How to coordinate an epistemic revolution by Nick Almond | Devcon SEA

[Music] morning everyone uh bit quieter after RAV isn't it um I hope you've had a lovely week um I've had a wild time and I thought I would take this moment to talk about some of the bigger ideas in crypto um I got very excited about this industry about sort of seven eight years ago for for a few things and I thought it'd be interesting to sort of revisit some of the the the sort of really interesting ideas part of which is I think it can coordinate an epistemic Revolution I'm going to tell you what I think that means so I'll dig into what um epistemology is and and kind of um what I mean by the title and then I'm going to jump into like prediction markets and info Finance some mechanisms for markets some mechanisms for our minds uh and disc discuss a little bit about the future of intelligence so I think we're in a moment of epistemic Crisis and by that I mean I think people are losing their minds um I think people are um starting to lose grip of reality and I think we're going to start to see this effect manifest um over the next few years um and what I think is happening is there are kind of epistemology is being corrupted now epistemology is essenti like a uh a lens that you put over the world that views that that shapes and biases your reality so you have a concept of reality and then your epistemology is what you consider truth within it um so you can consider this it's like you've put on you know when you put on Shades right and it changes the world that's what you your epistemology is like that but it's the kind of the entire history of you your biases who you are what you know what you don't um and everyone has one of these has anyone's seen they live it's a brilliant film it's a banger watch it uh 1980s John carpet film uh and basically uh it's a story about this guy who kind of uncovers a plot uh and the plot is like these people are trying to distribute these glasses when you put them on it unveils the truth uh and he realizes that everyone is being brainwashed uh and it turns out there's kind of like these ghoulish aliens have taken over and secretly in control of everything um and interestingly this is kind of what people think is happening now people are collapsing their epistemology to a reality where there's like a bad guy um the world is not that simple right but people want you to believe there's a bad guy and the bad guy that they want you to believe right and that's why truth is under attack um and I think this is like a critically important moment where um the kind of tools that we're looking at are just literally brainwashing our our minds right we are being brainwashed I know enough about algorithms and social networks and mechanisms to know this is true right um we have being brainwashed every day I don't know how anyone looks at Tik Tok for longer than like 10 minutes it's insane uh but these algorithms are like changing our minds and they do that because we are kind of like we are like we are socially constructed like I used to work in learning theory for a lot of the time and um this was was a big eye opener for me in terms of like how we learn right we used to think that you would just learn by just being told stuff uh and it turns out that the best way to learn is to learn together right we swap memetic units we swap memes to each other that we incorporate into our world viiew people can literally change your entire view of the world with a with a sentence right so we learn from each other and that's happening all the time um and increasingly in the modern world the kind of delug and the information complex that we're kind of under um is making us sad and upset and is really wrecking our heads in uh and I think this is a big problem um the feeds are the new Battleground um these two guys have kind of used these things um to their own gain in in numerous ways uh and we are in a world where we're about to take the breaks off um a lot of the sort of information World Donald Trump's talking about like making it illegal to do moderation basically uh on social media networks uh and this is a big problem right so you've probably noticed that X has just turned into a complete like hole over the last like year because they've just taken the breakes off right and it's just like top grifter uh every day on the feed uh and the most kind of like um dopamine inducing craziness um and the problem is it's going to get worse uh and it's going to get worse because of AI um we now have basically set up a pipeline of truth that's served up by an API um someone asked Rachel rosol yesterday um like what she thought of AI and she said it she thought it was like a demon that's been Unleashed uh by leeching on our uh on our data that's been scraped from us and she's right um and like the the reality of it is this thing has been basically just like the whole history of human language but in between it there's a policy filter it's a new epistemology for the world.org right um this is a A system that is filtering our reality and we're Outsourcing Our Truth to an oligopoly and they're Waring with each other for us right to shape reality so how do we coordinate an epistemic Revolution where we sort of create a new lens on the world that is closer to the truth um and to do that we need new sources of Truth we need um new information to look at the world that's more objective more empirical more evidence-based more real um uh sort now the technical bit right um prediction markets are part of the answer these have got very hyped recently I sort of um been working on these problems for about 10 years but I joined crypto fulltime four years ago to build a prediction Market except it wasn't really a prediction Market it was a little bit different um but the reality of it is there's a lot of work to do we've got a little with the hyped cycle is in full effect or just was because of prediction markets but very incomplete mechanisms in my mind um the oracles are very optimistic um and yeah like really optimistic and the reality of it is prediction markets put a bounty on the generation of new information and if we get this right then there's a large enough market um that the oracles are created for that market that actually have to research which reality we are in um so there's a lot of work to do on the oracles um we are we need to get better at curating predictors like plutocratic um approaches to prediction markets is like the one coin one vote of dows um there are better ways for doing prediction markets we need to create out who are the good predictors uh Crowds Are Not wise but there is wide people in them you need to find them um we need to get better at asking questions we've had one big prediction Market because of one question that was made for us um and you know they died four years ago cuz that was the only question that anyone that was found that was interested in this Paradigm um so we need to get better at asking the questions and we need to worry about incentive compatibility and this is a kind of game theoretic idea where there's like is is everyone revealing their true preferences to the market via the incentives and actually there's people in that game like market makers and Bots and airdrop Farmers that are messing with the signal um and we need to worry about the Dark Side of prediction markets some of these things have pretty dodgy and weird outcomes that uh you should be worried about um and we need to find ways of stopping that from happening and liquidity is the major limitation right we had one massive liquid prediction Market this happened four years ago and then it went dead again right and that was because you need a certain amount of liquidity in the market to get out what you need so I think info Finance might be the answer and what's interesting about this is vitalic has basically exploded the concept of prediction markets created a new idea called info finance and underneath it there's a whole design space of things that you can use incentives for generating new information um I designed one of these about four years ago um little one of the first mechanisms I'm going to show you um essentially you mint an ID uh the reality of it is there are many realities in front of us right a lot of prediction markets essentially have this binary Choice um so instead of voting with who's got the most money this is who gets an identity you can express a preference across a list of realities in this case a bunch of Market assets um so you quadratically vote on these sort your preference and you go into a kind of um like competition with other people so every identity essentially votes on what they think is going to happen in the next day uh we use market prices because that's what we actually have oracles for uh and then we find out who the winner is um and essentially what you get here is a kind of history of uh everyone's predictions over time uh what we built here is a permissionless protocol if everyone anyone wants to hack on this at like eth Global join this group or just contact me after um but essentially what we've got here is a system where we know exactly who made good predictions over time on chain um and now anyone can interpret that data and find out who the good predictors are so the next sort of wave of of prediction Market mechanisms need to go beyond Market mechanisms right because they're not that incentive compatible we all know the amount of dodgy stuff that happens in amm there's me there's slippage there's all of these things uh that are impacting the collective intelligence in the system uh we need to start thinking about reputation systems to find out who the good predictors are um and we need to incorporate like prediction histories uh and and think about sequentialized prediction markets there might be an absolute genius in that poly Market uh Market who knew what was going to happen and it kind of like there was one guy who literally went out and did on the ground ethnographic research and put $4 million on it and made tons of money right that's the guy you want in your next prediction market and you want to amplify their signal right there might be people that smart who don't have $4 million to AP into like Trump coins right so how do you find them um also we've got like AI as players in the game right another vitalism but this is definitely going to happen we're going to get AIS to play in the game and look at how everyone else plays and aggregate the intelligence across the market and be a player themselves uh importantly we need to validate this stuff scientifically um because people are already thinking well like poly Market is truth now um and in fact it got one thing sort of right um and there is things to it but like we are far from proving that these and Know No One Believes In collective intelligence more than me but we need to empirically prove these things as as as as real and valid and not get too overhyped about them uh and we need better articles and better tokens um here's a less talked about prediction Market Geo bodden token um now is this a rug or did it predict brion's resignation like months in advance um the reality is we don't know um because most meme coins are like bad memes generally uh and like nearly all rugs um but they could be prediction markets right if the meme of Biden was actually a true meme of it and not Geo bodden token and its market price was reflected to what was happening in the relevance of Joe Biden over time maybe it would contain information that would be super useful so actually we need like an infinite Garden of these mechanisms um which yes this is becoming a kind of like Zeitgeist thing right I believe in infinite games and what we need is many many new mechanisms all trying to tackle these problems uh but to get to that we need to completely fix the market structure now I don't think I've got time to do this because I need to talk about decentralized AI um but basically we've got a whole load of problems in the market structure we need to stop thinking tokens as Equity we need to find a line Capital uh and frankly the way the market is structured at the moment keeps out not just an infinite Garden of mechanisms but any mechanisms we don't have any apps not we had one app that's done well right so the market structure is completely broken this has annoyed me so much um that I'm trying to basically like smash this to bits with a bunch of like tools for like opening up um the market structure I there's no reason to me why projects can't be launched completely onchain all the way through so completely decentralized Capital formation all the way through to token launching and like not touching the real world in any way I don't see why that can't happen um this is a bunch of mechanisms uh that I made for these things this one's quite fun um and that one um but yeah I'm going to Swit on but if you're interested in like mechanisms and how to like break uh apart this kind of like Market structure uh join joers in that group uh I'll be going through kind of a range of mechanisms for how we can do this stuff um now I've got about five six minutes to to talk about AI uh this is something I believe knowledge is the most important commodity on the planet um and the sort of project that I'm interested in working on next um is called fact um and I'm interested in Building Systems about the deliberation of Truth um and how we might go about doing that so the mechanism I mentioned earlier um is a slightly more broader mechanism I consider it a kind of core primitive in in my sort of design space and essentially I call it semantic ballot voting essentially you have like a list of tags anyone can s uh put a tag that's a meme it's a uh unit of meaning it's an ontology um it's it's whatever you want it to be could be a proposal it goes into a list hence a ballot and then you vote on it to sort it by preference um and then essentially that allows you to aggregate the preference of meaning across many participants um but the interesting thing is is if you give those people an AI now to now talk about those hashtags you generate a really complex data set which is actually a kind of semantic graph and the mind of a da um this is like Wikipedia for AIS right it's a complete mess no human could be able to read it but it's absolutely lovely for LL um so now if you just imagine a da um like a good one uh nice and simple u a community of people um it's token gated you can only see what's inside if you're in it um it's Sovereign it's decentralized um it's social and consensus driven um it's a community of people and they're all equipped with tools to govern an AI many of the tools I was just talking about uh and each other right so they are basically in a group and they're governing a model together so not one that Sam's governing one that that group is governing on a topic um and in this group um those people are basically just chatting they're curating they're governing and all of that is their data they own it they control it uh and they can monetize it if they want but essentially what they what they can do is then leverage an AI system to create something that's a bit like a hybrid of like Reddit and Community notes right but in that specific topic so mimetically anchored in an ontology uh and what they're engaging is is a kind of like what I call semantic governance which is um essentially debating meaning um so I think this is potentially a new kind of name space uh where we kind of build this kind of like conceptual playground where people can connect it with other bits of meaning and you create a network uh so each of like these semantic nodes represents a unit of meaning um and then we can use this is this actually a great governance mechanisms for Dows by the way um because Dows typically have this perfect open scope decision making where anyone can make a proposal about absolutely anything which doesn't happen in organizations uh organizations have a scope and they bound their decision- making to particular themes and ontologies um and we don't right right so what we need to do is like semantically anchor what it is we do and what we don't do um and that will kind of help us with Dows um I also think this is the approach for um unifying expert systems which is basically the one like super creative thing in experts an AI system has ever done is move 37 um in the game of Go uh where Alpha go beats uh the world's best go player it create it played in a way that no one has ever done before and llms simply do not do that um so what we need is to kind of unify expert systems through these kind of neuros symbolic approaches um and you know I believe doing Dows in practice right we do lots of stuff you know I think there's like Dow governance uh use cases emerging and like protocol Dows and defi but it's really time to start taking these to the people um so we've been trying to like onboard people in citizens assemblies uh and give them access uh to a quadratic voting app and getting them to create these data sets and our goal is to create um a agent for those people called London voice and it's kind of a digital representative of those people um it's difficult like Real World Adoption is hard my wonderful co-founder Matt sent 50,000 invites to the people of London to get them to a Citizens assembly and we got 12 hits um it's really bloody difficult um literally Flyers through their door with an invite um 50,000 it's hard work but we need to do it um so really what we end up here is like a a a a tag a semantic tag an ontology London what does London mean the people of London decide what London means and they create a data set uh it's rather sci-fi sounding semantic hypergraph um but essentially what our goal is is to create a voice uh for those people and I think this kind of ontological engineering is what going to make alignment legible um so using this kind of tools you'll be able to find out who is aligned with you and potentially this is the way that we can do new kinds of public goods funding um so where this goes next is basically um you need a governance system for this kind of stuff there's some of those tags you probably don't want on your network so this is a kind of swarm of Agents so imagine a swarm of Agents with a group of people with each one a kind of Federated structure uh that kind of governs that Network and a sort of deliberative council structure um which essentially grows exponentially and this scales by the Fibonacci sequence so you keep adding these communities to sort of decentralize the overarching governance layer uh and what this gives us is potentially factually accurate Generations because you can uh root queries across this network of experts and synthesize queries generated by genuinely independent Minds so what that means is that every concept can have a dow there will be thousands of dows a day and every Dow will have an agent um so really what I think we could end up with is a kind of social network of AI powered Dows where each of these communities um uh are are building their own AI so what do we need we need more mechanisms we need more Innovation we need good token economies and we need to leverage AI that we control and govern in our own Dows uh and we need to do what we need to do is build systems for planetary scale decision- making to help us get find new spaces to figure out how the world is working thank you very much thank you so much Nick I see that the intersection between Ai and web 3 is a really hot topic because we've had so many questions about it so let's start with the first one what's the metal layer risk or benefit of an AI model guiding the scope of decision making that's a very interesting question um there is risks and benefits you're quite right um so I think the benefits is that we can build like like the models should be designed by by The Taste and and needs of that group right so we're creating a kind of new epistemology for that group a shared framework of reality a way of looking at the world a way of synthesizing information uh and that pluralism that difference in opinion in the way that data is rified to a group of people is incredibly valuable and in fact is the point uh the risks are is those people might be mental right and they might do some absolutely crazy stuff that you just do wouldn't want perfectly synthesized in an AI and like the prediction Market things you need the governance aspect on top of it moving on to the next question how do you think we can use AIS in a more Collective way such that it protects us from the dangers that centralized AI will likely bring what could their governance look like um again wonderful question and and like this is exactly what I want to be EXP exploring probably for like the next 10 years right at the moment AI is a very sort of like um unidirectional it's a onetoone relationship uh We've yet to even crack open the design space of what it looks like for our group of people to leverage and uh and and an AI and how they do that is through governance it's like for me it's the perfect use case for Dows right it's a killer use case for Dows and I think the um that is the question precisely the question that we need to explore and crack open um so really what you need is these pluralistic decision Mech mechanisms and lots of like deliberative decision- making and building data pipelines and then finding ways for people to own that data and be rewarded by it the system I'm designing I want to be able to give people to pay people per the generation as their llm is used we need to find ways instead of just stealing people's data uh and monetizing Humanity without their consent we need to find ways of people doing that yeah I really like that idea of building the incentive structure in to make sure people are being paid for the work that they are doing exactly exactly all right is there a scaling prerequisite for the AI models to work does it work with just 12 people are there any self- selection bias issues really wonderful question these by the way um actually you'd be surprised right a lot of the feedback I've had with this is people going oh you need like billions of tokens for AIS to work not the case at all actually you can can actually get incredibly good outcomes with like 12 people especially if you like this is like one of the core things right because what we've got is this incredible like block and Deluge of of data uncurated build from the internet that is what these bolt models are and they're all like fil filling it up with like we've got trillions of data points it's absolutely brilliant and it's like all garbage right and and like the reality of it is if you have highly curated data set that's structured in these kind of like ways that the people have carefully fine-tuned it you can use things like rag which just brings out the data that humans have curated for it so actually what you get is this like hyp specialization I think you can beat GPT on topics like I asked you know when did Biden resign for that go Bowden thing and it said Biden's still the President right like the reality of these things is that what we need is realtime AIS and that's what you can get with 12 people you can curate a topic and say What's Happening Now basically mhm and we have a little bit of time for the final question um so they were asking is the data factually accurate or socially accurate because social consensus does not necessarily equate the truth thank you for these questions absolutely wonderful all of them um the answer is is like truth is a social consensus problem right like it's a kind of uh meta stable thing that means that when it can be like Disturbed and we switch into to a new equilibrium like we used to think that Earth was the center of the universe right like facts are a social consensus problem and what we need to be able to see is what do people think the truth is and be able to look at the deliberative Providence all the way back and have that be your judgment call on whether it's truth or not all right thank you so much Dr Nick let's give a round of applause to him our next talk will be on