Recently had the pleasure to attend a seminar offered by the New England Complex Systems Institute, exploring the implications of advances in AI and their relevance to management and decision making. As would be expected given the specialization of the institute, the topics were evaluated in relation to complexity science, with one common theme being the interplay between questions of complexity and scales of address and how tools of renormalization can enable a reduction of relevant variables to make issues more tractable. The question of complexity characterization is an important one for purposes of AI planning, as these domains can help illustrate where machine learning outperform humans, where human and machines working together have the edge, where human specialists may still be appropriate, and at highest levels of complexity where teams of humans are the dominant paradigm.
The standout part of the program turned out to be the closing session when Yaneer Bar-Yam, the founder of NECSI, opened the floor to audience questions and then proceeded to walk through them one by one in an epic 3 hour marathon of high energy free form erudition (edging out earlier presentations by Alex Pentland or Nassim Taleb for highlight of the week). The question topics covered a lot of ground, some that were relevant to attendee industries, some that were revisiting portions of the seminar, and several that were from left field but in response were still somehow tied into themes of complexity. Just from assembling the list of audience questions it became apparent that we in attendance were in for something special, and a few of us took out our phones to record the talk like we were in the front row at a Grateful Dead concert (I bet there may even be a bootleg tape out there for those that know the right password).
The physicist Richard Feynman authored a whole series of books without sitting down at a keyboard — they were in many cases direct transcripts of his lectures and interviews. It was one of his many impressive traits that he spoke in a coherent prose that directly translated to the printed page. Bar-Yam has a similar talent in his speech, and I believe a full transcript of this talk could carry the same weight of hearing this in person. However there was so much ground covered here that in the interest of brevity I will in some cases base this post instead on my notes taken during the lecture, which are mostly direct quotes, but will in some instances condense the full speech into key points and lessons that I may have found the most relevant or significant. At times I’ll use the abbreviations “A:” here to refer to audience comments or “Y:” when transitioning back to Yaneer. Each passage will start with the audience’s question statement followed by Bar-Yam’s address. The question statements were abbreviated in my notes so may not match exactly the original phrasing. Oh and since I followed this seminar with an afternoon visit to a few Boston museums (MIT Museum and the incredible Museum of Fine Arts), I’ll throw in a few snapshots of their collection for color. And without further ado.
1. Complexity profile of the national air-traffic control system
This is the first thing to think through: at different scales, what are the things one has to pay attention to — which are dependent and independent variables at a particular scale? Start at the large scale (the global scale), then work down to maybe scales on order of 1,000km, followed down to individual flights. Time scales or spacial scales are all relevant. At large scales you aggregate or average up to that scale, then as you approach small scales ask what additional things do you need to consider. Once one has an understanding about a behavior being described, then you can start asking questions about it. We know air-traffic control is not optimal at all scales, even at the largest scales due to things like weather, i.e. what happens during a winter storm or hurricane. If you to ask me how to start improving the system, my question would be how to improve it at that largest scale. What is the central relevant parameter to the inability of the system to respond at the large scale, the big events. That will effect the system the most.
2. Mapping Models at Different Scales
The most important concept for multiple scales is nearby scales, how do you translate from one scale to a nearby scale, how to relate the two models — the crux of the methodology. When you take variables at one scale and then combine them together to get models at a larger scale, this helps you determine which of those variables will survive, a key concept. Ken Wilson, the physicist, in 1970 provided this renormalization question, the foundation of this approach. Take variables of the system and ask what happens when I combine them together. Every behavior of a system is in a space of possibilities, so you have to combine the spaces of possibilities in the right way, ignoring the finer differences. If things are independent they average like statistics, the dependancies are what survive. The idea of space of possibilities is a hard idea, writing it down mathematically is tricky. It is taught in physics as statistical physics and requires a certain sophistication of abstraction that is a challenge to teach.
3. How to detect or predict phase changes
Detecting is an empirical statement — in the empirical context detection is from systems beginning to fluctuate between phases. Those fluctuations are the fat tailed events of the coupling between the behaviors that make the behavior not describable by statistics, that’s how you detect a phase change. You predict by understanding the fluctuations empirically. By looking at different scales you figure out which parameters control the couplings in the system, then look at how they’re changing, which enables recognition when you’re approaching a transition. You need to map out the space of possibilities in macro space and look for boundaries in a “phase diagram” which are transition points in the macro space of possibilities — the meta view of macro space of possibilities in the tail of the distribution. Temperature and Pressure are relevant parameters in the example of a water phase diagram for example. In general there are two types of relevant parameters (this is really important), one is the external forces — if you change the external conditions, which are the ones that effect the behavior of the system. These are the ones that everyone used to primarily study. The second relevant parameter is internal interactions. These are the ones that give rise to self-organizing behavior. It is the combination of external forces and internal interactions that give us what we need to know. It turns out that there is usually only one relevant parameter for internal parameters, that’s what makes this easy. Geometry is also important, the space and dimensionality of the system. Internal interactions, external forces, and dimensionality are all important.
The point is the purpose of renormalization is to figure out which are relevant parameters, then in that space figure out where phase transition occurs, then figure out where you might be in respect to a phase transition.
4. Why write papers outside of academia?
We are all part of a global sharing community of which the academic community is only a small part. The global community is interested in new ideas, so it’s worth writing and putting it out there (on sites such as Medium). The global community is becoming a global collective organism, everyone participating in that dialogue is participating in this system.
5. Communication across scales
The aggregation process is not communication across scales, it is understanding how large scale behavior arrises. In communication across scales we are discussing how the behavior of an individual can affect the global system. There are built into certain systems amplification and dissipation, the process of creating cascades can thus be built into the structure of a system due to certain instabilities. If one individual can effect the system as a whole then the system is unstable with respect to that component. For example in the butterfly effect there are lots of butterflies, every flap of a wing will not create a hurricane, it’s only possible for a very very few of those flaps. Weather has limited sensitivity to finer scales. It’s not an easy subject. People talk about weather as a chaotic system, but when Lorenz wrote down his equations they were low dimensional systems which means only one or two variables — the space of possibilities is very flat, so fine scales influence the large scale. But in the real world there are only very few circumstances like this. There is almost no work being done on this. I gave a lecture on this some time ago explaining the problem of predictability — people don’t understand that there is only limited sensitivity to finer scales’ degrees of freedom except for very limited circumstances. It’s not an easy subject technically.
6. Template for developing questions
First let me say that my brain does its best work when it’s asleep. I think about a problem, go to sleep, wake up and know the answer — but the answer is “what is the question?” Over the years I’ve worked hard at explaining what I’ve learned, and in the process have teased apart the process. I use the frameworks we’ve discussed, the paradigm of a model in how systems work — interacting particles create collective behavior which is dissimilar across scales. If I understand how that system aggregates across scales then I’ll know what’s important and how to frame that question. For questions not describable at large scale I’ll look at them differently, there are different frameworks for different scales. For example, the complexity of engineering cannot be described at large scales. The framework of complexity profiles is an anchor for how to describe a system, with different tools available for different scales.
7. AI in decision-making complexity profiles
We talked earlier about how consciousness has components of attractor networks and feedforward networks. The ability to hold two different things in a mind at the same time and perceive the relationships or differences between them helps in thinking about collective behavior — the juxtaposition of two possibilities is a powerful frame. But turning back to AI in the complexity profile, it is relevant because you can’t always assume complexity profiles capture the right info about the behavior of a system. The information that’s the basis of a complexity profile doesn’t capture the difference between intricate patterns when there’s only one of it, for example a sequence like the Fibonacci sequence. A implies B implies C etc is an inferential process which carries the same information content as repeating digits but requires steps of inference to reach depths. There’s a term related to logic, ‘logical depth’ of a particular quantity, which is in some sense orthogonal to information. Computers are contributing logical depth that humans don’t have and thus adding a separate dimension to the capability of the human and AI system. AI will add additional depth of ability to go deeper into a sequence — like calculating the steps of a Fibonacci sequence.
8. Games and complex systems
Games serve as a way of learning skills in a relatively low consequence environment, which is true about play in general. Play is an important thing. Games are an environment to build teamwork with safe consequences such as when you want to help people to work together well. From scientific perspective, games are a powerful way to learn abut complex systems — I wrote a paper about it, and is also discussed in the book Making Things Work.
9. How to measure human intelligence
Different people are capable of different things. Exams are usually the wrong way of measuring intelligence, we have to ask ourselves “what is it we really want to know?” Most of what we currently talk about in human intelligence is measuring an autonomous capability. Why do we mostly measure capabilities of individuals in isolation? That doesn’t make sense in the world of today. It’s better to measure if someone knows how to ask a question than their knowing how to answer one — after all people now are usually a google search away if they know what to look for. Another valuable measure is less what people know and more what type of things do people forget. Those things that you relate and link, those things that you don’t, both are important. Every strength is a weakness, every weakness a strength. The process of selectivity is in both what you know and what you don’t. In a complex environment, it is always true that there are too many things for an individual to know everything, things you see verses things you are blind to limits what you can know, limits what possibilities you can distinguish. The blindnesses dictate what we can know and vice versa. EQ is also important, and is relevant to some kinds of success more so than what’s measured on the SAT. That being said, I have a friend with low EQ but who is still highly respected in his organization because of other skills. Integration and communications determine the success of a group. We need to enable each individual to do what they’re good at in a context with other people who are good at other things.
10. Risks of general AI
Is general AI a risk to humanity? If we create a general AI and give it control of decisions it will then outsmart us; is this true or false? If we tease apart this scenario we see that it misframes the nature of risk, misframes the nature of AI, it misframes even the framework of the question. Framing of this question is important. The first problem is that complexity is not something that AI does well, people do it much better. If we compete with AI in a complex environment we will win. This is something you can extrapolate into the future very far — farther than any of the other problems we will run into. They will win in a deep logic context, say playing Go. However it is not possible to translate dealing with deep logic to a complex domain. The technical origins of our ability to deal with complexity has to do with non-universal, parallel, and random architectures. There is thus a fundamental reason why we are good at this and computers are not. To the extent that computers are or will be what they are today, this is not a competition. People can do logic, but logic is a superficial aspect of human thinking, not a fundamental part of how we think. Logic requires assumptions. Human intelligence is not “one thing”, the key is to apply the right strategies to the right domains. Thus for the question of can we create a general purpose AI, the answer is no — as soon as you make it more close to what humans are capable of doing, it is now becoming special purpose. Even for the pattern recognition tasks we are giving it, it is now special purpose.
Will silicon be as good as biology, a system with millions of years of evolved structures? Well first you would have to embed that structure in silicon. If you embed all of that structure, what is it you’re doing? You’re replicating the structure, so can you replicate then one human being? Regarding the Turing test, I believe that more challenging than fooling a human is to what extent can you replicate particular human beings, a very different question. You can pass the Turing test of fooling a human trivially. You cannot make general AI before you have the ability to actually replicate an individual human being. (A: “the Bar-Yam test.”)
We are becoming a global collective with higher complexity capabilities. With respect to risks, the right question is not whether AI can replicate general intelligence at the level of human beings, but whether it can replicate intelligence at the level of collectives, of society. Our current society is stupid in many ways but smart in many others — look at the products we produce, at the growth of our economy. But for risks, we can kill ourselves off with lots of things, it doesn’t have to be with AI — things with global consequences from the connectivity of world. It would be dumb to create a program that has the capability to perform actions that can destroy humanity (as an example a program in control of the release of biological materials). Systems crash. There’s an example in the markets of Knight Capital in August 2012, they lost $440M in 45 minutes from a bug in their trading software. The vulnerability of the system is the issue. We need to protect ourselves from risks of our creations, regardless of their form, focussing only on AI misses the point of risks. We need to continually ask what are the risks we are taking, where is the global system vulnerable, and address them both as individuals and collectives. The potential failure modes of AI (e.g. as demonstrated by the paper clip factory thought experiment) are relevant and important, but not the only risks that need to be paid attention to.
11. Incentives in organizations
Skin in the Game is part of the incentive idea. While it does work under some circumstances it clearly is not generalizable. This is one of the key complex system challenges. Organizations aggregate actions into collective actions. Many people think the purpose of organizations is to align, I don’t agree with alignment of goals. For example if you’re designing security systems you also want people trying to find ways to break it. But you do want an organizational structure to align rewards to incentivize people to facilitate what you want them to do. To the extent that an individual naturally does what the organization needs, incentives become less important. In creation of teams where individuals are doing what they’re good at, this is its own reward. People want to do what they’re good at. An organization may have to worry less abut incentives when they can succeed at matching people to what they’re good at. Another part of this question worth note involves evolutionary dynamics — people want their group to be successful, which creates incentives for people to do things well at the larger scale as well.
12. Quantifying emotional or psychological weights in people
We just wrote a paper about understanding personality, revisiting the development that was done for the academic model of personality (OCEAN). This is a piece of a process that needs to be done to understand personality, which is only partly related to emotions. Emotions do have a space of possibility, so you can quantify them by mapping them. A: For generational emotional trauma, how does that affect behavior of groups of people on a societal scale? Y: Economics says people are rational, meaning emotions are not important, although how you value things can be based upon emotions with a hidden assumption that ultimately everyone values things essentially the same way so that the assumptions of economics work — which is not true.
If you look at the collective behavior of civilization, you see that values trump economics in describing the collective behavior of civilization. Different parts of civilization are manifestly different in their choices in a way not describable by economics. The emotional, value-driven decisions of people (emotions and values are linked) is a relevant parameter at the global scale — without understanding that we don’t understand the world. We have unpublished work developing frameworks for talking about these issues. The paper we published is about individual issues which is a first step. A: Are emotional wounds (e.g. slavery, war) driving the behavior of people? Y: What may appear to be wounds could actually be functional in the context of collective behavior.
13. How to improve the success of systems engineering
The important thing is to understand, recognize, and to some extent quantify systems complexity. When starting a project, if you don’t understand the inherent complexity of a project then you can’t understand which approach to use in systems engineering. In the context of every problem, knowing relevance of complexity is a relevant parameter in deciding on how to do things that may not be fully appreciated since it wasn’t as relevant when the principles of systems engineering were developed.
14. Social impact of AI economic displacements
What will happen when AI replaces say 80% of the workforce? The economics answer is “don’t worry it will work out” because the system is self-optimizing. Automation in agriculture was linked to industrial revolution, but freed up workers for industrial factories — do we have some similar process for automation on a reasonable time scale? We don’t know. But given the understanding of economic flows we can create a system that works regardless of which is true. What we have to do is understand that some transfer between owners and workers is necessary for the economy to be balanced. If we start that knob and watch it, we may end up in a Star Trek scenario (where everybody has stuff and can buy stuff and so on) or a working scenario where we are doing work that is paid for by society — with achieving these states derived from adjusting the knob. How disruptive this is will depend upon policy, it may require a regulatory system, or on the other hand an economic picture may work fine with more or less disruption, this is all still a question. It is clear from our conversations that eventually things may be fine, but in transitions what we choose to do will effect the amount of suffering, so it leaves us an important choice.
15. What shows you have enough data to validate a model
For the complexity profile picture, if we’re looking at something in the tail, although it may require the right kind of information it often doesn’t require a lot of information because the number of parameters is small in that space. One of the challenges is a lot of complex systems science is done by creating models that can’t be validated due to the number of parameters — there’s a famous (John von Neumann) quote, “With six parameters I can fit an elephant, and with 7 I can wag his tail.” It only takes a few parameters (on the order of 10) to make a model which is unvalidatable because of the need for large amounts of data which is combinatorial in figuring out the model. But if you’re in the tail then you can create a model with few parameters. If you want to validate a model, create one with very few parameters.
16. Traits that are newly relevant at scale
What are the things that are relevant? External forces are relevant, but internal self driving factors are also relevant. Self organizing factors are those cascades and coupled events that we have been talking about. Strength of internal ties are the most important trait often missed (not an individual trait, but a relational one), which helps determine how behaviors propagate.
17. How to go about analyzing a problem in a timeline
The first step is to understand and define the system, then to define the elements, and then to think about the interactions between them. Once you have done that, then you’ll understand the kinds of models to talk about, and the idea is to connect models to obvious behaviors of a system. It will be pretty obvious what are the largest collective behaviors of most systems, that leads to a process of building a hypothesis. This takes time to think about. Typically a project will get to a point for some understanding based on these factors after say weeks, then the big thing is to figure out whether you’re right. Real projects take on the order of years, where most of the work is a recursive process to make sure you get it right. You have to compare simple issues with real data. You’ll create a model and then compare it to real data, make sure you got the right data, then make sure you got the right fitting, then you’ll write it up, and in the writing it up the process of explaining will take 80% of the time as you discover new things to consider, and then do it all over again. That being said, it may be possible to tell people an answer in a day, but getting to a point you can explain it or to be published may still take this much time.
18. Walking and thinking
People think brains are where they think, but our nervous system is a single thing which includes all of the motions you make. These things are linked. You walk and that contributes to part of your thinking. Yes legs are somewhat separated from your thinking, some are more so than others influenced by other parts of the body.
19. Ways to feel better and think better
Human lifestyles have been developed over millennia. I think it is wise to adopt a historically traditional way of dealing with the basic needs of a person — eating, drinking, food, exercise. One of the challenges is that the needs of the environment are changing which creates changes in adaptability to lifestyle. For example we have heated homes. Physical activities are not required to earn a living. These create coupled changes in other aspects of life. There are two challenges I would point to: one is sleep — it’s an under appreciated necessity. It’s important to understand that sleep is integral to human function. Not sleeping for two nights can put you in danger. A second thing is food. There is very little food culture in this country due to so many immigrants so there is no consistency in diets. Food culture requires internal consistency. It is worthwhile to have a traditional food culture (perhaps adapted for less need for energy expenditure), but adhering to a well defined and consistent food culture that makes sense to you is important for good living. Presumably your genetics are tied to the culture you came from, the need for vitamin C is a genetically determined thing for instance. There is also a conflict between homeostasis and variability. The body protects itself by providing a uniform and consistent environment because it cannot deal with the complexity of variations that are possible — at the same time an overly homogeneous environment is not good either. Both structure and some variability are important. For example there is a power law distribution of heartbeat intervals variability, if you lose that variability you lose some mortality in new circumstances. Just as in the fluctuation–dissipation theorem in physics, the rest state variability is describing the dynamic variability as well — or in other words if you don’t practice doing various things you can’t do various things when they’re needed.
20. How can the greater New England Complex Systems Institute (NECSI) community advance the mission
- Within our various organizations, in a lot of people’s minds there is resistance to change, we need to start producing research used for practical problems.
- It has to reach a critical mass, which will be based on connections, interpersonal relationships, getting enough of a network for complexity out there for sustained growth.
- We’ve discussed the approaches to solving problems, the steps of framing. Would like to see a NECSI publication along these lines for using the renormalization techniques and the applicable tools available for different system scales and complexity domains.
- Most people don’t understand these ways of thinking, one fundamental part will be facilitating understanding of interconnectedness, including values, behaviors, and structures we live within.
- Someone who hasn’t spent 15 years in academia may understand these things better than someone who has spent their life in the ivory tower. (People in academia have been trained to think statistically and so will have a harder time making this shift)
- A lot of us here have more in common with our interests in the topic than with respect to our fields of application. As a result what everyone is sharing about NECSI is on the left side of the complexity profile, with a huge space of possibilities of what we think it means or how we think we might use it, which makes it hard for us to correlate or have cascading effects. The challenge and opportunity is to find more restricted domains or more general tooling of how to apply these frameworks.
- We need to ask the question “What does NECSI want to achieve, and how can that come into play?” And then how can we design an organization that people want to be a part of and contribute to. It’s about getting he NECSI community to see it as a resource for what they want to do for the world.
- People in power like it when you use their language, not new vocabulary from these tools.
Y: NECSI wants to be a research and educational institute, but also what I would call a “meaning organization” to help people understand the world that they live in. There are things we can do at scale, but also things that each can do individually. Everyone in this room is good at something else, please tell us what you can contribute to the collective effort.
21. Biggest challenge as a species
The world is becoming more integrated, but at the same time more in conflict in dimensions like values, culture, economics, social systems, or meaning. The biggest challenge that we face is that when you go through transitions you have fluctuations at all scales where statistics breaks down. These variations in variability appear at all scales. The challenges faced at global scale is the same challenge faced individually, which is the challenge to become diverse teams. Groups must coalesce in some sense into functionally collaborative relationships. The same is true at the individual scale — multiple individuals with different capabilities or values have to work together in some collaborative fashion in order to enable us to gain benefit of these diverse views. I fundamentally disagrees with the perspective that we all should have the same behavior or views. One view shouldn’t impose its view on others. Importantly, there is still some universal level of ethics, but that is a limited thing relative to the nature of human behavior. Right now we are conflicted from people trying to impose views on others at multiple scales.
The challenge for the species is to make the transition. This transition will happen, and almost surely successfully, because it is driven by complexity (and other reasons difficult to discuss) — but the point is that the transition will create huge amounts of disruption. The challenge isn’t to make the transition, but to mitigate the pain of that transition’s disruption. We can contribute by communicating the nature of these problems so they can appreciate the importance of individual differences, the importance of each person finding the roll that they play successfully in their collaborations so they contribute to the success of civilization as a whole. There’s an old statement: every person should carry a piece of paper in their pocket, on one side written “I am made in the image of God,” on other side “I am made of dust.” Every single individual is extremely capable at something, but at the same time has to be extremely modest relative to the capabilities of others, even those you can’t understand. This argument holds at multiple scales.
22. Acceleration of change
The pace or dynamics of variability of change is itself complex. There’s also the dynamics of change of the structure, which also contributes to the complexity. The rate of change is complexity. I do expect after the transition things will settle down in some sense.
23. How to move to other forms of education
There are two ways to change a system, either evolution or revolution. Given the nature of the system I expect that revolution will be the method. There will be local changes that will demonstrate what works. A piece of the education (or healthcare) system will adopt some new system which will perform so much better than everyone else that people will then adopt in mass. It’s easy to make wrong decisions, but at some point people will start making a right decision, and as other people recognize that it will contribute to propagation. The more that education or healthcare ties themselves into knots, the more apparent will be the advantages of some smaller system becoming manifest. I look forward to having that happen, and am happy to participate with anyone who wants to make it happen.
24. How to teach complexity
Some people have attempted making concepts teachable to kids in elementary school for instance — I don’t know if that’s right, there’s a progression of ideas that requires a foundation. It may be that you have to relearn basic math. Basic math teaches 1+1=2. 1+1 may not be equal to 2 in complex systems because they may be interdependent, it may be 1+1 is 1. Mathematics teaches a certain set of simplified assumptions because it is a powerful set of assumptions for doing a whole set of problems. Some people may need to learn early, others may never need to learn complexity at all. We have to learn how to teach complex systems at some level first which we haven’t done in large scale yet. A human’s brain capacity is exceeded by collective human knowledge — so the question is what do people need to know. And if you want to teach people complex systems science, you better teach it before their capacity is exceed by statistics.
25. Where can we go into depth in one area with complexity
If you go through my book Making Things Work, there are lots of examples, and each of the demonstrated tools is also applicable to multiple other domains. One way is to study these examples then go back and see how to apply them in some other domain.
26. Skin in the game
With respect to the complexity categories of random / complex / coherent, the question is whether the principle of skin in the game is more or less appropriate in any of these domains. I don’t know if these categories are the right ones to evaluate. The assumptions are important.
27. How to deal with absence of data
Very simply, you either have to see if the data is present in other information you already have, or otherwise go after more data. This is the bottom line.
28. Controversies in the complex systems field
Complex systems science has developed in an interesting way. The point is that it is really about walking into a different dimension. We have well developed sciences (physics, chemistry, etc.), and you can’t expect to contribute to any of these domains unless you go all the way into graduate school or beyond. But complex systems science adds this new dimension of a whole new set of tools and the range of possible tools is very large so what is only now beginning to happen is this space is starting to finally fill in to where people can ‘see each other’. So now there are opportunities to relate the advances that different people have made. There are many opportunities to make contributions.
29. How to study global systems
The answer is simply we have to roll up our sleeves and do the analysis. We’ve talked about how to do an analysis, which involves recognizing the fundamental relevant aspects of the systems at large scale. With respect to political aspects, we’ve dealt with collective behaviors, but in political environment there’s amplification, interactions between scales. There are particular individuals that can influence system at large scale (either from their roles or how they interact) so this has to be included in the frameworks.
30. Peer review journals
There aren’t many. The best papers get published in high profile general audience journals (such as Science, Nature, Proceedings of the National Academy of Sciences, or the journals of the Royal Society). There used to be a Complexity journal but that has recently changed management. We need journals that are broad audience journals, we don’t currently have it. If you have a great story to tell these general audience journals are great, but below that it’s all specialty journals that only gets to a particular domain.
31. Other books
A few recommended books:
Introduction to the Modeling and Analysis of Complex Systems by Hiroki Sayama
Network Science, by Albert-László Barabási
The Sciences of the Artificial, by Herbert A. Simon
Micromotives and Macrobehavior, by Thomas C. Schelling
32. Parallels between complex systems science and the universe
This topic has multiple legs. A: Do you think the universe is a cellular automaton? Y: Stephen Wolfram spent 10 years trying to prove that a cellular automaton was the foundation of the laws of physics. He didn’t succeed, and after this he wrote A New Kind of Science. Let me start with the statement that I find it interesting that people are more interested in new galaxies verses global food prices, there is some part of me worried about that. The universe is large, going back to Hitchhiker’s Guide to the Galaxy, it’s really really large. A: there’s even talk of a multiverse. Y: What does the universe look like? Here’s a story. Physicists discovered a form of superconductivity in the mid 1980’s that they’ve been trying to understand ever since. There are thousands of people working on it and they still claim that they don’t understand what’s going on there. The universe is large, which means we probably understand a lot less about the universe than we do about this piece of material. There is probably more out there than we know about and there will be lots of surprises. The fact that people are assuming that the nature of the way we’re going about trying to understand the universe is the right way to do it is a deep question. At the same time there is reason to believe that pattern formation plays a key role in what’s going on in the universe.
33. Common mistakes people make in applying complex systems:
One is making assumptions, whether ones not valid for that particular system, or they use a specific tool not valid to that particular question — I would say that’s the most common thing. There are lots of tools but each has assumptions. AI for example is a tool, but if it’s not appropriate for that particular problem then it’s not going to be able to provide a solution.
34. Beyond earth in the complexity profile
The answer is that this is the same question as beyond earth for humanity. I do not understand why we’ve lost that conversation. The solar system is an incredibly available place at technologies only incrementally different than today’s. There are obvious ways in which other planets have stuff that would matter to us, I don’t see any reason why we should think about it as being somewhere else. On the scale of space these planets are not that far away. It’s an answer that NASA would like. When I think about what human civilization would look like some amount of time from now not that far away, I have no reason not to believe that we will be among the planets. It just makes sense.
35. Humanity collapse due to connectivity
There was an example given about hyper connectivity in the financial markets creating collapse, could this happen on the social scale? The challenge of civilization is to go past the time of instability to a context of whatever is the appropriate level of stability and instability for the system as a whole. The question is will we make it? Will one of these fluctuations destroy us? The answer that I’m trying to communicate is that the fact that we have become so complex is itself somewhat a guarantee against real failure. But does it totally guarantee real failure? The answer is, it doesn’t. I think it is highly likely that we will survive, but if you think about the development of a complex organism, like a mouse or human being, well it works — wow. But does it always work? No. So I’d give it a good probability. But I’m not going to give 100%. The human civilization does not have any firewalls, the human civilization is one entity. So we’re either going to make it or not make it together. We’re all in it together. As to whether we need more firewalls, there is need for more local control for dealing with local problems. This is indeed part of this divergence. Local control helps create a kind of firewall, but we also need some level of global control to deal with global vulnerabilities.
36. Merger of complex systems and AI
I don’t really see it that way, I think these are two separate things: AI (right now) is a particular tool, complex systems is a science.
37. Is Artificial General Intelligence possible
We talked about general intelligence. Consciousness I explained earlier in the week that I think is just a mechanism. Consciousness isn’t this all encompassing thing or this mysterious thing, I just don’t believe in that. It’s not a global thing. It’s just a particular mechanism. So in that sense it makes it available to AI, but that demystification means that it’s not the answer to all of the questions. Human individuals I think will always be more complex than individual AI’s for a while, but human civilization will be more complex than AI is for much longer. Is it possible for AI’s to become more complex than individuals? I don’t know what the ultimate architectural issues are in bio verses silicon. But silicon is invented in order to limit randomness, and randomness seems to be important, so it goes to basic architectural issues. Those are kinds of questions that I think are a millennia away at the very least. We may find that there are shortcuts to answering those questions, to a fundamental analysis, but I don’t think that this is something that is in our lifetime. I really foresee that we will be doing all kinds of things, like bioengineering, hybrid systems, all kinds of stuff, and the point is that we will learn which things work and we’ll exploit those things that work and separate the things that don’t. The kinds of things that scare me are the fact that we may fragment people. Different parts of your brain are good at different things, so why don’t we couple into the different things that different parts of your brain do, rather than treating you as an entirety? That kind of scares me.
38. The university model and vocations
The way I see it is that there needs to be a much greater divergence of educational processes. Will it become more of a trades type thing? In a highly complex world individuals are going to really diverge a lot earlier and take different tracks and different directions, at some point much earlier. In our current system we don’t know how to do that. But I don’t know. This is the complex question not the simple question.
39. Fundamental limits to AI:
I think we’ve talked about that. The bottom line is biology is itself quantum limited. I don’t know that we should believe that there is another architecture that is necessarily better. Could it be better? Maybe. Better in some ways? Definitely. Almost any question that is more than say 20 years out to me is not the right question right now. I think talking about AI is clearly in that class. The 20 year question is how do we make it through the transition that we talked about, with the divergence of individuals and becoming a global collective. That’s the challenge of our time.
40. Will humanity make it
I think I’ve already answered that. Probably. It’s not a very satisfying answer. I’m optimistic. Probabilities don’t really mean anything here, it’s either you make it or you don’t. It’s a non-ergodic system, there is a ruin barrier with a probability that is not zero. The biggest challenge from my perspective is the pandemic one. I’m not saying that there are not others, but the pandemic problem is obviously one we’re not taking care of and we’d better take care of.
I’ll tell you for me what the issue is. I see the divergence of people and the challenge of having children that are diverging. There are classic stories of children going away and doing something different that is inconsistent with the value of parents, but in world that we have we really have to find people to be close to. It’s really important at the interpersonal context to find people we are compatible with, it’s not a simple problem. In the family this becomes an issue because the likelihood that one of your children will actually be one of the people that you want to be close to may become very small. It’s going to be reinforced by the fact that there’s transmission of knowledge and culture and so-on from parent to children, but in a highly divergent system the probability may become small. So the challenge is to either accept if the children are going to go elsewhere and appreciate the nature of the links that you will have with them. This society has much less strong family ties in general than other societies. For me its very painful. That’s the issue for me.
We answered this already. We said sure it will be done. Those things to me seem fairly scary but I’m sure that they will be done. We’ll figure out which ones work and which ones don’t work and so on. That’s not necessarily a positive thing.
It’s now 12:30, and we answered all of the questions, how do you like that?!
A: Cheers and applause
Y: It’s been a tremendous pleasure. These discussions from the beginning were very engaged, very attentive, and lots of great opportunity to talk about important things. I really appreciate the questions that were about the opportunities for everyone to contribute to what we are doing. I personally think that we have important work to do. There is a lot of opportunity for the science to have a voice, and I see it in both our ability to succeed in dealing with the challenges, but also in providing some relief from the stresses that are existing in a society from the personal up to a global level in terms of recognizing that we need to get along in a way that is not the same as we needed to get along before, and maybe helping a little bit in the perspectives that will enable that to happen successfully. I hope that you will stay in touch. I think the most important way that you can contribute is to keep interacting with us, so we can figure out together how to make things better.
Books that were referenced here or otherwise inspired this post:
Making Things Work — Yaneer Bar-Yam
Skin in the Game — Nassim Taleb
Social Physics — Alex Pentland
Introduction to the Modeling and Analysis of Complex Systems — Hiroki Sayama
Network Science — Albert-László Barabási
The Sciences of the Artificial— Herbert Simon
Micromotives and Macrobehavior — Thomas Schelling
(Book purchase links from this essay are to the Amazon Smile program for charitable fundraising.)
For more information about the New England Complex Systems Institute (NECSI), please refer to their website.
Many thanks also for the conversations and diverse perspectives from the fellow attendees of the NECSI AI and Beyond program.