Dr. David Krakauer speaks on the nature of simplicity and complexity | Academic Conferences

[Music] our next speaker is um quite an honor that he is here to speak to us david david krakauer he is an evolutionary biologist and the president and william h miller professor of complex systems at the santa fe institute he also co-directs the collective computation group at the santa fe institute and prior to all of that david served as the founding director of the wisconsin institute for discovery as well as the co-director of the center for complexity and collective computation and as a professor of mathematical genetics at the university of wisconsin-madison that's only part of his introduction part of his resume david has also held positions at uh as a visiting fellow at the genomics frontiers institute at the university of pennsylvania he's also been a sage fellow at the sage center for the study of the mind at the university of california santa barbara a long-term fellow of the institute for advanced study in princeton and a visiting professor of evolution at princeton university and if that wasn't enough in 2012 he was included in the wired magazine smart list as one of the 50 people who will change the world and he's here with us today so i am thrilled to have him join us today and it is my absolute pleasure to introduce to you all david krakauer hello everybody morning thank you very much those can you all hear me great yeah those sort of little introductions who are they talking about who is this person that's not me that's been introduced so um i i'm just going to jump right in um to my slides to make sure this all works um and let's see is that all visible to you great so i'm going to start in some sense angie system and then we're going to migrate towards system um this is a quote i rather love this is the quote from murray gilman's 1969 nobel acceptance speech murray is one of the founders of the standard institute as some of you will know um and he said how can it be that writing down a few simple and elegant formulae like short poems governed by strict rules such as those of the sonnet or the wakka can predict universal regularities of nature the wakka by the way is the ancestor of the haiku it would be a little bit too pedestrian for murray to say haiku so he's welcome so and this comes out of the work that murray had done that won him the nobel prize on recognizing fundamental symmetries in nature and he wrote down what's become famous as the so-called eightfold way which is the symmetry group um of the hadrons which are fundamental particles and it's an extraordinary achievement because you write down this what seems like an invention of the human mind that is the lee algebra for the special unitary group three and out of that pops a prediction of a hidden fat facet of reality right a subatomic particle this is the great mystery if you like of applied mathematics so that's that murray but there's another murray that was associated with the founding of the sun defense units the murray who asked this question which was think how hard physics would be if particles could think now this uh is in some sense the rallying cry for complexity science it says instead of a world dominated by conservation laws conservation of energy and matter and fundamental symmetries it's a world dominated by agency self-awareness consciousness reflexivity system two you know you pick your favorite phrase how do you understand those kinds of worlds where the uh system itself is learning and adapting responding to your observations in such a way as to stifle stymie your future efforts to predict them and that led quite naturally to this remark which i think would resonate with everyone at this meeting where he essentially makes the point that the the complexity of connections in the modern world is so dense um that there's a real argument for studying some emergent system uh which he calls the whole system rather than pursue that disciplinary um convenience of dissecting the world into its individual components of the kind that murray had worked on as a physicist and i've often wondered what murray meant when he said whole system um that's been confusing to me it's almost as if what murray did is he started as a reductionist physicist working on the atoms and quarks and gluons and took some kind of long jump uh by helping to construct the santa fe institute and landing on the whole system which is the planetary system itself and i think this is a significant challenge and i think what murray actually meant was looking at that point if you like of highest elevation where we're not studying the whole system i think that's actually arguably impossible but we're studying the systems of the world um and i want to argue throughout this presentation that the systems of the world should be thought of as verbs not nouns and i mean when i go to meetings people will say we should study the energy sector or the communication sector or the global banking system these to me are objects and artifacts as opposed to studying processes that actually transcend any given one of those systems if you like and that's what i'll argue for not to say that you can't do the former that would be perfectly reasonable and fascinating but i'm more interested in the generalities that span them and so this will be essentially the table of my talk in a nutshell we're going to dive into some of the details because i think the the the interest of the of this approach to to working on systems lives in the detailed way in which we analyze them and so for example um conflict it's obviously all of these topics i imagine are of great interest today in the world can we understand conflict both at the individual level right that is individual fights if you like all the way through to wars at the level of nation states is it meaningful for example to describe them both as conflict or is that just a sort of historical accident that we use the same word to describe events at different scales and i'm going to argue that the only sense in which it's meaningful is if you can find the mechanics or the models that are constitutive of the process right and demonstrate that that actually yes they are because what we mean in both cases is a certain notion of criticality and i'm going to jump into that in a second another one culture how does culture actually work um you know where we're suspended ambiently in a world of beliefs and ideas how do we acquire them how do we transmit them and that this notion of culture is very um diffuse right so we go from the culture of individual beliefs and norms all the way through to you know enforceable legal systems and again is that real i mean are we correct in asserting that these are elements of culture and i think the answer is yes because in each case we're talking about attitudes towards imitation transmission if you like the sort of essential modules or atoms of cultural evolution and so forth again what is constitutive of that notion of culture and i'm going to focus there on on the evolution of constitutions and finally revolutions social revolution scientific revolutions um why do some institutions emerge and then freeze and are extraordinarily difficult to change whereas others seem to be in flux all the time can we come to some fundamental understanding about why that's true um and again here i think the answer is yes we can we have to think about structures and orders that have multiple time scales and the way learning rules percolate through those time scales um so as to either engender fixity stasis or evolution and change so that'll be the sort of that sort of the talk and i'm going to jump into empirical case studies with mathematical models don't worry too much if it all looks a bit heady um just to say that i think that you do this just study says you have to do this in other words with the essentials of this scientific method that is empirical data model building and experiment are no less important here than they would be in any disciplinary pursuit it's just that we're doing something rather differently so let's start with this conflict of contagion issue um let me just talk about criticality first this is a a plot that you would have all seen if you're watching the news at any point in the last three years which is sort of flatten the curve right so this is a series of projected epidemic profiles for of covid for example um assuming different critical exponents are zero or r naught in england this is that thing which we've all become familiar with which is the number of secondary infections initiated by a primary infection so in that little picture on the top right you can see that if r naught or r zero is five then a single initial infection will lead to five secondary infections and so on and we've all learned that as you move the critical exponent r0 closer to one we flatten the curve and if r0 is below one the epidemic would peter out so this is what's meant by contagion this propagation of infection and it's also what is meant by a critical exponent which is this threshold value magic value below which an infection is subcritical and above which it's super critical okay many people have claimed in the literature that conflict is something like a contagion it spreads from one person to another through a mechanic which is somewhat dissimilar i imagine what we know to an infectious disease and moreover it shows critical phenomena that is that there is some threshold value of aggression um or adversarial interaction above which conflict becomes supercritical and below which it would be subcritical and the idea i guess of the pacific dream would be to get those critical exponents below one right so as to move into sub-critical regimes is any of this true there's a huge literature well to answer that question to look at the data and we have to build mathematical models and answer this question of whether or not conflict really is a contagion or not um so here's the kind of basic questions we might want to ask ourselves is it true does it have a critical transition is there a relationship for example between the conflict duration the number of fatalities the the geographical area that it encompasses and if that's all true what are the normative implications that is we're not just doing um descriptive natural history of conflict we want to intervene in such a way as to minimize it the way for example a physician might make recommendations for mask wearing so as to minimize the speed of a pandemic so those are the sorts of questions so you know what a chance it's like now i wonder if i can see if you can see this is anything blocked here by the way laura or can you see the whole screen because of my own feel good we can see the whole screen okay great so um so the way we do this is we look at two kinds of systems we look at experimental systems non-human systems that we can intervene into and um so these are small populations every individual can be monitored and we're looking at the individual conflict rules that individuals are following so very micro on the other hand we have large-scale observational data of human populations very macro and in that case we're inferring quite different kinds of scaling relations not individual decision rules so i just want to make that point about model systems versus observational systems it leads to a lot of um heated debate i think one should look at both because they each have strengths and weaknesses um so let me just give you one example this is something we've been studying for many years non-human primates macaque spider-monkeys and so on and here just to give you a sense we can look at any given moment in time these are animals in captivity but who is fighting or not fighting at any given point in time so very micro we can see who interacts with whom we can reconstruct social networks of their interactions and so on and out of that kind of data you can get all sorts of interesting insights but let me give you one example this is the scaling of the on log log axes of the conflict duration and conflict size just notice that as the uh number of individuals in a particular conflict increases so does the duration of the conflict and it increases polynomially it increases as the square of the number of individuals you can also look at the variance not just the mean so this is the distribution of how long peace bouts last so for example no one's fighting anyone for 10 seconds 10 minutes an hour and so on and conflict durations and you'll notice that if you look at the distribution of durations as they get longer the variance increases so the scatter around that distribution increases so here's just some data what one can now do is write down a whole family don't worry about the details of mathematical models and explore the full configuration space of possible contagion processes not just one or two but a very large number and ask which ones of those mathematical models are consistent with the scaling relationships for example the one observes in the data the scaling of the mean and the scaling of variance and you'll see that most contagion processes actually can't recover the observed data and this is quite important because in many studies people will assert contagion without checking carefully against the empirical data which would either refute or confirm transiently the hypothesis and what we find just is is a critical role of agency it is a contagion process but with one really interesting characteristic which is long-term memory that is the memory of the first conflict and its duration is felt throughout all subsequent conflicts that is not a property of a mathematic of a biological epidemic this is a very different kind of contagion it's contagion with significant memory in it which is that sort of agency fact that's self-awareness fact and we all know that and resentments are very long-lasting so an attempt for example to intervene on a human conflict contagion as if it were a biological contagion without taking into account the sort of psychological fact of memory would fail so that's just one example uh we've also looked as i said at the more core screen data we've looked you this is the so-called acled database it's extraordinarily rich and we've been studying conflicts in continental africa for two decades uh looking at things that are very similar actually to the micro data i just showed you and here you don't have who's finding whom at a given point in time but what you do have is data like this where over the course of time you have battles the number of fatalities um exact geographical locations and so forth so we can study time and space but at a much more low resolution than we would in the experimental system and as with the experimental system we can look at things like the distribution of conflict sizes uh the distribution of fatalities duration and spatial extent and as with the micro data we can write down mathematical models this is a rather complicated model and when we cannot describe this one in any detail but just the idea that contagions grow as a kind of a fractal like branching pattern on a two-dimensional surface and one of the things that these models predict is that at critical points like for example r zero equals one that is that critical point below which an epidemic is um sub-critical above which is supercritical something really interesting happens you get a complete collapse of complexity and all of those scaling exponents become related it's actually a really beautiful feature of complex systems that are critical points they become simple and the for those who are interested in the form of analysis this is using something called renormalization group theory it's a it's a way of looking across scales and you can see and just take my word for if you like that if you look at the ratios of the empirical data against the model predictions are extraordinarily close which suggests two things again yes this is a contagion-like phenomenon so that's a correct statement but it's also a contagion-like process that's near criticality and this is very surprising um because it's not clear why that should be the case so if you think about an epidemic with our zero why would our zero hover around one why would it live by a critical point and we all know the answer actually because we're observing it today and that is that if arnold or zero i never quite know i've sort of been in the united states long enough to want to say zero but my answer my early life was in england someone said naught so i'm going to be at the critical point where i say both and um if you think about it when the epidemic is in full blast we start taking precautions we wear masks we get vaccinated we get boosted and we shift our zero down but when our zero falls below one the epidemic starts petering out and what happens we take off our masks right because we think oh look we've won and then what happens the new strain emerges and it propagates back in and we put a mass back on and this is one of the so-called routes to endemism and that phenomenon of a system hovering around the critical point is called self-organized criticality and it seems as if conflict is manifesting this phenomenon of hovering by a critical point so okay so is it contagious yes uh with a strong agency memory effect does it have a critical transition yes but we don't know exactly where it is we don't know the r0 of human conflict um you get this incredibly interesting simplification of the system of conflict at the critical point where all the scaling exponents become related um and the normative implications for us are deliberative techniques to move systems away from critical points and that's a whole area of investigation we've pursued and i perhaps read any questions we're going to deal with it but if you look at my website there's some things on that there so that's just one example let's keep jumping look at culture um how on earth should we model culture and this is again my view is you have to take a representative model system where good data is available and you could look at twitter you could look at newspapers it almost doesn't matter as long as you're rigorous and uh you know consistent in the way you do this so we've been looking at constitutional history and so this is a very famous painting it's painted in in 1940 it relates to the signing of the constitution the american constitution in 1789 and you can see in that picture i'm sure a few familiar faces and what we do working with constitutional historians and lawyers is we take a constitution and we atomize it we break it down into its essential themes and for those interested in machine learning or natural language analysis these atoms are called topics so for example oops these are my questions let's just skip them we'll come back to them so here's a constitution we're going to break it up into its basic lexical atoms the way it treats general rights the way it treats sovereignty public order and we're going to ask how do these themes diffuse forward in time through constitutions the way that constitutions are written is you don't sit down in a room and write one for your nation de novo out of nowhere ex-ovo you look at prior constitutions and you borrow from them and uh what you can do mathematically is reconstruct the phylogenetic tree of all constitutions that have ever been written and we if we zoom in on this you can see here's the constitution of egypt of 1923 it was highly influential on burundi 62 albania 25 yugoslavia 31 year round 1925. these are obviously a reading chronological order um what's going on here uh which topics are circulating how is borrowing taking place and why what's the impact of local culture and geography and i'm going to skip all that for today but just give you a sense of what you can say in this structure we can sort of leap in and look at the micro geometry of that tree what you're looking at here if you look at the center you have the focal constitution of interest it has a number of parents that it borrowed from that we've inferred using some techniques and a number of offspring that it influenced that borrowed from it and we can reconstruct this kind of zoo of geometries of all the constitutions all of these are essentially that structure on the left the giving of a sense of the diversity of constitutional forms that we can reconstruct going back to 1789. and these actually correspond to particular constitutions you can look at the bottom one that's the constitution of montenegro we had three parents no offspring right um the interesting one is the south korean constitution of 1948 that's that one sort of third from the bottom it has a small number of parents but a lot of offspring that's a very influential constitution right and so we can start asking which constitutions are highly inventive which ones had disproportional impact which ones are sterile actually never led to offspring and so on um and so complicated plot but what if you look at the lifespan of a constitution the lifespan is defined as when it was written and when it started it stopped being borrowed from right no one looks at it anymore the u.s constitution is very interesting it was very influential early but it was so influential on secondary constitutions that they borrowed from those instead of the founding constitution and top left you see a tree of all the constitutions you see three clear epochs found in 1789 up until the second world war this is a period of a relatively slow rate of constitution rating and all the constitutions are very long-lived then during the war subsequently you get an acceleration of the production of constitutions most of which are very short-lived so quite distinct cultural phases uh in relation to this particular object which is interesting because what are constitutions they establish branches of government the separation of powers property rights issues about freedom of expression these are in some sense the operating systems of societies and they have a very interesting dynamic and out of that we can then answer these questions how is culture transmitted culture is transmitted a little bit like genetics it's not genes it's core constellations of concepts like rights property rights uh structures of government which ideas are most influential in strong early mover advantage we see quite unequivocally that if you're early early in the writing of the constitution you will be an influential one it almost doesn't matter what you write about because you become a source of um influence on subsequent constitutional authors um this is a preferential attachment dynamic that many of you will be familiar with it happens for example with uber happened on ebay happened with etsy the being there first and becoming the preferred market for ideas is a long lasting effect and it's no less true for these ancient cultural forms and it is for contemporary technological forms even though many have claimed the dynamics of software um is different it actually isn't you it seems to be a more universal dynamic of culture than a particular mechanic of the contemporary world and this really interesting point that it's better to be revolutionary than evolutionary if you're writing constitutions that is the highly inventive constitutions that that add new topics if you like new categories of thought are the ones that people look to not the ones that add one or two increments on what came before them so be first and be radical is the sort of if you like the message of constitutional history finally revolutions like everyone else i'm interested in revolutions particularly scientific revolutions but during the last few years social revolutions um the whole black lose lives matter protest is of great interest to me personally um and how can we understand what's going on here why why do society seem to go through periods of stasis and then these incredibly rapid bursts of change is there a sort of unifying framework for evolution and revolution in social systems uh the kinds of things i'm interested in things like this you know why do we go from extraordinary intolerance of homosexuality to accepting it uh going so far as to allow for gay marriage uh which to me is a real fantastic evolutionary move of society what what changed in people's thinking by the way there's a nice typo here this is oscar wilde 1985 which is not quite right of history um so this you know attitude toward this cause of fiction on the left for anyone in rock history you realize these are two very different people but this sort of scare tactics that people have used to sort of discourage people from taking recreational drugs and maybe they're right maybe they're wrong but then this shift towards a much more permissive um society political beliefs i mean endlessly oscillating back and forth uh between political parties can we sort of wrap this system up in a common mathematical framework so that we can understand properties of all of them um and as i said i think the answer is yes because they're all characterized by having multiple time scales with individual learning rules that percolate up through the time scales often in invisible ways leading to very dramatic shifts the kind of data that we look at to see if your senses is in like this all across the board why for example did seat belts once they're adopted lock it's very unlikely that we're going to move back to a state where people say you know let's get rid of the mandatory seat belt law it seems like a bad idea um whereas political parties uh fortunes are constantly shifting attitudes towards the death penalty are constantly shifting and so on i mean it would be kind of nice right if attitudes towards same-sex marriage or the death penalty look more like the seat belt in a progressive sense and once we'd adopted them we stayed that way and didn't constantly fluctuate can we understand all of this and so these are the kinds of questions we're asking here how did these systems evolve how do we as individuals or as collectives contribute to these institutions so i use this word institution by the way to mean scientific belief systems or organizations companies could be ibm how do individual micro learning rules percolate through to the emergent structure of the of the macro properties of the institution and what um an institution do that wants to change um but finds itself like the the seat belt locked in you know can we are there normative implications just as there were for conflict this is the basic structure of these models you have collectives at the bottom um they express their support or opposition to an institution through voting mechanics and we think of the institutions as a ledger that records in its entries votes it aggregates preferences and beliefs and it has a public position and that public position feeds back to reinforce or inhibit uh the collectives that are learning to support or inhibit them in other words members of of of a republican party will support their own party but they will do everything their power to oppose the democratic party and democratic beliefs and vice versa so the same kind of logic applies by the way to scientific revolutions here's the so-called copernican revolution you have copernicus saying you know you know everybody the sun is at the center of the solar system it's it's kind of the past millions thing but you've got the judge saying well no that's kind of a little bit at odds with what we've been saying for the last however many thousand years we'd rather you shut up and so again that sort of dynamic of voting uh opposition and support with feedback from the institutions because that's crucial right because if the institution helps you um you gain power and we can take that and mathematize it for those of you with keen eyes you'll see this looks like a perceptron it's a very simple neural network but it's all analytical um but it doesn't matter um mathematize it and explore this is gonna be the slight technical point and it's coming to the end which i would like to sort of get your heads around the crucial role of learning rules which are the rules that reinforce your beliefs and support or opposition based on the success of your corresponding institution so here let me just show you what i mean so here's a case let's say x supports i x and y support i y the institution y one learning will be simply to say look if there's lots of us lots of x's lots of let's say supporters of the copernican revolution we're going to support governors in all publications that promulgate a heliocentric position vice versa right if you're a member of let's say the time uh religious fraternity or church if you're more numerous you'll support your institutions in that case they say the catholic church that's one kind of rule it says you learn in accordance with your success another kind of rule is quite opposite to that is that um actually i might just attend if i'm being successful why bother why bother doing more work i've already won i'm going to divest from supporting my institution because i'm in the leading position here's an interesting one an arms race an arms race says i'm going to support my institution in proportion to the success of my rival so think about uh nuclear expansion if if your opponent doesn't have weapons then why should you so your support of your institution is in proportion to your competitors abundance so you can write down all these simple rules uh which tell you how to construct institutions according to their abundance if you like or power or size and on and on you can go and you can actually mathematically look at all possible permutations and if you do that you make some interesting discoveries that if you use for example this rule which says invest in your institution in proportion to its size you get lock-in you're going to fix on a single institution and never change it and you can sort of see intuitively why that's the case because you're not attending to the competitor at all and the bigger you get and the bigger your institution becomes the more you support it lock in the same is true for other rules the arms race is the opposite right it says i support my institution in proportion to the competitor that leads to endless cycling periodicity so what this sort of work is showing is that the incentive system that society puts in place that is the reinforcement learning rule has strong emergent consequences on the stasis or liability of the corresponding institution and that suggests that you could change them so for example let's say you were in a lock-in situation and you're a company like kodak and you say well look we make the best film in the world we should make more of it and then all of a sudden the market changes then you're extinct you could as a leadership team say i'm gonna create a new learning rule amongst my employees that instead of investing in proportion to our market share we're going to look at the competitor change the norms change the incentives and move into an oscillatory region where you allow for the possibility of flexibility so this is just an example of these very counter-intuitive ways in which local incentive and learning systems have emerged in properties at the institutional level and i just wanted to make that point that rule there is an institutional black hole this is kodak this is research in motion um follow these rules stasis so okay so how the institution evolves in this particular case feedback from the institutions that are established through some collective voting dynamic um how do you construct use different learning rules to do so um these rules in a very non-intuitive way bias the stability institutions and how do you change you have to change the learning rule you have to that's the key and that's very hard so let's just end with this murray statement mary making that point that we're living in this highly connected world someone should be studying the whole system i think that the correct interpretation of that kind of insight is that we should be studying global representative systems that have a mechanic that is constitutive of that system um we should be studying culture conflict institutions cities pandemics these are the kinds of processes where we have a principled means of attaching to them mechanics and dynamics um the whole system i have absolutely no idea it's a spaghetti and you'll note that these aren't disciplines they span them but they're as disciplined as the disciplines which is the sort of mantra that i i try my best to follow so if for those who didn't know if i go to a webpage you can if you scroll down to the bottom of the santa fe institute webpage you can subscribe to our publications um you can join our podcast there's all sorts of information available there for you that you will either hate or love according to a disposition but it's worth having a look because there's a lot of material and with that i'll wrap up thank you very much thanks david um that was fantastic uh just we need a whole day on this uh fantastic uh and i do encourage everybody to go to sfi and check out what they're doing it's a very special place you still do tea david of course we do tea [Laughter] tea tea was sort of the most amazing thing where you get to meet these just people from all over and talk a little bit about tea well it's interesting you know i mean this is we've all been debating this move towards this world right this online world which has extraordinary that like this i mean you can do this one thing that's fantastic um but what what is missed and i think the obvious point is that this is very telegraphic right in other words i just waffled on for 30 minutes but we'll have questions for 15 and most people don't have a chance to ask you know it's whereas if you're in person someone can really pester you and it's great but also you can move back and forth you know clarifying what people mean because it's very often takes two or three questions to sort of get to oh that's your sense that's what you mean sorry i misunderstood you and so um t is our sort of mechanic for doing just that we pretend it's this informal thing which it is but it really is it's recognizing the complexity of human interactions and that it takes time to to understand each other right into what do you know so i don't that's what she's for so at sfi they have a courtyard and people just have tea and you know all all manner of conversations are sparked uh at t and and you could have a physicist and a sociologist and a mathematician and a you know all kinds of different people sitting at a table talking about a an issue it's fantastic well one thing to say i'd like to point out i didn't talk about sfi and its structure but to do the kind of work that i just described is a tiny slither of work it happens to be work that i've been involved with i could have talked about other people's work right um you need a different kind of organization and um so we don't have obviously departments we don't have deans we don't have all that stuff and so we and in offices which are typically shared you'll have an archaeologist with a quantum mechanist right and by the end of the summer they're working on a problem together because they like each other i mean it's sort of simple as that and um so the structure uh respects i think the the mission um in terms of what we want to accomplish yeah so i'm curious you started with the the one of murray's uh quotes uh think of how difficult physics would be if atoms could think is a fantastic concept um and murray gelman also wrote a very short paper called uh let's call it plectics which really got at the root definition and root uh word forms of simple and complex can you talk a little bit about that relationship between simplicity and complexity and why that's important yeah i have lots of slides on this i didn't want to do this because many times but i will do it verbally um so one way to think about this is that where have the because we are a quantitative department i don't think you need to be to work in systems or complexity quite frankly i think philosophers do it very well in natural language but we could do it that way um the one way to say this is that you know physics has been so successful the kind that murray worked on that i illustrated because it's so simple meaning there are laws like the conservation of that of energy matter and so forth the second order genetics fundamental symmetries and that's what mathematics likes okay um it captures in a very compressed elegant form regularities it's the essence of mathematics like pythagoras theorem at the other end of the spectrum have random things like the ideal gas laws it turns out maybe counterintuitively that highly random systems also can be compressed with very simple laws yeah but in the middle between the world of extreme regularities like the celestial mechanics or on the right you know thermodynamics and statistical mechanics you have this domain which has randomness in it but randomness that accumulates regularities like evolutionary history right that is the domain of complexity it's the domain of living systems and people say well you mean you're just biologists and we'll say well in one sense yes but we're studying markets as well and we're studying ancient civilizations and archaeology it's the living world that we study but we study it through that hubristic lens of can we find theories that are general for the living world and it turns out that theories in that domain have a different character in part because they have to contend with agency and i would say if you ask me what is the channel between the challenge of our century it is this it's that two paradigms have emerged to contend with that machine learning and complexity science and i and i'm willing to make that a very broad church but i am identified and um one of them is highly predictive and opaque machine learning the other one is less predictive but gives you the possibility of understanding how it actually works explaining to people what's going on you know pedagogy um and i think that we're at a very interesting bifurcation point in our history where that's never happened before and um if you look at the 17th and 18th centuries those two things were very close newton could be both predictive but also teach you how calculus works and you can write down people then may i mean it's that simple nowadays that's not true and i think we're now contending with these two churches the predictive opaque um and the explanatory transparent and i'm very interested in that in that schism and you also talk a little bit about that schism with with the human dream and and how that how how that relates uh can you share a little bit about that can you just expand on that you know yeah well well just that that then there's human thinking uh and and so you have these agents that that don't aren't seeing the transparency of the system and they're behaving a lot of a lot of what you just talked about in contagion and culture and how do we have that loop back to the agents so that they can change their learning rules or or whatever to change yeah i mean that in a way but is this is our this is part of our agenda right to understand um individual freedom in the face of institutional constraint um i don't think we have answers to that i think we can analyze cases um we still valorize the individual certainly in in the united states west is a very strong emphasis placed on individual creativity perhaps overstated you know we all know that theories aren't really invented by single people through hidden history there are hidden figures they made huge contributions that get neglected or ignored um but i think that is right carving out the space for individual freedom and free will in the face of these very powerful constraints which are market driven i mean this has been one of my crusades you know i mean there's nothing wrong with platforms that harness collective power but when those platforms have an incentive that is misaligned with human freedom i sort of go to war on them and so and it's you know an example in my own field the the journal system scientific publication peer review it's a perfect example of a corrupt system because we all know it everyone's nodding and yeah of course it is it's embarrassing here we are researchers and we are completely beholden to a system that does not have the progression of science at the forefront but sales and so it's not that i'm anti-market it's just when those things become misaligned then we have to rethink so we're constantly having these battles i wish we had them more often and we had more victories under our belt yeah so an audience member uh asks how can we use these uh complexity models to appreciate better the external drivers for the changes we see yeah i mean i think yeah that gets back to your question derek earlier which is they in some sense formalize the contributions of the constraints i mean that's partly what they're all doing i mean they're trying to you know where do we have free freedom to move and and where are the constraints established by the institutional feedback and and i think in every case we sort of have to analyze how much is being driven by the constraint and and how much is being driven by our decisions which is made even more complicated by the fact that our decisions emerged through a history of institutional feedback it's not even clear that we are free i mean there are some very difficult issues hiding in these circular feedback systems where you how to attribute causality but but they do i mean just to get expression you can actually do that you can partition for any given synchronous observation how much comes from the constraint and how much is coming from your internal state right you can start getting at those things i want to give you a chance to talk a little bit more about learning rules that the based on the success of the corresponding institution there um say a little bit more about are the learning rules merely evolutions of of the original rules meaning like their schema that are just evolving as a result of feedback yeah so what um the paper actually that goes into detail is it's actually going to be published on may 16th and it's called on institutional dynamics and learning rules so that will go that might help people um that was for that particular model a full discrete enumeration of all possible learning rules so we can do that in this very simple setting um and then pick any given one right and ask what are the implications if that were being used um so we don't have we don't look at the full continuous space between all rules and so on um but i think the purpose of that work was to show that there is a i think quite counter-intuitive relationship between the local incentive reward system and the ensuing institutional dynamic that that's the purpose of that work because if you look at the literature and institutions it it tends to work at the institutional level it doesn't have these multiple time scales which is mathematically by the way at heart and this is something that's a hidden idea here that one should surface which is that much of theorizing for good reasons is led by analytical expediency right right i mean especially economics um where they love that and the question for our community is when are we willing to forfeit that and that's really the move to machine learning or simulation right where you say you know what no i mean it's so far from reality i don't care if you understand your model it's useless you know and that's a kind of uh people set that threshold at different points somehow yeah it would be very interesting you might know that why do we do that psychologically why do we have different preferences for that uh i think we have time for one do we have time for one more laura yeah one or two more okay good uh so another audience question uh it's actually two questions does your work provide any insight into the role of relationships of the agents of systems and second does the fact that contagion hovers around criticality have anything to do with enhancement of adaptive capacity of the agents okay so yeah i mean i think the latter that's very interesting um i think the answer has to be yes to that second part because as in the example rate of of hovering by r zero equals one it's precisely because we are so adaptive that that happens right because as soon as we observe the information bearing on the sub-critical regime we take our mass off if we weren't adapted we wouldn't we would fix and the actually the pandemic would go away it's sort of a peculiar perverse situation to be in um so yes absolutely i think you're absolutely right adaptability is i think the key ingredient of tuning to criticality and um but you know what are we going to do get rid of adaptability you know it's enhancing and the first part seemed to be a more obscure i wasn't sure i understood the first part of the question well i i think that's interesting and i want to follow up on that so a lot of the systems that our students deal with are is is you know so you found this this uh results from a mathematical model but how do you feedback that that learning to the agent like in the case of covid if the if the individual agents could change their schema or their mental model so therefore changing their you know behavior you would get we we could get rid of covid right yeah so how do we take you know it seems to me the hard part is we learn these things about the at the systemic level but how do you get the individual agents to change based on that awareness i mean that seems to be like the crux of the whole thing and actually there it's even worse for climate change because it's for interesting i was in um sabbatical at harvard actually when the pandemic hit and i was actually working at the center for the environment which works on climate and immediately the conversation we were having is oh dear um how do we deal with systems that are massive global tragic collective commons problems yes like ameliorating climate change um but also operate on time scales where the the change of our behaviors know the implications of changing behavior are not obvious yeah so you just gave a good example i think that um in covet at least we see it right now part of seeing it is we adapt in the wrong direction that's the earlier question um so climate's a problem because there's no feedback it's like there's no the learning will come no learning no learning rule happens and there's so much delay there's so much delay and what i've heard discussed here is how do what are the appropriate metaphors that we could start inculcating in people's minds that would help them understand delayed feedback for example yes gardening gardening anyone who plants a garden right and i'm very impatient i'm not i'm not going to breakfast but there are people who are very good at it and they you realize i'm going to do all this now but i'm not going to see the fruits in my labor for about three seasons because it's gonna be really scrabby next year you know and so on and i think there are things that humans do that have those longer time scales where we're willing to operate with delayed gratification right delayed right um it doesn't seem to be the case here and i think i don't have an answer but i suspect educating people to think about pandemics and things like them the way they would think about gardening or something yeah would be very powerful and but unfortunately we do the opposite because we say we're going to have an rna vaccine in a month right right and so whatever virtuous long-term thinking we started to mature is immediately obliterated by the expediency of an instantaneous vaccine right yeah no i'm not blaming vaccine development that's fantastic but these things are actually kind of in competition in our minds yeah i love the idea of the metaphors and kind of metaphors we live by like the fruits of our labor really is built into that is a metaphor of delay yes and and so if we had these sort of systems metaphors that we could you know raise children on yes then when they get to adulthood they would they would sort of understand that uh cause and effect aren't always neighbors on the timeline and derek you just recursively used the metaphor by saying raising children it's it's nothing exactly yeah that's right well thank you so much david uh i really appreciate the talk it's an amazing talk and uh amazing ideas and uh i wish we had many more hours to chat but thank you for for uh your your talk and answering questions thank you very much everybody thank you derek and laura for inviting me and hope to see you at some point at sfi virtually one person but um yes much appreciate it okay bye bye you

Channel: Cabrera Research Lab

Views: 4,339

Keywords: cabrera, knowledge, dsrp, thinking skills, think, thinking, critical thinking, systems thinking, creative thinking, education, podcast, emotional intelligence, metacognition, intelligence, mapping, academic research, thinking conference, cognition conference, cabrera research lab, cornell university, think x, military, military science, complex, complexity, complex adaptive systems, CAS, learning, learn, simplicity, santa fe institute

Id: vgrZR81QYH0

Length: 57min 43sec (3463 seconds)

Published: Tue Jun 07 2022