you kill all biodiversity.
01:03:18 - There that is well known and then you have to manage how do
01:03:21 - I put back my lake in a state that I can have fish and biodiversity so this
01:03:27 - kind of general question I think has to be addressed also by the same kind of
01:03:31 - tools and again the non-linear dynamical system are good tools to address those questions but also that,
01:03:40 - type of model shows that the system will be intrinsically unpredictable.
01:03:44 - Certainly if this individual agents become more non-linear like in the case
01:03:49 - of having humans around.
01:03:51 - So if you talk about inserting, for instance, autonomous technologies in society, then.
01:03:57 - We should think very carefully about, indeed, what kind of reactions will trigger
01:04:01 - in the environment in which they have to operate, right?
01:04:04 - And these environments will, this will include humans.
01:04:07 - So now, so we look at this complex interaction between, in this case,
01:04:10 - artifacts, humans, and environments, right?
01:04:13 - And also in your model, you would say, well, the behavior we observe is indeed
01:04:16 - the function of, let's say, the controller of the agents, their morphology,
01:04:20 - as you showed us with the cockroach, and the environment itself.
01:04:24 - How many shelters are there, as an example, right? but juxtaposed to that,
01:04:30 - you're saying, but there's no hierarchy, right?
01:04:34 - Or you don't feel that the notion of hierarchy and hierarchical relations is
01:04:38 - helpful in trying to understand that system.
01:04:41 - So how should I be able to, how can I combine these two positions in a consistent framework?
01:04:48 - Yeah, but it depends what you mean by hierarchy. In the case of the cockroach,
01:04:52 - it's hierarchy at the social level.
01:04:54 - But then in terms of complexity, if you start to think that it's the system
01:04:59 - that is the system, you cannot cut part of it because the whole thing is a system.
01:05:08 - It's like people are doing now system science for Earth.
01:05:13 - With this controversial hypothesis of the Gaia hypothesis, but even if you get
01:05:17 - rid of that controversial hypothesis.
01:05:20 - Most people are now thinking about the Earth system as a whole.
01:05:26 - Because if you made a change in the climate, it's going to change biodiversity.
01:05:31 - But if you change biodiversity, you're changing also the local climate.
01:05:35 - So you see that you have to take into to call the whole system.
01:05:39 - And then what I'm claiming there is that, okay, the system is the system.
01:05:44 - There is no different level, but we have to work with that.
01:05:49 - So there is a methodology to work with that, is to select different level of description.
01:05:55 - But they have been chosen to answer a question.
01:06:00 - That doesn't mean that they have their own existence.
01:06:04 - They are not ontologically existent there.
01:06:07 - But it's a natural, sometimes it's a natural way to cut a system.
01:06:11 - For instance, if you're thinking about a solid, okay, you have the molecules, the atom.
01:06:16 - That's a very evident way, natural way to cut your system.
01:06:21 - But then if you have more complex way, more complex system than a crystal,
01:06:28 - a solid crystal, if you have the earth as a system,
01:06:31 - the way you're going to cut the system to answer a question is open.
01:06:37 - It's up to you to find the relevant elements in the system to build a model
01:06:45 - to answer your question.
01:06:47 - You're not stuck to a given, obviously given level of description because there
01:06:54 - would be the level of description.
01:06:57 - And it's one of the open questions in complex systems with emergence.
01:07:03 - You know, it's also a controversial term, emergence.
01:07:07 - Choosing,
01:07:10 - What you want to include in your description, your model, depends on the kind
01:07:15 - of question you're going to address.
01:07:16 - For instance, with the fish or with the cockroaches, if I just want to describe
01:07:20 - their social behavior, I don't
01:07:22 - need to go to discuss their sensory motor system or their neural system.
01:07:29 - I don't need that. But no, if I want to study the impact of the sensory motor
01:07:36 - system on social behavior, then of course I have to include that level also.
01:07:41 - But that's a different scientific question. I want to know the impact of the
01:07:45 - sensory motor system on social behavior.
01:07:48 - My other question was simply, I want to understand social behavior at the level
01:07:53 - of the population without taking into account other levels.
01:07:58 - If you want to now to study what is the metabolic impact on social behavior,
01:08:06 - because the state of the agent is changing depending on its metabolism,
01:08:10 - yet another question, then you start to include the metabolism.
01:08:16 - So we have to think as a system as a whole, but then as a methodology,
01:08:21 - we have to decide what are the relevant pieces that we have to take into account
01:08:25 - to have a good description,
01:08:28 - to get a good answer to the question we're addressing.
01:08:31 - And we have to be open-minded about the pieces we are including.
01:08:35 - So you had this talk about disease, mental disease and stuff like that.
01:08:42 - You have a whole system that is the human being, and then you can.
01:08:47 - Re-cut it in many different ways to take into account all the pieces that are
01:08:51 - relevant to answer the question, how do I cure that specific disease?
01:08:55 - One thing that's distinct about your approach is that, as you say,
01:08:59 - a lot of people will take an approach where they try, for instance,
01:09:03 - to understand social behavior in terms of sensory motor systems.
01:09:07 - So you're trying to describe what happens at the group level in terms of some
01:09:12 - process going on inside the individual.
01:09:14 - And most of the science you're describing, is saying, let's look at this level
01:09:19 - and see how it impacts on this other level.
01:09:21 - So how does metabolism affect social behavior? How do genes affect social behavior?
01:09:26 - But the dynamic systems approach doesn't seem to do that. It seems to say,
01:09:29 - let's try and capture behavior in terms of behavior.
01:09:33 - There is no crossing between two levels.
01:09:36 - Yeah, but that's, again, complex systems. Complex systems have,
01:09:39 - let's say, a level to simplify the discussion at which you can describe them.
01:09:45 - You can take into account the temperature of a gas.
01:09:50 - You don't have to take into account the kinetic energy.
01:09:53 - Simple temperature is good enough because you have this different way of measuring
01:09:58 - that or describing that.
01:10:00 - Sometimes you don't need all those details in terms of kinetic energy or vibration
01:10:05 - or quantum effects in the same way.
01:10:08 - The normal temperature that everybody knows with a thermometer could be good enough.
01:10:13 - But sometimes not. Sometimes you need to get into the other kind of details
01:10:17 - because you want a different answer.
01:10:19 - For instance, with the fish, what we're trying to do now is,
01:10:22 - okay, we have plenty of models of bird flocking or fish schooling and schooling, right?
01:10:27 - And the Vichek-like model that they've been done in physics in the 90s.
01:10:32 - But I say, okay, I have a different question. What kind of information do they
01:10:37 - take into account and how do they process that information, right?
01:10:42 - That's a different kind of question. Then I start, I have to,
01:10:45 - I have to start to open the black box of the individual.
01:10:49 - So I have to think about what is the vision in how vision works in fish,
01:10:55 - in that specific species of fish.
01:10:57 - And I have to kind of some kind of have a model of that, a minimal model.
01:11:01 - I don't have to take into account all the details of vision of the eye and stuff
01:11:05 - like that, a minimal model.
01:11:06 - And then what's the action we get in response to that perception that,
01:11:12 - and we did a model like that.
01:11:13 - You have the field of perception, the fish sees where are the individuals,
01:11:19 - and then it makes a probabilistic decision to go to certain direction.
01:11:24 - Then you have the other step. Okay. In that model, there is no processing of the information.
01:11:30 - There is this probabilistic description of the decision that the fish makes
01:11:35 - to go towards a certain direction.
01:11:37 - Now, if I want to understand how that information is processed.
01:11:42 - Right? Yet another question, yet another layer to add to the system.
01:11:47 - Yeah, but there's a challenge here, you know, because Poincaré already,
01:11:52 - you know, was talking about what concepts as scalpels in which you sort of carve reality.
01:11:58 - And Plato in his dialogue, Phaedrus, has this famous, this motto of that you
01:12:05 - carve nature by its joints. That means there's some intrinsic structure,
01:12:09 - and by accessing that structure, we can gain knowledge.
01:12:12 - But now, if we discuss hierarchy, right, across these many levels of organization,
01:12:19 - you seem to be saying, well, we should not commit ourselves too strongly about
01:12:24 - hierarchical relations.
01:12:26 - But does it imply that also intrinsically there is no hint of a hierarchical structuring?
01:12:32 - And don't you run the risk then of basically advocating, let's say,
01:12:35 - okay, there's an amorphous structure,
01:12:39 - with many elements that have no specific relation in a hierarchical sense and
01:12:44 - I can now arbitrarily group them in any way I like.
01:12:48 - Because, of course, that search space would be huge, right?
01:12:52 - So do you really believe that there is no intrinsic hierarchical organization in these systems?
01:12:57 - No, the patterns exist. I mean, living organisms are not amorphous stuff,
01:13:03 - a bunch of things interacting.
01:13:05 - The structure exists. The shape exists. The patterns exist in space and time and stuff like that.
01:13:11 - But the whole thing is a system.
01:13:15 - So some people say, okay, you take a thermostat, you cut the wire,
01:13:20 - you don't get the feedback, then it doesn't work. So you see, aha.
01:13:24 - Of course, because the thermostat is the system, you cannot cut the wire,
01:13:28 - otherwise you don't have a thermostat anymore.
01:13:30 - You have to consider everything together.
01:13:33 - But that doesn't mean that you don't have a wire, a temperature sensor,
01:13:38 - and some way to act on the radiator that you need. So, all those pieces do exist
01:13:44 - and they are displayed in a certain pattern.
01:13:48 - You have to put your temperature sensor at the right place.
01:13:52 - You need to act on the heating device in a certain specific way.
01:13:58 - So, the pattern exists, but you get regulation of the temperature,
01:14:05 - homeostasis of the temperature, if you have the whole system.
01:14:10 - And then if you want to understand how it works, okay, for a thermostat,
01:14:14 - it's pretty easy, but still you can get instability in thermostat in terms of
01:14:18 - negative feedback and you get oscillation and stuff like that.
01:14:21 - You see, then you start to analyze how it, and then you can recut your system
01:14:26 - any way you find it interesting to answer the question, how do I get a stable
01:14:32 - temperature, a chosen, given stable temperature?
01:14:36 - No oscillations in the temperature, no instability and stuff like that.
01:14:40 - So for living systems, it's still an open question.
01:14:43 - How do you shuffle the things? things, but you see that you always have an incomplete,
01:14:48 - again, talking this morning about this genome,
01:14:53 - proteome, neural nets, immune system, well, when you take into account,
01:14:59 - if you want a biomedical application,
01:15:01 - you have a disease, it can involve many, many things at the same time.
01:15:08 - So how do you choose that?
01:15:09 - Or you may decide, okay, it's a kind of cognitive dysfunction, so it's the brain.
01:15:15 - How do you know? It's a guess. I mean, it's an educated guess.
01:15:18 - It's probably involved.
01:15:19 - Okay, it's obvious. But how do you know it's only the neural nets?
01:15:23 - No, it's not Glyosense, it's not the other state, the physiological state.
01:15:27 - It's not a genetic defect.
01:15:29 - So you have many, many reasons to think that many other pieces can be involved.
01:15:35 - So some people are obsessed by the brain in itself and not even the brain in itself.
01:15:40 - You know, Paul, it's only a set of neurons in itself. Like there is only one
01:15:45 - type of cell in the cortical column.
01:15:49 - Some people are doing models where you have only one type of cell.
01:15:53 - And they pretend they are capturing some. Well, are they? Maybe.
01:15:58 - But you know there are all the different ways you can describe just a cortical
01:16:03 - column because there are many other things involved there.
01:16:06 - So taking the only way, the only methodology you may have is to have educated
01:16:13 - guess of what pieces to take into account because it's a system,
01:16:17 - it's a complicated system, not only a complex system, but it's a complicated system.
01:16:22 - And then you have to try to grasp the elements that are going to help you to
01:16:27 - have a model to answer a specific question.
01:16:31 - But do you believe, also as a physicist, that the math, the mathematics is all
01:16:37 - there, the framework is there, we just have not worked hard enough to apply
01:16:40 - them to this domain of biology and psychology?
01:16:44 - Or do you believe the math has just given us a starting point and most of the
01:16:49 - work still needs to be done?
01:16:51 - Well, that's again a big question. There are some people who think we don't
01:16:55 - have the math to describe a complex system because there is something difficult there.
01:17:00 - And some others are claiming, yeah, but we've done already a lot of,
01:17:04 - we have a lot of mathematical models to describe the system.
01:17:08 - I'm not sure. I have no specific answer to that.
01:17:12 - But clearly we are, if we look at the history, let's say again of physics only,
01:17:17 - because physics is clearly linked to mathematical methods.
01:17:21 - What physics has been doing since the beginning, if you place the beginning
01:17:25 - of physics with Galileo, Galilei, let's say, physicists and mathematicians have
01:17:30 - been working hand-in-hand to invent at the same time the physics models and
01:17:34 - the mathematical tools.
01:17:36 - So maybe we still need to invent new tools.
01:17:41 - And in physics, there is this famous problem of the N-core problems.
01:17:48 - When you have N bodies there, N entities,
01:17:53 - then it's becoming a mess because if you want to
01:17:57 - describe them that the so-called fundamental laws of physics as
01:18:00 - soon as you have three bodies there it's becoming
01:18:03 - a mess in terms of you cannot solve the equations and imagine
01:18:07 - you have avocado bodies there so so it's a it's a it's a big issue that's why
01:18:12 - statistical physics has been developed and those methods so now we have complex
01:18:16 - system with uh with a huge number of elements there so the question is.
01:18:24 - Will we need new models or new mathematical techniques? Why not?
01:18:29 - Why not? I'm not going to invent them myself.
01:18:32 - But the other thing is that you are in this tradition, you come out of this
01:18:35 - tradition that since the late 80s more or less was also advancing this link
01:18:41 - up of dynamical systems and life and also feels like artificial life and so on.
01:18:47 - So now we're almost 40 years later, I'd say 30 years later.
01:18:52 - How much progress did we then make on really understanding what life means so
01:18:57 - does this model you advance actually help us to understand and to define life.
01:19:04 - So little piece of it so we're still far,
01:19:09 - away from understanding well there was the famous quote a fine man on his blackboard
01:19:15 - which is you know it better than me what I don't understand I cannot construct No,
01:19:21 - I can only understand what I can construct, what I can build.
01:19:26 - So if you don't understand the system, you cannot build it, basically. Or the other way around.
01:19:32 - Actually, it goes back to Gian Battista Vico, the 18th century philosopher,
01:19:36 - who says that the truth and the fact are reversible.
01:19:39 - But then again, you think about the people in synthetic biology.
01:19:43 - There is no such thing as a synthetic cell.
01:19:47 - Purely synthetic, I mean, from scratch. So what people are doing for the moment
01:19:53 - in the field to try to understand how it works is taking piece apart and reassembling
01:19:58 - them or injecting pieces from there.
01:20:00 - You take the genome from one cell, you inject it in another cell,
01:20:03 - you take piece, you take a subset of a genetic regulation, you inject it.
01:20:08 - So we are tinkering still with what the cell should be.
01:20:13 - But we don't have a full understanding of just a cell, a bacteria,
01:20:18 - or a yeast cell. There are still things that we don't know.
01:20:23 - So now look, so we made the grand tour. Yeah, we made the grand tour of dynamical
01:20:29 - systems, behavior in our life.
01:20:33 - And you came also a long way in that whole adventure over the last decades,
01:20:37 - starting as a physicist, now trying to understand complex behavior.
01:20:42 - So if you would like to follow in that tradition, the José Alloy tradition,
01:20:46 - what would be José's law that we have to adhere to?
01:20:55 - It's more a way of thinking and a methodology than a law.
01:21:02 - I don't think for, and I'm skeptical that for complex, real complex systems,
01:21:07 - we will have the law or the simple set of equation that, like we have Maxwell's
01:21:14 - equation, Einstein equations, or Schrödinger equations.
01:21:17 - I think that doesn't work for real complex systems.
01:21:20 - It's a conviction. I cannot prove it.
01:21:23 - I have no theorem that proves it, but, um, when you're facing such system like
01:21:29 - living system or the earth system that includes living system,
01:21:32 - it's so complicated that we will not have a single set,
01:21:37 - simple, single set of laws, but we need a huge collection of good models capturing
01:21:44 - pieces of the system to, uh, interact with it.
01:21:49 - And so we have to be quite modest. That's why people are afraid of geoengineering.
01:21:54 - And there are people who want to interfere with climate by geoengineering.
01:21:58 - Well, that's a big question.
01:22:00 - You better think twice or even more than twice before starting to do that because
01:22:05 - we have no single idea what's going to happen.
01:22:08 - At the same time, because climate
01:22:10 - change is becoming such a big existence issue for the the humankind,
01:22:16 - well, you may start to think about how to interact with that complex system
01:22:22 - that is the earth system.
01:22:23 - Yeah, but look, we want to put your law, José's law, on a bumper sticker for Tony's car.
01:22:28 - So you cannot have like two pages.
01:22:32 - So what's José's law that we can fit on a bumper sticker that fits on Tony's car? Think again.
01:22:39 - Okay. Then three years from now, we're going to come visit you in Paris where
01:22:44 - you're going to take us out for dinner.
01:22:45 - And then we're going to interrogate you about a prediction you're going to make today.
01:22:50 - And the question will be three years from now, did you confirm this prediction?
01:22:54 - So what's the prediction you're going to make today that you're going to give
01:22:58 - us the answer, the confirmation to three years from now?
01:23:01 - What's the main hypothesis you would commit yourself to within your domain of research?
01:23:07 - It depends which domain you're talking about.
01:23:12 - The one about animal-robot interaction? The one we discussed today?
01:23:17 - I think my prediction is that we will not have made a lot of progress in the next three years.
01:23:26 - Well, José Aloy, thank you very much for this conversation.
01:23:36 - The CSN podcast was produced by the Convergent Science Network of Biometrics
01:23:42 - and Biohybrid Systems. a project funded by the European Sevens Research Framework Programme.
01:23:50 - For more interviews, recorded lectures or upcoming conferences in the field
01:23:55 - of biomimetics and biohybrid systems, go to csnnetwork.com.
01:24:01 - Music.

CSN Podcasts

José Halloy on collective behavior and bio-hybrid robots

Season 2018

Description

CSN Podcasts

Episodes

Ton Coolen on immune networks and neural networks

Paul Verschure & Tony Prescott on synthetic psychology and robot models

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

Timestamp

Be the first to leave a comment

Leave a commentCancel

Legal

Follow us on

José Halloy on collective behavior and bio-hybrid robots

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