 Thanks for having me here. I'm actually, this is going to stay, I think I'm going to try to, because it's 20 minutes, I'm going to bounce through a lot of ideas and sort of hopefully tag them for more detailed and involved discussions during workshops and stuff this afternoon, rather than trying to go really deep into anything in particular. With that in mind, I'm going to talk about complexity in social and economic systems as sort of a leading point. So the reason that we have these complex emergent behaviors in our socio-technical systems is that we are fundamentally modifying the environment that we participate in. Actually, there's some interesting analogies to discussion of sort of biological evolution where our bacteria was functionally changing the environment and that changing environment changed the fitness landscape for the participation in that environment. And we're actually doing this in sort of macro human scale by engineering on the world and inventing things and building new technologies and fundamentally modifying the means with which we interact with each other and our environment, which in turn our social and economic behaviors adapt to. So this is sort of behavioral evolution. I'm going to be focusing on this from a sort of, social behavior and social ecological perspective, but I always find it fun to talk about how the sort of complex feedback dynamic emerges in these sort of what I would call cyber physical systems. And anyway, so this is sort of the main entry point. We have this overlap between the sort of temporal dynamics, how we behave, how that behavior changes the environment and how in turn the change environment changes our behavior. The sort of paradigms for complexity and engineering are not particularly new. I like to point out that the socio-technical systems paradigm was introduced in the 50s, the sort of second order cybernetics and the discussion of sort of control systems and the way we organize our societies and organize systems becomes part of those systems itself is sort of introduced in the 70s and was very popular in the sort of, I don't know like late Cold War post-Soviet application of engineering to sort of human systems. And now actually we have something that is a little bit newer. I like the term cyber physical systems there, things like our power grids where there are the economic loads put on the grid, the physical network infrastructure, the plants and various other forms of sort of business entities that actually collectively source power, source materials, source labor, like all the way down to the economics at the individual consumption of power level. Those systems are highly networked and in fact we see things like cascade failures when sort of part of a power system goes down, the loads get shifted and you can actually have the same systems that make the resilient in the short term can make it fragile because you sort of pick up the loads that were dropped and if that sort of overloads the system then you see more things go down. These are actually some of our best physical networked sort of multi-layer systems that exist today and I think it behooves us to study them as part of our sort of what I would say more social system studies as I get into sort of decentralized organizations, this is gonna be more about human coordination in a less physical sense. So this is a slide that I've actually given a 20 minute talk just on so I'm not going to do that right now but I am gonna highlight the fact that there's sort of a partition here across silos and across what I'm gonna call, I'm gonna call it scale here. I don't know that's the right word, at least in this context but our silos here are areas of expertise or sort of places where you can have understanding of the mechanics, the dynamics or behaviors in the social context and economic context. Here technology is here generally information technology and physical is the sort of more, the environment or the physical environment that you're interacting with and there are lots of cases where we look at one or two of these things together but when in our increasingly interconnected systems it's very difficult to fraction them apart from each other and still have a coherent view of what I'll call a generalized ecology. Like if this is a social and economic ecology then the environment is defined by technology and the physical environment but as the rate of change of the technological or physical systems rises it becomes less appropriate to separate the time scales and so we kind of have to reason them as that the systems that are co-evolving between social and economic and technological and physical but in order to do that when we pull that all into context we can't necessarily look at the full depth at the same time so we sort of trade our slice in silos for a slice in layers and the way that I've done this breakdown in my work is to focus on our sort of global systemic goals value flows like the desired emergent properties of a system which are the byproduct of the specific local agent behavior which is itself constrained by the patterns of interaction so patterns of interaction basically say what does it mean to interact? What is interacting? That's different from what interactions actually occur which is different from which sort of global properties emerge as a result of those interactions so this is still quite a sort of evolutionary dynamics like and as we get down into this trusted computation and durable data now we're talking about the way in which sort of novel information technologies are allowing us to sort of program these environments because we have cryptographically cryptographically guaranteed execution of code this is a bit like writing your own rules of a mini universe you're saying you are allowed to do this and not that and if you attempt to do something that the system doesn't allow it just doesn't happen unlike our traditional economic systems where basically you do anything that you get away with you pre-prescribe what you can and cannot do in these environments and an attempt to violate those sort of coded natural laws breaks but what's really tricky about this is that you don't really do this at a very human intuitive level to make it work well you're generally working in very abstract spaces you're defining you're defining the very notion of what is possible at an abstract level and it backs up to the fact that you have durable the last layer is what I call durable data or trustworthy state information which means that these computations can be state dependent which is what gives them real world like properties if you just have trusted computation but computation isn't over isn't able to write or read a trustworthy state then you don't get the capacity to sort of embed state dependent dynamics to internalize externalities like you want things to have the same property of the physical world if I want to move the state of the system I need to expend some energy or expend some resource or I can't simply arbitrage an environment I have to actually pay the price of moving it and therefore I only change the system in so far as I really, you know the cost is worth it and so there's this Pareto response that says yeah okay I wanted to move the system but I don't want to move it ad infinim because moving the system costs me energy and as long as my capacity is finite I can't diverge the system and so if we want to talk a lot more math later during breakouts I'm happy to do it this is a conceptual framework for being able to go basically to create sort of programmable environments where you actually have reasonable predictability of emergent properties but the point here is that emergent properties are not defined as equilibrium in the sense of exactly this will happen rather they're quite the opposite they live in much higher orders like if the system can be changed according a certain set of rules that means it has a set of reachable spaces and you're actually just restricting those reachable spaces and that means that you have some underlying universal properties that can be thought of as some Lyapunov function in a high dimensional space over a part of the space maps to zero or stays within a ball around a set because moving out of that set is costly and that the system is sort of attracted back into that set but this isn't a statement about the state you would, the first order state this is often a statement about the second or like even like the third derivative and it's constraining how much the system can move away from a subspace in a highly expanded view of the world so the last conversation has me really like wanting to go into math but I'm not gonna do it I'm gonna talk a little bit more about the technology because one of the reasons I came here today was to try to give this community more knowledge about what this technology stack can do starting from the sort of low level sort of rules that define networks decentralized networks and configurations through which peers interact at the computational level you can form peer to peer networks which are essentially communication and computation networks where each peer is essentially provably following the rules of the protocol and their behaviors are not controlled just the rules about what valid behavior is and the byproduct of that low level technology is to enable sets of higher level interaction patterns which are sort of if this is computational communication network this is sort of social and economic behavior where what one can do agent to agent here is a property of what this peer to peer network exposes so this gives you the rules of this world and enforces them and thus individual interactions are hey whatever follows the rules but then collectively those fit into these larger sort of networks of interacting activity sort of ecologies of behavior which actually results in emergent properties and from a design perspective you have to basically make claims about what you want what you're trying to achieve up here work backwards through this conceptual stack towards what I can actually build and roll back up to defining these things and essentially enabling and measuring the extent to which you're achieving your desired properties we generally approach this using a sort of formal scientific and engineering process which mixes basically three kinds of science right we have analytical most of the system design is Lyapunov theory a lot of it because I run a private research company is part of projects so I'm working on writing some papers and some tutorials when times available but the core sort of high dimensional representations of systems in abstract mathematical spaces allows you to do computer aided design essentially run simulations of those systems and refine your understanding of those systems and then ultimately implementing things observing real data which creates a sort of standard scientific feedback loop but we're trying to basically understand the implications of our decisions about social and economic rules one of the probably most important points that I make ever is that whenever you're doing this you are making subjective choices about objective measures there isn't an objective fitness in the sense like people like to reason as if there is one and more often than not you have to pick one to reason and then you have to back out in sensitivity test and understand what is the implication of making that assumption because you don't get to say well this is the social utility of the system like even if you go into social choice theory like there's a million ideas and none of them are inherently right and often they even conflict with each other so you have to basically say I've systematized this social and economic co-evolutionary system and at various points in the process I have made subjective choices of objective measures and so one of the most important parts about being a designer of these kinds of systems is to actually be explicit every time you do this and find ways to A, B test or sensitivity test those assumptions so sensitivity test when you parameterize things A, B test with alternative metrics and repeat the same experiments computationally when possible with different measures and see how robust the results that you've gotten so something might have nice dynamics and converge to an equilibrium under some and here I mean equilibrium in the sense that there's a low dimensional attractive subspace not that the whole system somehow converges to something static whether or not those spaces, the behavior patterns the sort of ultimate emergent non-equilibrium dynamics that have some properties actually are desirable is also subjective so it's a really weird place to be because you're sort of forced to simultaneously objectify but to do it well you can't actually trust your own objectifications so anyway, so I'm gonna talk a little bit about decentralization and organizations because I think these terms semantically cause a lot of challenges and in order to actually make any meaningful progress on any project you have to sort of get on the same page so I define organizations in terms of networks and just simply say that you have it can be multi-scale but for simplicity we'll talk about a network of individuals a network of individuals is basically going to say well, I have some form of interaction with another party and I have this interconnected web of labor activity, exchange trade, whatever and my organization can be defined as essentially a network of individuals now for the most part I'm dealing with sort of open social and economic environments so an organization might be a community in the sense that we know each other or have people in common or it might be something more formal like a corporation or a community that you join or have to get permission to join in order to keep something formal and abstract we're just basically defining them as networks over agents and when you have a network over agents you need to understand what does it mean to have a link and a link is a function of coordination and for whatever the context is we can talk about well we want to do something together, whatever it is and I have some rules, requirements, limitations whatever and you do and then the feasible space of links that we could form between us is basically the overlap between these two things and if there's no overlap then we don't link and it's no big deal because the system doesn't require that everybody's linked to everyone we just have to have this idea that there's this network formation game embedded in the idea of an organization as a network that determines whether or not we choose to persist or a relationship and that is an abstract concept because all we really want is the graph at this point you need a specific problem space a problem to be working on where this collapses to a specific something specific like an agreement in one of the projects that I worked on was to teach a class it was a project called Odom it's an on-demand education marketplace and the agreement pattern was a multi-party someone offers to teach basically puts out a thing that says I'll teach this class for this much money if so many people will sign up, et cetera, et cetera and people can sign up and once the preconditions are met for an event it's essentially locked in and there are penalties for reneging and breaking the collective contract but that's a feasible agreement space followed by events that transpire and then ultimately if the conditions are met it's determined to be an agreement and at which point reneging it has costs but that's a specific example the point is that you can construct sort of arbitrary coordination problems of the form each participant has a set of requirements and there's an intersection and only the agreements that fulfill the needs of all of the parties would constitute a feasible complete agreement and then I'm probably not gonna try to go into this too much one of the biggest challenges here is that these things are multi-scale so you can't just flatten it always and look at one sort of community or one group a lot of the times you have to examine the way multiple communities interact with each other so it's like the network itself is got a network implied by the sort of co-participation and I use myself as an example individually I am sort of motivated by impactful science and engineering research but I also have sort of a context as a member of a family where I expect some physical health and emotional connection I run a firm, I have a professional community I'm part of a government and society I care about this sort of world system and it's sustainable health I ultimately at every one of these levels of scale I could be a participant in multiple communities and I have individual relationships but I also have firm to firm relationships and I have even professional community to professional community relationships I ended up here because one of my collaborators in the sort of engineering research domain around social and economic networks was invited and he lives in Berlin so he was like oh hey you live in San Francisco can you go speak to this community and so that at some level was a individual connection that was manifest as part of a professional connection so one has to be really careful even about slicing these networks to finally when there's multi-scaling going on this really though comes down to scoping problems well so if you acknowledge the sort of superscope then you can prune it down to find the right dotted line boundaries that help you solve the problem in hand without having an a priori assumption of here's the template, you just model it like this and I found that to be particularly- Then if I mentioned that then this network become more like a field, don't they? Like a school of fish, it's not a network. It depends on how you like to define your networks. In math it's like- It's just static, like you're writing the lines and structures, fields is dynamic. Right, so networks can be dynamic but the mathematical object isn't richer it's like saying time vary it's just a mathematical state object that contains essentially nodes and edges and even it can have multiple classes of node multiple classes of edges the presence or absence of a node or edge can exit or leave over time so I generally work with dynamical systems and optimal control frameworks having come from a much more sort of robotic AI decision theory background and so I basically assume everything is dynamic state dependent and stochastic but that's when I get into my math and then we start to collapse as much of that out as is appropriate but you start with the assumption that basically things are big honk and mess and then you try to figure out which things you can collapse out and retain good paradigm like paradigm fit you want the model that you're imposing to be as useful as fit as well to the problem at hand as possible but you actually start with something obnoxiously general and abstract and you collapse it onto the problem at hand rather than sort of start with a template and say how do I fit my problem into the template I have at least, that's an element of process for me I find it to be pretty useful. There's actually an area of math that I probably I'm sure you probably have some adaptive networks where it's like dynamic upon the network and then also dynamics of the network topology itself changing over time and so there's different layers of dynamicism kind of in the math to explain it on the network a couple more minutes. Yeah, sorry. And yeah, so that's like multi scaling in time in addition to multi scaling in space so when you talk about the multi layers of time scales you have to deal with much like we were discussing in the biological sense you have environmental evolution you have this network evolving essentially what I would have called the ecology evolving which is different even from the individual agents strategies or behaviors evolving so the issue is that you can't always handle all of those things at the same time so you need to figure out what things are have separable time scales what can I assume is constant because it's moving too slow what can I assume is constant because it's so fast that it's at equilibrium and then there's some temporal phase in the middle that you can actually focus on but it depends on the question at hand there's not an a priori absolute right scale you kinda have to have this really big complex view of the world that's too messy to work with and then collapse it onto the problem at hand if you wanna be able to sort of make meaningful progress that's the experience that I've had so I wanna talk briefly about what decentralization is because this term is also very very loaded especially in the technical space that I'm in right now we have decentralization in a bunch of different metrics so we can think of it in terms of peer-to-peer versus centralized it's decentralized if you have a more peer-to-peer architecture it's centralized if you rely on a central authority but the notion of a central authority or a bottleneck can depend on a lot of different things so in the sense of transparency this is accessed information the system is decentralized in the sense that anyone who is participating is sort of permissively access has access to information which is different from the way the decision authority is exercised you can have systems that have high degree of information access but highly centralized decision making authority and actually my favorite examples are around the evolution of things like Uber and Lyft and Airbnb as centralized technology platforms that allow for decentralized activity so you have open access to information at the level of who's available to say provide you a service you have individual rights to make a decision to say go out and drive and there's not a lot of intermediation but there's still heavy intermediation at the technological level and the blockchain community in particular is pushing to make this technological system more decentralized but often they sort of conflate this with the decision making authority decentralization because even if you decentralize the technology the social and political and economic behaviors can actually remain very centralized even though they're ostensibly structurally the information access is available the platform allows it to be decentralized but the behavior is still very much like oligarchy so you have to recognize the difference between something technologically decentralized and something where the actual emergent behavior is decentralized okay yes Lightning Moon Per位 place cool so I won't show these examples from my work page but later this afternoon yes