Combinatorics and cryptoeconomics

Transcript

hello hello that's a better crowd than I expected thanks very much thanks for staying around so late in the afternoon I'm Oliver I'm a little bit about myself three things about myself which are important about this talk is I use I used to be an industrial engineer that did operations research in the in the early 90s wrote a master's thesis on a topic of neural networks in in combinatorial optimization and and especially on the topic of machine scheduling and after that during that during a short blip when when neural

networks were kind of like a hyper topic in the 90s then I moved moved to the US just around the time when the internet started and I decided to switch sides from from working inside the company to thinking about how markets work especially how markets will be changed by the upcoming internet this is working better okay so I did my PhD on putting putting the network in network externalities and I went back into industry went back to Germany for got a park forgot about economics for a while and then sort of like the Decrypter economics boom started which really saw

it brought me back into seeing that like a lot of the questions that the emerging Internet era the attention economy asked have not been really addressed properly by economics as it is and crypto economics is really sort of like the for me the distorting point to ask for very very fundamental questions about the how the economy works and sort of I start I start a little bit of a program and this is for me a bit of a kick off in a way of like trying to understand how things changed since the mid 90s and how we could use crypto cryptography kryptek

anomic stew get things right and this is so like the first talk it will be 90% economics 10% crypto economics just as a warning the other thing is will be not not a math talk I was trying to make it up like keep it on a very conceptual level so I hope to get everyone on boards not only for for a techy years and the math people so I'm trying to try to talk a little bit about two things is sort of what are the fundamental questions a crypto economics and the crypto world asks of economics and on the economy and then for me the second thing is sort of like how can I extract the things I

learn from this and so the right a little bit of a what I was like set up a framework of like very very fundamental building blocks to understand how the economy works how the economy worked over the last 25 years and how we you hope that you can change some or a thing so what this is the first first of talk as I'm expecting a couple series the first one will be about how we think about space and structure within economics and how it is sort of like blocks of view of things we really need to understand about economic organization right so I'm a wonderful

wonderful little story by Herbert Simon a Nobel laureate I don't know if you know him is that was a political scientist and it was an economist then he was a computer scientist was like one of the fathers of artificial intelligence and also famous for like doing very fundamental things about economic organization is the the inventor of bounded rationality for once so yes like this this little little story which goes like imagine you're a visitor from Mars if you have the ability to look at Earth from space you have a telescope a very special telescope that doesn't show you the wall is but it actually shows you the social structures what are you going to see so

first thing we see bubbles green bubbles for some reasons Alma green bubbles emerging their firms their organizations they are structures that were bound people in a way so on the next thing that happens what he said you don't really see is you see market transactions in between those bubbles there's a little bit of mistake this is actually not really market transactions yet because market means you aggregate things it's right now it's trade transactions between those bubbles all right and the next thing that happens cracks appear cracks are the lines of authorities this is hierarchies and

there are many many many types of organization but it didn't almost all firms almost all productive organizations tend to turn into hierarchies and the next thing you observed is so if the Martian was basically sending message back home what do we say like it doesn't matter what society you in if it's if it's capitalist society if it's if it's the market society a socialist or anything else almost all the things you see is like mostly green bubbles and very few red lines and it's like why is that like why do people like to hide I remembered this quote when I thought I'd like thinking

about a crypt economics is really sort of like the goal crypt economics is in my way to in in in my interpretation to flip this around to think more about green lines red lines sorry and and shrink those green bubbles we don't like hierarchical organizations too much we're looking for participatory ways of organizing and we're looking for ways to protect those red lines so they're not getting sucked into the green bubbles so this so my starting point to think about this there's one thing missing anyone noticed something that was missing so

far the cracks there were a couple more cracks many ways to getting too lazy ya know looking at like us one of my most famous quotes within the women the like economy of the post-world War is Friedrich Hayek's claim that we need this organization to ensure that knowledge base II drives production right so we have to make sure that the people who know things about what took going on these points are the people that actually make the decisions so what's

missing really here in in Simon's wall is flow points we're not címon címon swirled is very static so and one of the things critical economic started to do always like talk about disintermediation meaning that we have a complete graph between all those bubbles which most like will not work out this is a really simple thing that you have if you have multiple steps between between you and and receiver usually there will be multiple parties taking over so basically the question we have

to ask yourself as a question of what I call value chain integrity and there's two things to to look at this is basically the key question is like if you have a flow from its origin and it's a transaction flow or goods flow information flow whatever whatever we are moving around within the economy does anymore everyone within this flow get the return that reflects their contribution and I could go like this problem can go and go in in two directions and it was like okay you have someone in the middle and this is very like very common in

in in in supply chains that you have only one bubble extracting all the wrens all the returns from this process is like the like the middleman problem usually or the other way around is so you're basically you introduce spoilage within this value chain and you're only at the very end of the value chain you notice something went wrong within this value chain you cannot figure out who and introduce this problem into the Shana's like why the the big thing that that cryptic economics did is really flip around the audit trail the audit

trail we had so far this was the the green bubble audit trails this is common corporate accounting which is basically creates this bubble it basically creates a boundary they sell a business everything is within within the the organization this is accounted for everything that happens in between we don't know which means we cannot we cannot really say what happens here to here to here it here we're losing connections and this is sort of really one of the biggest biggest contributions of satoshi by offering this shared audit trail that we can actually start thinking about how did how to trace these problems so this is my interpretation of what are the the

keys or key contributions of the crypto world to economics so we really really started looking at things from the opposite directions we're no longer thinking about bubbles green bubbles that are protected from the outside world there are hierarchical they're bounded and they're kind of accountable which is the problem but we can actually think about how can we protect this red network and how it can we set up a networked world rather than hierarchical to questions we have to ask yourself Camas like we're not gonna get rid of

the dirt green blocks we just try to make sure can make sure that individual green blobs cannot take over the whole value creation or get all returns on valuation so on the other thing is of course there's a more technical question how do we protect what using cryptography the integrity of of the value chain red light okay so the problem is economics as it is I'm not sure like assuming everyone has ever

taken at least one economics law so we know like supply and demand so I have don't have to go into this detail economics has a very very particular conception of space and structure sorry yeah economics has a very particular conception of space and structure and this is like conceptualizing space is very intuitive once once you think about certain things in in a certain spatial ways it's very very hard to snap out of it and I think this is like probably a much much better cut try to talk about those things so this creates problems so let's have a look at how we think of choice space within economics it's not like we're the very simple things you

have mine are choices between bread and potatoes or other things we have other feels too in technology famous examples from the 1980s oh you're gonna use a VHS video recorder or a beta video recorder there was use any early early papers on on network externalities economics also a technology topic are you going to use the the QWERTY keyboard or you're gonna use those horshack keyboard also used in politics so are you progressive or liberal or are you conservative so this is like the most simple way to to think about choice to simple and someone came up okay is actually this is a false

economy it's a false binary there's actually aligned as moderate people in between right and so we can actually have a distribution of of people's political views over this line where the distribution of progressives distribution of conservatives so you have a one-dimensional line and we can continue this go on into multiple dimensions we have a conservative bubble progressive bubble a few a couple of here people's their liability area north or eight Aryans and we can continue this from the two-dimensional space this is

like undergrad economics if you take an undergrad economics class you know everything happens in in a two-dimensional space and if you go to the PhD level we go into the n dimensional space and we're looking for ways to separate this n dimensional space into cut hyperplane so we cut trying to cut through this space with a line will a budget line or similar things right so really it's getting mathematically much more complicated but in the end we're still doing the same thing we're just moving into into higher higher dimensional space but I'm trying to look for for a good example to to show how this conceptualization of space

loses your information and losses you critical information at the wrong point in time so there's this famous paper by philip converse about belief systems of mass publics and like one other things he says is a little bit paraphrase here is that the belief systems are a knowledge of what goes with what and this is sort of like information we're dropping when we put everything in a space and from the get-go and this is sort of the information i wanted to bring out again by using comm unit Oryx and

combinatorics translated means networks right so what you're talking about this one type of network was a hierarchy which is a search tree and the other type of network which is these decentralized model of network and we want to find out how we can Marvis so now in the world of combinatorial optimization there are a lot of operations research problems are being done with the community of the optimization it usually means that you start with discrete choices you have and you compiling information or only on what goes we want but also of course afterwards what I have to go together

famous examples is the what can expect which systems can you come up with wrote through seven breeches in in königsberg these are their problems so machine scheduling is one matching assignment parcels and contracts which is of course of interest to crypto economics things that go after each other are always interesting to community oryx routing problems branch and bone problem search tree problems or combinatorial and as well we're starting to get interested we can talk about network structure and network boundaries so we can actually go away from thinking like like herb Simon did as every every transaction is market

to there's a network of transactions and we if we aggregate them they turn into market so the other thing is the last thing is something I did is called unit choice inference and the thing is if you take any of these mostly solved operations research problems you have here and you add coordinated in front of them so what is coordinated machine replacement another word for that is innovation right it's basically it's a easy easy for Vitaly kufi wants to decide he wants to wear whatever replace his laptop but if you placing your laptop influences you for

some reason or whatever switching switching from one messaging client to another which is like the tipper example right so like you're losing connection this is the type of things we're considering this is sort of why call me no turrets it's really the right tool of thinking about how to solve coordination problems right everything else matching assignment uber this is all over and tinder problems search and routing is of course uber - where do you want a market is not all the all the cars all the uber cars in San Francisco a market is you want to go from the mission to Castro at Friday afternoon

right this is your market you only interested in the cars that are willing to to bring you there the drivers so this is like all questions where we start thinking of a market not as they is a an aggregator thing but a network of transactions so does the tool helps us what's happening in economics now - if you want to design a economic model you go through a couple steps usually you have a real-world problem is fairly complicated you want to simplify this you choose your your vector space everybody has in as the same vector space now you make some

behavioral assumptions typically I suppose a rational actor assumption but you can actually changed a lot and you lose each is resolution criteria which usually means a Nash equilibria or any kind of equilibrium criterion writers economics gets criticized a lot on those two things that we think too much about rational actors and we're thinking too much about about sorry actual Librium but it's really the thing and unselected look the last thing is how you bring in a special circumstance that I make your your model publisher and I you see it as a bubble missing and I was divorced with what bubble we sort of ignoring the special space and structure we're in and

then what only if you do that we can make a decision if it makes sense to go into a plate in vector space like normal space we think or we skipped a step completely and basically try to model economics within a network ok assume this was a little bit too abstract now so let's try an example I picked an example that's probably on almost everyone's minds which is like go on right now this is basically American political situation which is very much in turmoil right now and said okay we have a starting point is our

one-dimensional policy line which is used in in in in almost all publications from liberal to conservative but we can actually take this apart and the interesting thing is about this is not that this exists but it's actually one-dimensional right the interesting thing is that if you look at very very different issues we're looking at abortion gun rights free speech or other things like more local things people who have a progressive opinion and like these are all individual issues people who have a progressive opinion or something will most likely have progressive opinions on very very

unrelated things this is the what goes with what right the same thing conservatives tend to be conservative all across the board all the kinds and then we have the moderates in between and the thing is like what's like what is happening here in in the middle and there could be two two explanations one is like what's happening in the middle so there could be two things moderates could just have like weak preferences between or they could be inconsistent right they have progressive opinions on one issue or conservative opinions on the other issue

so they're really we have mixed in the middle and therein consists and about their their beliefs or because we are sort of facing a situation that American electoral politics gets more and more polarized and this might offer an answer to this because if like say if I'm a progressive I'm gonna agree with this guy across the board mostly if I'm conservative I'm gonna increase agree with this guy across the board if I moderate I might actually be couldn't uh progressed on this one conservative on this one progressive that might actually not be basically in agreement with the other person right the other moderate so

we actually saw people are progressive share the same beliefs people are who are conservative share the same beliefs people who are in the middle can be can be across the board so this could explain why electoral politics in the u.s. dries towards towards two polarizations right so something I did so you can see is like very old very old jpg it's from two thousand something in like a couple of hours for a term paper on a political science class I was looking at how the US Senate was voting in early 60s and you see like okay

here's this is actually from progressive to to conservative here so if you have progressive senators tentative vote weeds the other we have a middle and then we have a couple of conservatives Republican senators voting in the middle you see a couple of stripes here this is like the Southern Democrats but you have some sort of kind of continuity the same thing for the Senate back then and suddenly you have very polarized voting and it's like at this point like

everybody assumed all a Polock assignable reassumed okay we're still in this world in comparison to like the early nineteen hundred's voting was was not very polarized there was a lot of lot of coalition formations between the tooth voting bloc that's what everybody assumed and so there's like it's the first inkling that it was actually something going on which we didn't quite understand the question is how can we see this if you're moving this into a a uni-dimensional world just one policy line and I'm going to use so I forgot to source this so I posted us on Twitter

it's a paper from 2015 that sort of did the same thing a couple years later and you see sort of they're kind of splitting up eventually but it comes out fairly late right it still 999 is still sorry sorry here okay this is actually starting to separate the two parties are starting to separate so they did the same thing with kind of a network and realize that you started the separation becomes clearer much much earlier right so this is the one thing we see if he go on from the one dimensional space to

what looks like a network but it's really a two-dimensional representation so they basically decided we're not going to look at one dimension we're going to look at two dimensions it's getting more obvious what is missing here and it's just like the crucial thing this is like sort of the thing you have to understand about American politics that American politics is mostly uni-dimensional in most of the cases most of the times their mix but in sort of you can still tell what is progressive what it was a conservative for most votes but in times of crisis there's a second dimension emerging or something if you decide beforehand that

you're looking at things in a union or medicinal way or in a two-dimensional way if you're pressing a network into two dimensions you're losing this information and what is happening when a second dimension emerges is a party split up - is exactly what are we seeing they're not splitting up in in in sort of like deity the major ideological way they are actually splitting up on issues that cannot be captured with from it's completely new usually new things create new tensions and this is sort of if we can do this we can actually model on real data to

economic organization which I showed earlier basically the bubbles and the lines this can see it has a blue levels and red bubbles in this case here we can separate and we can actually answer the questions like how many dimensions see we do we see what drives this separation you're basically watching right now sort of like the the Democratic Party splitting in two we kind of watched the the Republicans split in two after Mitt Romney lost his presidential election and then sort of like what used to be the Tea Party is basically taken over the whole Republican Party and it sort of it it is it's it's moderate wing

instead of now that the Democrats have the situation where they have to socialize and the moderates and and and sort of the former Republicans and they have to find it find a way to to bring them all under one umbrella this is sort of really what case yeah the pension in here and this is something you miss out on if you start looking at things on the Euclidean vector space because you're dropping information at this point right little site no this is something also most of the empirical work on on politics used to to use issues scores based on voting behave

yeah published by Singh tangs usually ever liberal think tank they want to spread all the Democrat Democratic caucus you want to say okay this is the whatever legislators closest to us and is the one it's just for is away and they also want to like press the other party all in the extreme corner I see on the other hand as there's also conservative think tanks that do the same thing and then you certainly have the situation that you have conflicting spaces and this is something because it's something called perspective bias that you have this tendency to see things that are closer to you more

spread out more pronounced than the things that are further away from you and this is sort of part of the most important tool to analyze what's going on and in politics right now so if you're say you're your moderate year sort of okay I would I like to vote Joe Biden I think it's an it is too extreme for me but if whatever if by lose I vote for Sanders but I hope a that people who are supporting said let's do the same and I amazing in here okay Sanders people say oh no I like Biden and Trump may see the same they change so this is really a perspective saying right so people tend to to shrink things that are

further away and this is sort of really really happens in our perception if sorry if you drop this if you sort of move right into a Euclidean space to model your yet scenario this is something you miss out on this is sort of why can you use networks and ryu is coming at oryx to understand what's going on here so I said ideology politics tends to be you know dimensional almost all other preferences space if it's music or film to something or so usually very very complex very high-dimensional is sort of created the whole industry of recommender systems from Amazon to Netflix everyone who is basically trying

to figure out Spotify every it's trying to figure out what is if you like something what are the other things are in that neighborhood that you might like to write this is very very important for them to always feed you with new things and have a high probability to solve this and it's the interesting thing enough is like it I've been watching this this industry for 20 years it's pretty sort of this is the ones that end up using mostly neural networks because sort of like spatial economic regression models didn't really help them turns out they're doing it using they're very complex mixes between

the two but this is sort of the problems you have when you start with a choice model and I don't know what the time is wrap up quickly so I do I run through a couple okay good so I run through a couple of other other examples quickly so every paper on this technology standards discussion was discussing the the QWERTY keyboard which everybody uses versus T is this alternative keyboard called the Dvorak keyboard but how does the space look like what is in between turns out there are if you're just using a twenty six letter keys there is less

like four hundred septillion or something ways to place those 26 keys on these three lines of a keyboard alright so why did we end up with mostly this there's about 20 20 keyboard designs used worldwide about 10 or variations of QWERTY the other ten are our alternatives how did we end up with those 20 out of this situation and you can say okay like yeah there's a lot of that's a lot of keys but all the line has infinite points right there's actually easier but you cannot just simply go from here to here and say okay

what is like what does it mean shifting on this line from here to here so how do you resolve that is two ways so basically do you do combinations you're flipping keys around and you're always interested as there's a couple a couple of ergonomic criteria how this works are basically not using a pinky finger of a frequently used keys switching hands what left and right so you can measure performance now and you but basically you have to flip out just go along the line and do marginal analysis to find the peak points you have to actually switch or the other famous example is the shelling

segregation game which is also a way to switching to reduce friction in the system which in this case is major performance criteria so if you if you take this typical economic definition of Technology it's just it's transfers the the factor in famous factor in post labor capital on soil into products and so it's just a production function but if you change that definition to what we've just seen it's a combination of features that need to work at conjunction and most of those conjunctions most of the combinations

just simply don't work you have to figure out the ones that actually work then we are in the uncommon entire world once you there almost everything turns into into your technology language is a technology is basically you can take a word you can switch out letters like Hamming distance or something it's still comprehensible this happened over I know when in the European came around like 3,000 before crisis or something so 5,000 years a lot of singles were changed to the point that so many language started to emerge that are now mutually incomprehensible so this is

like really long-term switching causes we see this over and over again is also a refutation of this idea that because you have positive Network externalities and languages have positive Network like Excel IDs I hope I can communicate with you right now because we speak the same language even if it's our second language still you have this conflict between common shared interest of communicating and individual preference and individual preferences usually come from the way people's throat boxes are shaped right so this is where dialects

come from all right so yeah these mechanisms and you can actually use this for almost anything once we're talking about this once this is a technology and market as a technology because the market is anything and it's like talking over you know no longer talking about right cheering you're talking about going from from this spot to this spot at this point in time and this is only the interesting other interesting people in the same spot are your market so you have a lot of combinations and this is why those matchmaking platforms emerged because this is a computationally hard problem right and this is basic a key problem economics has because a market

should be a market and a market should not be an enterprise ok well you know economics as it is post-war economics emerged to solve production problems and this is sort of right now it's still doing fine in this area we have two other problems this is a typical production function so this is where decreasing returns to you to your input start to other problems we have this is the one that's most important in crypto economics is the problem of governance this is one you have in your situation that you have many many inputs and you have to hear equitably this is where any

comment oil and the other one is at the the lower end and this is just like the the the second part of a start-up hockey stick here right this is entrepreneurship when you have so many combinations of things you could possibly do but only a few of those combinations work and the only way to solve this is trial and error or community riyal and evolution and we can continue this production if you if you dis a great production into supply chains you're in in a combinatorial world markets said any any kind of matchmaking market is a combinatorial problem so we're actually moving or removing the big thing that happened is

a massive amount of data coming out of Silicon Valley that showed that things are not quite the way economists assumed and this is sort of why we're in this this questions hey this was awesome maybe an abstract question as well to this like I I really like how you approach that you can see these problems coming out Oracle and then like by iterating through all the combinations we find that but isn't there like a call like a line way where

we can say like in the same way as we know like Oh in numbers we know like there's much more numbers that we cannot iterate right yeah all the all the numbers that we can iterate is actually the smallest set so there's problems that we cannot solve community Oracle with C end and the next step is actually something we cannot promote 8 anymore like all the original numbers like if I think of my is that like your my favorite rock music is petroleum and dimensional exactly I'm thinking more of it like a irrational number at least that's what I hope for and if it's a rational number you don't get there was comic combinatoric right is that is that is that maybe the reason why people use neural networks because they just don't have a clue well your neural networks

the reason is because like it's it's a very complex search space and basically by deciding you're doing things coming on tour er you basically structuring the search space in a way you can actually go through this is what caught what's called the branch and bound algorithm so and a coordinated branch and bound where you have a group of people trying to go through a labyrinth together it's like they usually data set of a lot of horror stories right because you know you not only have to decide individually am I going this branch or I'm going this branch you basically have to agree on which branch to go which makes it as very very very complicated problem this is why we try to use neural networks like try to do anything we can this is why like uber uses massive computational

capacities to solve these problems right so now as I said like economics the way we be phrased it in in in 50s really stall still stalls production problems works well but now we've shown in a situation that we are trying to solve market problems which are actually much much more complicated and we thought they were so and this is like the starting point is okay there is the of course the thing that we have real irrational any kind of numbers but there's still a structure that's like okay up for complication this is why it always try to simplify your real world

into model that you can work with all right so it's like basically simplifying this walled into discrete steps so you can capture the structure it's better in this case than using you and there's like one one anecdote George Stigler famous Chicago economist he was charged during World War I World War two to find the lowest-cost combination of food that would nourish people with like whatever vitamins some whatever and so he came up on a yeah he invented linear optimization in a way doing this when he came up with like a really low number and I was like and it

created huge SPAD with a food scientist who basically said like okay you cannot do this and this together it doesn't work right so people like okay they're getting vitamin C and whatever else but you're not really getting fed and stickler but who cares they just they just want to basically ramp up the price and this is sort of kind of where I started why so we ignoring the structure getting the advantage of being able to use real numbers using the Euclidean space and I've worked for for 80 years of rusting so we're really pushing this boundary now that it's no longer really working for us to model economy so how do you decide how do you know I break

this space how do you categorize you know at some point you need to do something and we know how many basket we'll have because I can go to I guess like totally different conclusions based on how many options the point I was trying to make whether it the politics example you can during a normal time the normal way with a policy of one-dimensional policy line explains almost everything so it's very cool to use that but if you have turmoil that it doesn't work anymore this is the problem right so like and this is what economics should should focus on be able to explain it but this is a design decision I said a design decision that's like 60 80 years old so it's it's really

ingrained so people don't know like economists usually don't think about that stuff anymore but it's sort of I did sort of within was in the afternoon and like trying to solve this one one Trump paper problem I sort of looked at it okay what's going on was what is like how do these things relate to each other and this is how how I came up with these these matrices and sort of a sort of thinking you know what how can i how can i express this relationship between those people are there some general finding about like you know like you know probably there is an optimum yes but are there general findings about like you know for measuring this stuff to rather to say merging these categories or rather splitting them I

mean do we usually start with like few and then build up or we usually start with many and then it's always good to start with a simple small like binary model as long as it by neri conservative progressive conservative model as long as it answers the question you asking it's the simplest way and I realized starting really okay this model this model is not good enough to capture things I like I'm looking for yeah but isn't that like if I start with the you know binary model I can go right to the wrong conclusions and I'm just not aware about it I mean I mean this is like you can tell when when political candidates have the wrong

consultants my impression is like I would start with I don't know based on some assumptions yes the like to have you know many options and then try to merge them okay I think people want to leave so sadly I did a podcast for Sally in German explaining why Nate Silver got the the Trump election wrong crazies base he was basing using a model a mental politics which was really based on a spatial line that worked for long long time and it created a lot of successes for him and I like didn't realize what was going on between between the second

Obama presidency and and the Trump election so he did not change his his political model of course he did his it changed his Pauline model but his sort of this is a very very different question so like once so you start realizing the model you're running does not answer your questions anymore you have to start refining your model and sometimes you're not refining your model you have to switch domain in this case would be like to go from the marginal domain to the comment on toy domain because it sort of really answers you a very very different kind of question that's the question of what goes with what or who goes with who who goes after who what goes after what so basically everything everything you're thinking of

in terms of coordination which is a very big topic in crypto economics you should start really thinking from a combinatorial perspective and that's means networks really so thank you we don't thank you very much