Watch Episode Here
Listen to Episode Here
Show Notes
This is a ~25 minute talk and another ~25 minutes of discussion between Benjamin Lyons (https://benjaminflyons.com/) and myself, about his work to understand the commonalities between economics and collective intelligence in biology. Some links mentioned in the discussion:
https://interestingessays.substack.com/p/prices-as-memories
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4878765
https://www.frontiersin.org/articles/10.3389/fpsyg.2019.02688/abstract
https://www.mdpi.com/1422-0067/24/1/285
https://www.cell.com/iscience/fulltext/S2589-0042(21)00099-7
https://direct.mit.edu/artl/article/20/2/183/2768/The-Cognitive-Domain-of-a-Glider-in-the-Game-of
CHAPTERS:
(00:00) Prices As Cognitive Glue
(25:52) Minimal Rational Agents
(32:09) Price And Collective Selves
(38:37) Learning Across Systems
(45:39) Information Flow And Pathology
PRODUCED BY:
SOCIAL LINKS:
Podcast Website: https://thoughtforms-life.aipodcast.ing
YouTube: https://www.youtube.com/channel/UC3pVafx6EZqXVI2V_Efu2uw
Apple Podcasts: https://podcasts.apple.com/us/podcast/thoughtforms-life/id1805908099
Spotify: https://open.spotify.com/show/7JCmtoeH53neYyZeOZ6ym5
Twitter: https://x.com/drmichaellevin
Blog: https://thoughtforms.life
The Levin Lab: https://drmichaellevin.org
Transcript
This transcript is automatically generated; we strive for accuracy, but errors in wording or speaker identification may occur. Please verify key details when needed.
[00:00] Benjamin Lyons: This is going to be conceptually pretty similar to what we talked about last time, but it's a little more condensed and focused onto just the core: prices are the absolute basics of understanding the economics of collective intelligence. I'm going to go through it. It's fairly dry and abstract, but the idea is if you can understand everything in this talk, then you've got the bare minimum. It's about 20 or so slides. It should take about half an hour, and then we can discuss for the rest of the time. This is a talk of prices as cognitive glue, an introduction to the economics of collective intelligence. Here are our goals. Our primary goal is to learn the bare minimum economics for understanding the general abstract principles of collective intelligence. Regarding bare minimum: I don't know a lot about biology, but I do know some absolute basics. I have a basic concept of what a cell is, a basic concept of what DNA is, and this is helpful for understanding your work. Similarly, this presentation should give you the absolute bare minimum for understanding the basic economics of collective intelligence. If you can grasp everything in this presentation, then you should be good to go. Regarding the abstract side of things: biology is very useful for understanding collective intelligence because it lets you see concretely that you start as a single cell, you're just physics, and there's no magical transition where you become a real intelligence. Economics allows us to do the complete opposite where it naturally gives us a bird's-eye view and we can see the whole forest without having to navigate through all the trees. Having the concrete and the abstract is very powerful. Specifically, we're going to understand three ideas. The first is that economic production, even building something as simple as a pencil, is actually incredibly complicated. You have to coordinate across millions of people, dozens of countries, and tens of years. You need some way of managing that. The way that gets managed is through the price system. The price system is the cognitive glue that makes this possible by connecting people across space and time. We will focus on how the price system works, with an emphasis on the idea that prices erase the distinction between other people's stresses and rewards and your own, turning one person's problem into everyone's problem. The hypothesis backgrounding this talk, which I cannot prove but think is quite plausible, is that all collective intelligences work according to the same basic economic principles. On the surface, they might seem very different, like a cell versus a person versus an economy, but underneath the hood, it's all just the same economics. Let's get into it. Let's talk about making a pencil. This is a very familiar object. This down here is a little diagram of what happens once you've got a nice block of wood. It seems very easy. You cut down a tree, you divide up the wood into these slats, you slide some graphite in, you glue the slats back together, then you dice it up and you attach some brass and rubber at the end. And you get some nice lettering for Dixon Ticonderoga, who are not sponsoring this, but just a little free advertising. It seems very easy to do. But in reality, making a pencil requires an immense degree of coordination that seems virtually impossible once you start breaking it down. To chop down a tree, you need a saw. To make a saw, you need some ore. These are going to be in different places. Maybe your tree's in Oregon, but your ore is in West Virginia. Then you need refining processes to take that ore and make it into steel and manufacturing processes to take that steel and turn it into saws and helmets and so forth. As you keep iterating, it grows like this. All of these processes are very complex. They take a lot of people, a lot of training, a lot of knowledge, and somehow all of it has to come together in a very complicated way. This isn't just true for chopping down trees, but it's true for the graphite, the brass, et cetera. All of the elements of a pencil are very complicated to actually acquire, and each and every one of them requires an immense degree of coordination. Not only are they individually very complex, but they all have to come together at the same point in space and time so that you can actually combine it into a pencil. They all need to be coordinated with each other.
[04:18] Benjamin Lyons: And in the modern age, these processes don't even exist in the same country. You might have manufacturing in China, wood from Brazil, rubber from Mexico. These processes take years to complete, and they're not on any kind of shared timetable. Managing this process seems incredibly complex, and you have to manage it without a pencil, because you haven't built a pencil yet. So you can't even write things down. The question is, how are you able to do this? What's most incredible about this process is that the degree of coordination necessary is achieved with so much robustness and reliability that I can take this incredible pencil, which we now know takes millions of people and decades to produce, and snap one in half and nobody cares. Because these pencils are so widely available, the system works so well that its outputs are entirely unremarkable and you can easily predict that you're going to have access to them. There has to be some kind of system managing everything. We obviously can't leave this up to random chance. Whatever that managing system is, it seems reasonable to call it intelligence because it does what we expect intelligence to do, which is to navigate around obstacles to reach a goal space. When COVID-19 broke out, you didn't rush to the store to buy a bunch of pencils because you knew there would be pencils if you ever needed one. One hypothesis is that there might just be a set of instructions in the economy somewhere telling everyone what to do. But in economics, there is nothing analogous to DNA where we might think a set of instructions are hiding. There just are no instructions. So then we've got to really figure out how this is possible. The answer to that is the price system. Our primary focus for this talk is the basics of how the price system works. This is the famous supply and demand graph for pencils. All the numbers are made up just to illustrate. On our y-axis, we have the price of pencils. On the x-axis, we have the quantity of pencils. The graph is done backwards for historical reasons. The y-axis is our independent variable. The x-axis is our dependent variable. Quantity is a function of price. We've got our two curves here, our demand curve in red and our supply curve in green. First, the demand curve is the sum of everyone's plans to buy pencils at every price point. How this curve is constructed: if you could peer inside everyone's mind, you would say, at a price of $5, how many pencils does everyone plan to buy? The total amount is the market demand for pencils at $5. Then you iterate down. The demand curve slopes down because as the price goes down, people want to buy more. The price goes down and the quantity goes up as we shift out to the right. The supply curve is the sum of everyone's plans to sell pencils at every price. You take the total amount of pencils that everyone plans to sell at each price. The supply curve slopes up because as the price goes up, people naturally want to sell more pencils and make more money. There is one special point, the equilibrium point where supply and demand intersect. This is the point where everyone's plans match up. No one is trying to sell a pencil that can't be sold or buy a pencil that can't be bought. The quantity of supply and the quantity of demand are the same at this equilibrium point. This simple graph illustrates the basic concept of how you could have a system that could build a pencil and how you could have a system that's a collective intelligence. If you can arrange a system so that everyone's plans are mutually consistent, or even in the limit if everyone has the same plan, that's the key to large-scale coordination. Everyone is going to be naturally working together rather than conflicting or at odds with each other. One final thing to notice is that the reasons behind the plans don't matter. Your goals stay private. Why you might want to buy a pencil or why you might want to sell a pencil, none of that matters. Maybe you want to write something down, maybe you're making pencil art, it doesn't matter. This system works for any kind of plans, any kind of goals. It doesn't need that information. Given your plans, which only you need to know, it's able to coordinate everyone regardless.
[08:37] Benjamin Lyons: This is the basic concept of how prices enable large scale coordination. But the question is, why would an individual be interested in participating in this system? It's good for the collective. Is it good for the individual? And the answer is yes, because prices help you plan. To plan effectively, you need to know how available one resource is relative to another. If you want to eat fruit and you're thinking, do I want an apple or an orange, you need to know how available apples are relative to oranges. If they're very available, they're probably easier to get, and you'll have to give up less time and less energy to acquire one. Maybe you like the other one more, but you need to know how the trade-off works. Economists say that when a resource is less available, it is more scarce, and the scarcity of one resource relative to another is called relative scarcity. And the brilliant thing about prices is that prices track relative scarcity. When people plan to buy more of something, which is demand, or plan to sell less of that, which is supply, the resource becomes less available and therefore more scarce. By planning on price, you can plan based on scarcity. It would be hard for you as an individual to figure out how available apples are relative to oranges, but it's very easy for you to figure out how costly they are relative to each other, because you can just go to the store and see that. This graph shows our original equilibrium point where the supply and demand curve intersect. If demand goes up, it now intersects here, so the price is higher. Or if supply goes down, it intersects here, so again, the price is higher. When things become more scarce because people are trying to use them more or make them less available, the price goes up. Similarly, if demand were to go down or supply were to go up, that would make it more available and therefore cheaper. So prices naturally track relative scarcity, which is what you need to know. The elegance of this system is that you can plan based on your budget. You have an amount of money, say $100. You can naturally combine that budget with the existing set of prices that you can easily observe at the store to tell you every possible bundle of resources you can purchase. At that point, it's up to you and your competencies to pick the best of all available combinations of resources that you want to purchase, which is what intelligence is all about. You have a lot of alternative ways of achieving a goal state, and you're trying to pick the best one. The price system makes that very easy. You just say, how much money do I have? How much does everything cost? That really simplifies the problem. If equilibrium prices actually do reflect relative scarcities, then that allows you as an individual to confidently choose your favorite available bundle of resources, secure in the knowledge that your plan will not conflict with other people's plans, as long as the prices are in fact accurate. We'll talk about what happens when they're not. What happens if plans are wrong? We know that the price should be at this equilibrium point where they intersect so that what everyone wants to do is consistent with what everyone else wants to do. What happens if the price is too high? Let's say it's at $4 here. The demand for pencils: people plan to buy about 20 of them, and people plan to sell about 30 of them. Sellers are planning to sell all these pencils and use that money to do other things, and no one's going to buy these pencils because they're too expensive. People can't accomplish their plans. Similarly, if the price is too low at say $2, sellers want to sell maybe only 16, but buyers want to buy about 35. This creates what economists call a shortage. Everyone wants to buy a pencil and there just aren't enough for everyone to accomplish their plan. The system's in conflict, people's plans aren't consistent with each other, and that makes it difficult to achieve allostasis. This matters in real life. This is an image of the famous 1970s gas lines. This is hopefully a familiar image to some people.
[12:55] Benjamin Lyons: You can see what this issue was. In the 1970s, the price of gasoline was kept too low. There were some price controls on it. People were seeing low prices and planning to buy a lot of gas, but people weren't planning to sell a lot of gas. You had these issues where people were waiting in line for hours to buy gas. Often they couldn't even buy gas because the place had run out. When you get the price wrong, you really do have this issue where people's plans are in conflict with each other. Not everyone can accomplish their goals simultaneously. The system really gets jammed up in a very physical way because people are trying to do things that don't make sense based on what other people are trying to do. We know why it's useful for prices to reach equilibrium, but why would they reach equilibrium? The answer is simply because of people's self-interest. If you're a seller and you're trying to sell pencils, but the price is too high, you actually want to lower the price so that you can make at least some money selling the pencils. You make zero money selling the pencils for too high of a price. Sellers want to lower the price. Buyers, if the price is too low, actually want to raise the price, because if you're offering too low of a price, the seller just won't sell to you. You want to raise the price so you can get something. Sellers are pushing the price down, buyers are pushing the price up, and the only logical stopping point is equilibrium. Just by people's natural self-interest, the price is going to be pushed to the equilibrium point. The basic idea here of what's going on is that your plans are functions of other people's plans. To accomplish your plans, you need to make your plan consistent with everyone else's plan, not because you care about them, but just because you're selfish. The price is the variable you can manipulate to make everyone's plans congruent with everyone else's. Everyone's just naturally trying to bring things to equilibrium. What happens if there's a new stress? This is the key section. Let's say that you are writing something down when all of a sudden your pencil breaks and you need to go buy a new one. If the system's in equilibrium, then as we saw, every pencil planned to be sold is already planned to be bought. There's not an easy way for you to get a pencil based on existing prices. What you need to do is bid up the price of pencils to outcompete other buyers. You need to convince a seller to sell you a pencil by raising that price. What we get is this diagram. This was the old demand. Because you want to buy more pencils now, there's new demand based on your new plan. There are two key facts about this. One is you can see the intersection point, which is the equilibrium price has moved up. The quantity of pencils in the market has also gone up as well from about 25 to about 30 or so. Price goes up and quantity goes up when you plan to buy more pencils. What this does is the stress sharing mechanic. You bid up the price of pencils just because it helps you, it gives you better chances of acquiring a pencil, but because you're connected to everyone else, the price system transmits your stress to them. An increase in price stresses out buyers, right? They see the higher price, they go, whoa, I was planning to buy a bunch of pencils. Now that's harder to do. I don't want to buy as many pencils, which is exactly what you need. Similarly, an increase in price excites sellers. They go, oh boy, I want to sell a bunch of pencils now and make money, which again, makes more pencils available for you to buy. Just by bidding up the price of pencils, you're communicating your stress to everybody without necessarily even realizing you're doing so. It just happens anyway. It looks like they're generously choosing to help you out. But in reality, what's going on is they just can't tell, is this stress mine or his? In fact, they don't even know about you. They're not even asking that question. They're just saying, I'm stressed out, so I got to behave differently. That takes care of you. This is the diagram, right?
[17:14] Benjamin Lyons: You're stressed out, price goes up, everyone's stressed out. If this diagram makes intuitive sense at this point, then we basically accomplished the primary goal. We can see that prices achieve mind-building and the expansion of perceptual fields, which is what we need for large-scale collective intelligence. When you bid up the price of a pencil, your stress becomes the buyer's stress and the reward you would receive from a pencil becomes the reward of sellers because they make more money. We've got stress sharing, we have reward sharing. This causes perceptual fields to get bigger. Everyone acts as if they know about your plans and they change their behavior to fit with what you're doing just because the price went up. This is basically what lets people make pencils. The task of creating a pencil is too large for any one individual on their own to understand. But because of the way that conceptual fields get expanded, everyone's able to see what they need to see to make the system work. One important thing is that the system depends on competition among the subunits to achieve this cooperative outcome. Suppose a pencil seller tries to cheat the system and pretend that there's a new stress in the economy and raise the price of pencils to make more money. This would immediately incentivize all of the other pencil sellers to undercut them and steal their customers. Lying and trying to cheat the system in some way immediately gets punished by your competition, who are going to take advantage of your poor decision to make more money. My hypothesis is that all collective intelligences work this way. We're all bound together by cognitive glue, such as bioelectricity, that are in some way mathematically analogous to the price system. I think there's one way to do this. It's this basic abstract economic system. What justifies calling this a collective intelligence is a shared model perspective. Prices are a set of publicly available, electronically transmitted numbers that condense a huge amount of decentralized information, which means they constitute a model of relative scarcities. It's a shared model because we all see the same set of prices. If we think about an active inference internal model, it seems like you can associate a self to an internal model. If everyone has the same model, then everyone's part of the same self. It's a collective intelligence. Because everyone's sharing the same model, everyone perceives the predictions, in an active-inference sense, of the expected perceptions and planned behaviors of everyone else, which makes it very easy to figure out how to achieve your goals while not interfering with, and therefore being interfered with by, other people's goals. To return to the original question of how we're able to build a pencil, this is the basic picture of how that works. All of the facts necessary to coordinate people across space and time are things that some individual is aware of. Maybe it's someone, a worker on the ground or a CEO high up in the company, but everyone has some piece of the puzzle and they incorporate that piece of the puzzle into their plans because that's what people do. Then those plans, as we've seen through supply and demand, are condensed into prices such that updates to relevant information cause updates to plans, which cause updates to prices. Information causes plans, causes prices, and then changes in information yield changes in plans, which make changes in prices. All people have to do, rather than having to think, "How do I make my information congruent with Mike's or how do I make my plan congruent with his?" is make their plan congruent with prices, which is easy to do because they know what the prices are and how much money they have, which then unwittingly makes their behavior congruent with others' behavior as well. The simple signaling system allows everyone, in the US, Mexico, Brazil, China, et cetera, to coordinate with each other without realizing they're doing so. Fascinatingly, this causes a massive scaling of plans and goals. An ordinary human on their own plans to acquire food and shelter so they don't get eaten. When you combine people into the system, people start planning to create millions of commercial-grade pencils every year. This system really scales up people's goals by making it highly attainable to achieve those goals. One interesting note is that if you can prove that this is how organisms work, then you would prove that cells are agents, because one obvious way to think about agents is that agents are entities that take information and turn that information into plans. One important point: we can see how helpful prices are to us. Why are they so helpful to us? The answer, of course, is their own self-interest. They're trying to help themselves. The price system has two goals. It's to be an accurate model and it's to be a rational model of the relative scarcities of all goods. I suspect that's the goal of all models: to be rational and accurate. A reminder: scarcity is the availability of a good relative to people's plans. The more people plan to use it, the more scarce it is. Relative scarcity is the scarcity of one good compared to the scarcity of another good.
[21:32] Benjamin Lyons: What accuracy means in this context is that if apples are priced twice as high as oranges. If apples cost $2 and oranges cost $1, then accuracy means that apples are actually twice as scarce as oranges. The prices really are tracking the relative scarcities. Rationality means rationality in a Bayesian sense: prices should update as a random walk, not moving in a predictable direction. The way we know prices have these goals is because they pay people to improve its accuracy and rationality. If you're a stock trader or a commodity trader and you can make the system more accurate or more rational, you can become very wealthy. The price system effectively pays us to make it more accurate and rational, which is exactly what we need in turn to enable large-scale coordination. This takes us to a primary conclusion; then we'll have a couple of miscellaneous slides after. What I think are the basic things you need to have collective intelligence; this should be a universal principle. You need some kind of system that collects all relevant information and condenses it into a single shared set of mathematically useful objects, like prices, which are numbers. Numbers are very convenient. You want these objects to interact with the subunits of the collective, such as humans or cells, such that they persuade people to behave in the intended way, which would then allow you to manipulate these prices or other objects to get the system to do anything it can conceivably do. You need the system to incorporate new and updated information and have that transmit rapidly and accurately through the system, which is what the stock market does. You also need a process for inducing honesty and accuracy in communicating information such as competition. The result of all this needs to be a shared model that is an accurate and rational model of relative scarcities — relative scarcities being what sums up the congruences and conflicts between everyone's plans. If everyone can see something that models the relative scarcities, everyone's going to be able to act in ways that are consistent with how everyone else is acting. This, I think, is the basic details any collective intelligence needs to have and is going to have. A couple of miscellaneous points. I'm arguing prices are analogous to bioelectricity, which means they should be the morphological code of the economy, and indeed they are. This is an example I got off Twitter. Taxes are an artificial way of manipulating the price system. You can raise the price above what it would otherwise be. If you tax windows, you don't get many windows. If you tax roofs, you don't get many roofs. If you tax frontage, you don't get a lot of frontage. If you tax finished buildings, you don't get a lot of finished buildings. Economists believe that, in principle, you can get the economy to do anything it could conceivably do just by manipulating prices up and down, which is convenient because that leaves the problem solving up to the subunits. A high price says supply more of this and buy less of this, but how to supply more of it and how to live with less of it is left up to the subunits and their competencies. If you want less carbon dioxide in the air, you can put a tax on carbon dioxide. The challenge of how to adjust to that new price is left up to people and businesses. Managing the system in principle is very easy as long as you can move prices up and down. The final point here is: what if there's a flaw in the system? What if there's a missing link? The system works very well as long as everything relevant is being tracked, accounted for, and condensed into the price system. But what if something doesn't get into the price system for whatever reason? That's called an externality. An example is carbon dioxide, which isn't priced properly. The effects aren't included in the price. People end up doing things they shouldn't do, like overconsuming and buying SUVs. I believe that externality is analogous to cancer; cancer is an example of externality. A firm producing excess pollution does so because that activity isn't tied to the rest of the economy via the price system. It treats the rest of the economy like the external environment, dumping waste into it rather than taking care of it. I believe externality explains economic tumors. Relatedly, externality causes developmental disorders. If there's something weird going on in the firm that doesn't make sense, or if the economy is creating goods and services that it shouldn't create, that's going to be because of an externality. Externality is my thing. If externality interests you, check out my paper on my website, "The Problem of Externality" (2024), for more information about how that works. That takes me to the end. Thank you for sitting through this. Here's my contact information. If this theme interests you, I have an essay on my website, "Prices as Memories," which goes into more detail. Thanks very much. I hope that was interesting and educational. I'll turn it over to you now. What do you want to discuss?
[25:52] Michael Levin: Thanks. In this system, which is consistent with our multi-scale competency architectures, the individual subunits that are acting on the price, buying and selling, are rational agents, right? So they have a little bit of intelligence themselves that allows them to execute these policies based on the price. Can we say what you think about what is the minimal set of things that they need to know and be able to do for this all to work? How smart do they actually need to be?
[26:34] Benjamin Lyons: I believe it's extremely minimal. The basic concept of rationality is a preference order. You have preferences, a set, and then rationality is a complete intransitive order. If you can just see these three dots connected by these lines, this is a rational preference order. This is an extremely simple structure that exists literally everywhere. If you have this, you have rationality. This is the minimal competence necessary. Part of why this is something I noticed very early in undergrad, and part of why I'm attracted to your work, is that rationality should be everywhere. It's in no way restricted to humans. At its core, it's a very simple, very basic thing.
[27:37] Michael Levin: So I can see. Can you just connect the dots a little bit more? Preference order is something that is common in biology. Why do you get from preference order to—draw the loop for me. So the price: you have a sensor that can sense the current price. You have some kind of inner state that says how much do I need. Can we just sketch out what's the minimal agent here that you need to participate in the system?
[28:07] Benjamin Lyons: You would have preferences over every possible state of the world, but at least when we're considering a standard shopping example, preferences over different bundles of goods. You have an amount of money; let's say you have $10, apples are $1 and oranges are $2. You could buy 10 apples and zero oranges, or 4 apples and three oranges, et cetera. The basics of economics is you arrange all of those possible bundles into an order based on preference, with your most preferred at the top all the way down, then you just buy your most preferred options. That's the basic competence. Can you take your options, such as your shopping bundles, and arrange them in a list where you can then select your favorite one on that list? And if you have any ability to do that, I'm working on a couple of papers arguing that power does exist to a much greater degree than people realize. I'm working on some stuff about how human beings are diverse cognition. In human beings themselves, cognition does not work the way people think it does. Economics exists in human beings in places where people don't realize. You don't necessarily need to see people making shopping lists; as long as there's some ability to create an ordering over things that map to options, you're good to go.
[29:26] Michael Levin: But presumably there's some, if you were trying to implement this, either somebody makes an offer, I want to sell you this for this price, and you can compare that with what's the calculus that you then do to decide what you're actually willing to pay.
[29:45] Benjamin Lyons: There are two things. The basic calculus is you're really comparing it to your opportunity cost. You're comparing it to your other best option and saying, is this worth it? If I do this, I don't do something else. Is that worth doing? That's what you're comparing: your two different plans, your two different predictions. The price itself is all relative. This is actually a key point. Prices just track relative scarcity. The only thing that matters is the relative prices. If apples are $2 and oranges cost $1, or if apples are $4 and oranges cost $2, it's the same situation. The price is almost secondary in a way. The only thing that matters is that it should be in some way a mirror of the relative scarcities, as in ordering.
[30:30] Michael Levin: In order to do what you just said, to say this over that, you have to have the ability to consider counterfactuals or store memories of other options.
[30:46] Benjamin Lyons: What I'm working on in these papers, do you know Paul Cisek's work at all?
[30:54] Michael Levin: Sure, yeah.
[30:55] Benjamin Lyons: He's got this action selection hypothesis that I think is really congruent with some recent findings in neuroscience. What's happening in the human brain, and I suspect similarly in other organisms, though I don't know, is when you're considering an option, you're constructing several options simultaneously. A motor behavior: I'm constructing different possible motor behaviors all at once in my brain and anticipating how they're going to plan out. Very quickly, the brain compares all these different options. They compete with each other, literally in your head and fight for dominance over the brain. Whichever one wins that competition is probably your best option. That gets implemented as a result of winning that competition. You do literally construct alternatives and have them compete. This is interesting because economists for a long time thought, since you don't feel yourself doing that, this is just an "as if" — you're not really doing this, but it's useful for modeling you. I think neuroscience shows that this is literally happening. I have a paper titled "Beyond As If" describing some of this stuff. You do construct alternatives in your head. I suspect cells do the same thing, although I don't know.
[32:09] Michael Levin: I'm thinking now we should take a look at how to cash out that kind of model in terms of, for example, transcriptional decisions, physiological decisions. What does it look like to be teeing up a bunch of different transcripts, gene expression change responses following a stimulus or a stressor and then having to choose one and the idea that they can compete with each other, it's interesting. We'll have to see what that would look like for cells. It's really interesting to me that the price here plays a number of different roles, which is much like what we've been thinking about for some of the things in the biology morphogenesis field. In a certain sense, it's the middle node of some kind of bow tie, because it encompasses in a very compressed representation a huge amount of other information, but also from that it has all these fan-out consequences that it encodes. There's a thing you pointed out about the shared model: that it's also, in code, a shared view of the world and it's binding everybody together into the same picture of the world. At the same time, it's a control knob because by impacting that price you get to have widely distributed consequences. As you point out at the end, it also has a perspective and a degree of agency because the price itself has certain interests in this, which I think is one of the most interesting things. That aspect — watching how things can run that spectrum and simultaneously play all of those roles — I think is very interesting. I wonder if we can say more. Could we say, in the system you've described, there are a number of layers you can try to imagine taking the perspective of. The easiest for us is the individual humans or the individual pieces that react to the price, but could we say what the world looks like from the perspective of the price itself? What is its world? When Randy Beer does the cognitive domain of a glider, when you're looking at unconventional agents and minimal patterns and asking what the world looks like from their perspective, why don't you start there? What do you think the world looks like if you're the price?
[34:42] Benjamin Lyons: Yes, I have given that some thought. I don't think I've come up with anything terribly useful. But what does occur to me is that what the price seems to naturally want to do is to preserve itself. It wants to preserve its relationship with other prices. It doesn't want to change in any kind of predictable way, just like an organism. It naturally achieves that because the equilibrium price incentivizes people not to mess with it except as new information comes in, which then should update the equilibrium price. So it does seem to me there is at least some kind of minimal model in which you can think of a price as something like an organism that's trying to maintain its state with respect to its environment. I have no idea what to do with that.
[35:42] Michael Levin: There's something going on with your Wi-Fi.
[35:48] Benjamin Lyons: Yeah, it might be a little unstable.
[35:52] Michael Levin: Sure. Seems better now. You're used to being caught up now.
[35:58] Benjamin Lyons: Did I cut off there? Did you get all that?
[35:59] Michael Levin: Is there any notion of, right now the price is a single scalar, and that's great for many things because it's an enormous compression into a simple thing. Is there any idea about what would happen if it were richer, a higher dimensional concept?
[36:25] Benjamin Lyons: People have talked about that. I'm not too familiar with any specific models. Right now getting everything incorporated into the scalars is hard enough. I think that's a rich area for exploration, but I don't have any specific ideas about it.
[36:42] Michael Levin: And similar to the question about the perspective of the price itself, what do you think, usually what in biology, what we have is we have the organism, and then we start asking questions about the cognitive glue and how the pieces actually add up. So in this case, we're going backwards because everybody believes that the individuals are agents, but then there's this cognitive glue that you've identified that binds them together. What do you think the collective is doing? In other words, looking at it from the perspective of the collective, having been bound together by this economic model, what are the goals, the differences, the competencies, what does the economy, is that what we're talking about. What do you think, what does the world look like from its perspective?
[37:31] Benjamin Lyons: I think the answer to that is analogous to the answer for an individual, which is less semantically clear than people believe. So, as an individual, when we talk about our goals, it's easy to say, oh, my goal is to do some economics thing. But in reality, that's not what's going on. My goal is to achieve allostasis by managing my metabolism and energy-regulating the ins and outs of what's going on in my body. I think that's also true of the price system. You could say, oh, it's trying to make pencils or trying to do this or that. It's trying to achieve allostasis, which it does by coordinating all the subunits, getting everything to work together. That is, in principle, its only goal, I believe.
[38:10] Michael Levin: Do you think that the active inference perspective, surprise minimization, would be applicable to that agent?
[38:19] Benjamin Lyons: Absolutely. I borrowed a lot from that perspective. It was moving away from economics to some neuroscience. I keep bringing up Lisa Feldman Barrett. I have to mention her name once every time we talk. Her work helped me make a lot of progress and prepared me to understand your work as well. I think that's big, and I wish more economists were aware of that.
[38:37] Michael Levin: Got it. Do you think that the system as a whole is trainable? In other words, is there a notion of memory here, some kind of historicity, where future behavior would be different based on a history of prior stimuli.
[38:58] Benjamin Lyons: Yes, it's tricky to give a clear answer to this because I think it gets so mixed up with the trainability of humans and firms. The economy is trainable, but that's not surprising to economists because people are trainable. There is a great deal of research about how firms learn and, based on previous instances, they'll change how they behave in the future. The trainability of the price system itself — it's conceivable that certain elasticities might be influenced by past behavior, but I don't have any specific ideas. I'd say it's tricky to disambiguate that from the trainability of people and people-made organizations such as businesses.
[39:43] Michael Levin: I think this might be an interesting thing for us to do some research on, because I think you can disambiguate it by analogy to what we did in gene regulatory networks. If you have a model of gene regulatory networks, which is just a set of coupled differential equations that say how each gene turns the other genes on and off, we showed six different kinds of learning, including Pavlovian conditioning in that model. Specifically, the pieces A were completely dumb, so they did not learn. Unlike in what you're talking about now, the nodes themselves have zero learning capacity. They all have is state; they just have whatever their current state is. They don't have any memory, they don't have any of that; that's all I know. Zero, zero memory. That's A. B, what you're also not allowed to do in these models is change the topology of the network. In other words, how strong are the connections? The typical things you would do to train a network like that, which is to adjust the synaptic weights and the connections between the nodes, you're not allowed to change. The network is completely fixed. No intelligence in the nodes, no changes of the topology. Yet, just from the dynamical systems behavior of that—of certain, not all; it's not a universal thing the way, for example, Stu Kaufman's Origins of Order network stuff is, which is widely applicable to random networks. Random networks don't do this as well as biological networks. I think evolution probably has selected for ones that are good at this, but nevertheless, there are plenty of networks that can learn without the pieces being completely dumb. I wonder if we had an in silico computational economic model where you clamp the behavior of the individual so they cannot learn. We can even clamp the relationships between them, and examine the behavior of the system after stimuli patterned in the way that behavioral scientists do for testing habituation, sensitization, Pavlovian conditioning. I wonder if we would see learning there. My conjecture is that we would. I think this is a very widespread capacity, and we could potentially show that if you clamp the behavior of the individuals.
[42:08] Benjamin Lyons: I'd be inclined to agree because I think these systems should be analogous. There's a Nobel Prize winner, Vernon Smith, who became known for doing economic experiments, which we usually can't do very well. I wonder if there are some analogies in his work. Maybe I'll try to read up a little bit on the gene regulatory stuff and see if I can find any connections to anything Vernon Smith has done. Maybe not, but it's worth exploring.
[42:32] Michael Levin: That's cool. As you look at the GRN stuff, I wonder if we can define, for example, in these networks, what would be the equivalent of the price, right? Could we say that by watching their behavior? How about this: do you think that watching is like a weird Turing test? If I showed a bunch of economists the behavior of a distributed system, could they say that this absolutely looks like an economy or not? Is that something, or can you make anything look like an economy, in which case we could actually do the mapping and figure out what is the price in this GRN?
[43:19] Benjamin Lyons: That's an interesting question. I never thought about that. I suspect that most economists would be cautious about drawing any conclusions that haven't been hard verified. I'd be very interested in seeing the results. That's worth exploring for sure.
[43:38] Michael Levin: One way to get started with that is if you identify for me what an economic data set looks like. If you were going to say to a colleague, "I'm going to show you the parameters of this economic system, what do they expect to see? What does the data look like? Are they looking at prices over time or distributed buys and sells?" I bet if we understood the format of that, I could give you a data set. We could do a blind thing where I wouldn't tell you what it really was. You can show it to people and they say, "a classic case of an economy of this type." Then we could find out that actually this is a bunch of cells communicating during axial determination or something. I think it would be interesting to use that as a way of finding these kinds of mappings.
[44:39] Benjamin Lyons: Yes, that makes a lot of sense. I will look into that. That's a really good idea.
[44:43] Michael Levin: Does everything you've said today assume perfect efficiency? Immediate, instantaneous transfer of information.
[45:01] Benjamin Lyons: No, you won't achieve perfect outcomes without it, but you don't need that to have something that's perfection within a body, but I still feel myself. And I believe the same is true of the economy. And as technology improves, we will improve the efficiency of it. Right now, a tremendous number of people are not terribly bound into the economy. Billions of people are in poverty. And so over time, that will increase greatly. In the first world, certainly the stock market is bound together tightly enough, but I think it's reasonable to think of it this way.
[45:39] Michael Levin: I'll tell you why I asked, but I'm also thinking about information efficiencies when you buy. I know nothing about economics, so this could be nonsense, but to me, it seems like when you're selling something you want to get rid of it; for that to work, somebody else needs to disagree with you and think it's a good idea to buy it. There are going to be differences in how informed people are, and somebody's going to make money because they're more informed than somebody else. People can make bets on what's going to happen with the economy. The reason I bring this up is the biological case. You got a set of cells, and the cells are now coupled by gap junctions. That's cognitive glue because the bioelectric signals are spreading through the tissue, and it's harder for any one cell to be an outlier in terms of its memories and its model of the world because it's sharing physiological parameters directly with its neighbors. In the real world, it takes time for information to propagate through that tissue. Much like the speed of light is a limit in physics that ties together space and time, there's going to be a constant like that in any tissue that tells you how fast information propagates and how fast signals and information get across the thing. You're going to have scenarios where for some amount of time there are regions. Here's your cellular network, your tissue: something happens over here to trigger it. There's going to be some time where the ones over here know about it and the ones back here don't yet. I wonder to what extent that delta has something to do with the thickness of the cognitive "now" moment. You're not instantaneous. In neuroscience they tell you there's about a couple of hundred milliseconds when things are propagated. Below that, you can't really answer questions: Did I see it? Was I aware of it? You can only answer those questions definitively from the perspective of the larger cognitive agent after the information has had a chance to propagate and all the subsystems have had their say. Eventually you can say yes, you did or you didn't. There's a temporal resolution below which you can't really slice yourself, that temporal thickness. I wonder if something like this is going on here, and if there's a helpful mapping. Economists know a lot about what happens when there are discrepancies in information in the economy and the ability to act. I wonder if some of those things could help us understand development or morphogenesis more broadly.
[48:32] Benjamin Lyons: I suspect so. Information asymmetry is a huge field of research where you're trading with someone who knows more than you about the good. You have to ask, why are they selling this to me? That can make it very challenging to create markets. You also have things like no-trade theorems where, if we're all trading in the stock market, potentially the only reason you're selling to me is because you think it's going down. Why would I buy it? There are instances where differences in information can make it very difficult to trade, even though the whole point of trading is to ultimately bring everyone's information into sync. Yes, there is a tremendous amount of research about that in both finance and the economics of information asymmetry. That's absolutely something that could be incorporated in some way.
[49:32] Michael Levin: Okay, yeah, the wi-fi is really...
[49:35] Benjamin Lyons: And maybe I cut out again.
[49:36] Michael Levin: The last thing I'll suggest, and this is something we can do in later work, is I think you've painted an amazing picture of the parallels between the two systems. And I think the other complementary thing we can do is look at the parallels of the disorders — the various ways that economics goes bad and has breakdowns of optimal functioning. I wonder if we can map that onto specific disorders in other spaces in living systems: developmental defects, cancer, failure to repair.
[50:19] Benjamin Lyons: I think cancer is an externality, and an externality is a generalization of all of those issues. Currently, economists have a very hard time talking about externality. It's hard to make it consistent with economics. I have a paper on my website talking about that issue, citing some of your work as to how we could try to make progress on that. I believe those parallels are definitely there, and I have been thinking about it. What we might need is biologists to figure out cancer and then use that so we can solve externality. The direction might flow that way.
[50:51] Michael Levin: I think that's very interesting to think about which direction, but the insights could flow. We've had some luck normalizing cancer, at least in the amphibian models. We're moving on to mammals and to human cells now, but we've had some really good luck in frog models. We could think about what that might be. Some research that we could do together would be to get hold of a computational model. There are computer models of economic systems that people can play with.
[51:27] Benjamin Lyons: Computational models have not been huge in economics because they have not been able to tell us anything surprising. What I've been thinking about, because you mentioned this before, is maybe they could make a comeback because it would be surprising to see economics in biology. I'm not familiar at all with computational economics, but I do believe that is a very useful path. Relatedly, one thing I've been thinking about is there's a field called economic geography, which explains why finance ends up in New York and tech ends up in San Francisco. That could be useful for explaining why your heart ends up here and your kidney ends up where it is. Computational models could be interesting in that context, but I currently don't know anything about them. I definitely have to do some studying.
[52:12] Michael Levin: Let me know what you turn up. I think it would be cool. I remember from 100 years ago a lemonade stand. You could easily—something as simple as that, I'm guessing it's too simple, but there must be things for students to play with and where if we rip the graphic interface off of it and use the underlying computations, we could test things like, can we induce and repair various disorders in a biology-focused way? Can we try the learning and the training? Can we use it to generate data sets to compare to other things and see if economists can pick out which one's the economy and which one's the morphogenetic signaling, all those things. I think we should look into it.
[52:53] Benjamin Lyons: Absolutely. I think so too.
[52:54] Michael Levin: Yeah, that'd be super cool.
