# Machine Learning Resistance for Human Rights on the Blockchain by Santiago Siri (Devcon 5)

- Channel: [Devcon](https://streameth.org/devcon)
- Date: 2020-10-09
- Duration: 32:25
- Topics: blockchain, smart contracts, distributed ledger, ethereum, Science & Technology
- Watch: https://streameth.org/watch/yt-tZ_76AZtOf8
- YouTube: https://www.youtube.com/watch?v=tZ_76AZtOf8

## Description

Visit the https://archive.devcon.org/ to gain access to the entire library of Devcon talks with the ease of filtering, playlists, personalized suggestions, decentralized access on IPFS and more.
https://archive.devcon.org/archive/watch/5/machine-learning-resistance-for-human-rights-on-the-blockchain

Bitcoin’s original white paper description of “one
CPU one vote” (Nakamoto, 2008) shaped the software industry to think governance
centered around machines, instead of humans. Although a fundamental right to privacy
bent early blockchain design toward anonymity, this property facilitates the malicious use
of Sybil Attacks (Douceur, 2002) significantly undermining the governance capacity of
protocols.

The importance of formalizing human identities can be inferred from centralized networks:
major social media platforms were established when they achieved a sufficient level of
consensus over their identity protocols, thus creating a trust layer on top of which a myriad
of social applications could be built. This networked social infrastructure has in turn
facilitated the emergence and spread of multiple borderless political movements.
However, being highly vulnerable to several artificial intelligence exploits such as addictive
algorithms, bots and deep fake technology, the underlying architecture of current protocols
exposes society to political manipulation as well as data theft.

On this talk we'll explore the different approaches to reach a “one human one node” graph,
providing protocol specifications able to serve as a source of legitimacy in the governance
of digital networks and strategies that can keep these protocols free from AI.

Speaker(s): Santiago Siri
Skill level: Beginner
Track: Society and Systems


Follow us: https://twitter.com/efdevcon, https://twitter.com/ethereum
Learn more about devcon: https://www.devcon.org/
Learn more about ethereum: https://ethereum.org/ 

Devcon is the Ethereum conference for developers, researchers, thinkers, and makers. 
Devcon 5 was held in Osaka, Japan on Oct 8 - 11, 2019.
Devcon is organized and presented by the Ethereum Foundation, with the support of our sponsors. To find out more, please visit https://ethereum.foundation/

## Transcript

[Applause] all right welcome everyone this is my name isn't LLC co-founder of democracy earth were a nonprofit organization that basically tries to understand what democracy is in the information age and we have been doing a bunch of crazy Democratic experiments over the last five years we tried everything from liquid democracy direct democracy participatory budgeting quality building you name it we work with all kinds of organizations we work with political parties we were with the parliaments congresses nonprofit organizations and also decentralized networks and here's the thing I can tell you after five years of doing Democratic experiments usually when people go out there for a boat it's because it's a very high-risk decision and the higher the risk the higher the need for when it's a contentious decision there's also the interest in trying to corrupt or subvert or manipulate the decision-making processes over these five years we have also been hacked a lot of thanks democracy is one of the most hostile environments to deploy assistance so we learned a lot and today I want to talk about the core of the problem that is a decentralized networks we know very well what happened with centralized networks with centralized networks there's one company one Silicon Valley corporation that brought democracy and that's Facebook and the recent Facebook growth democracy is simply because they were the best ones at formalizing humans on the web and that means that Facebook today has a database with 2.7 billion people larger than the Chinese government by than any government on earth now they're even considering their own currency so when you become the largest repository of identity with people they consult be straining your face AI algorithms every day you you you'll likely end up breaking whatever technology is present technology democracy's being made with me now when we look into blockchain space there's a tremendous amount of innovation happening but there's one key missing ingredient which the way God in an utterly languages identity humans were are the humans in the merchants we are blocking tools to promise humans over these networks so the problem with this is that it's limiting the social impact of bludgeons blood Chiefs have been extremely good and the financial part of the spectrum of the economic activity the promise of what budgets can achieve in terms of transforming our society beyond finance beyond capitalism out there if we look quickly a group of work which inaugurated our whole space within satoshis paper the explicit mention of Satoshi and in his paper in relation to governance is one CPU one boat and it's basically one CPU one ball this ruled by the machines not by the people ruled by capital not by society and this has no this makes sense in the world of finance for privacy is a fundamental right and this has spent and the design of these protocols around finance and around protecting privacy but at the same time were still witnessing how democracy is being disrupted and very broken all around the world because of digital technology as well so the protection of disinterest and logic based networks who use this this tremendous destructive capacity of this technologist to improve our democracies or to level up or democracies for the challenges of the 21st century must not be lost now when we look in to proof of existence something we all know here in this room is a group of state systems voting with your money whoever has more money wins it's a good authentic way of governance and maybe this makes sense for private companies where you want the largest shareholders with the largest skin in the game to have more influence in the decision-making process and when it's about public goods when it's about public infrastructure not necessarily skin the game is a good alternative to wait the conflicting interests we find within society and even if the skin the game argument was to be put under sometimes investors have scheme in multiple games and sometimes you want some games to lose for other games to win so scheme the game it's not compatible under contract of interest an interesting case this here is in the governance of animal on one of the a GPS you know this is a time line but on the on the x axis you see the different balls of people staking tokens on the a peahen at the last minute a whale appear and not seek single whale pretty much to control of entire election now the other goals became relevant because their way simply waited until the last moment to decide how he she or it would allocate its tokens and bring the election in its own interest so who will stay can work for certain certain aspects but it's it's you know they have the prophetic and way of governments so here's the question that we're facing and this is I would make a disclaimer right now I'm more confused than ever about this problem of trying to fill out proof of human which is whether but we have identified after five years of doing work at democracy earth as the core challenge to sort out in order to understand how we can do democracy over the Internet is proof of human possible under more confused than ever but I'm convinced that we forget this we will get all of these stops that are that have tremendous potential if we sort out proof of human we will get democracy universal basic income portal craley better alternatives to kyc firm electrons even love sri communism who knows a lot can happen if we can do proof of human I don't know I read involved with some we're going to say the party leader in Argentina support Li I have had the increments will suit to found that eight years ago the mobility he tell you by which is exactly like the popular reading when cells and they are running for the presidential election this year or a Lou Young who is trying to make ubi this this is the demand side these are the folks that we don't need a technology like proof of human in order to deliver on some of the promises in order to take watching based technology beyond the means of the finance and start building technologies that and also print solutions to society but here's the principle that we must abide with here here's a very important that a consideration that I think it's great that Edward Snowden in doing what three in Berlin this year he said this and I think was you know definitely highlight of that extraordinary conference give FDA in the summer we don't need to verify the entity we need to verify the right face and it's a very important distinction because if we put our identity on change in a mutable ledger further to zero forever and it bins happen and there's a totalitarian regime then regime can use the information in very dangerous ways so when we're working they were working with a very sensitive piece of information but we - we - prepares us for the future we need to design systems that can help us understand in you know what context in a world of our computation will be far more abundant is today which is far more about that it was yesterday how much computation look in that world and found that world we can create systems and we are sure that we will not be taken away by a is and all kinds of machines and also we must avoid recreate the auto Berg and Facebook for the Chinese Communist Party which are the two largest thanks to privacy our identities in the world right now when they wealth began 25 years ago the work was also this dream of free information of connecting humanity no one ambition during the 90s even though there was a lot of people trying to make money but no one envisioned that something like Facebook will come out out of the way and you know Facebook started breaking the burger ceasefire way before Donald Trump they started bringing the Morrises actions during the Obama election Facebook was critical for the election of Obama the the exercise was not be stuff because well done because there was the Americans using Facebook to elect an American president when the Russians are using Facebook to elect an American president or any foreign powers using this Trojan horse to elect an authority over another country then the very concept of the nation-state start exploring the very concept of non-domestic intervention in each other's affairs is no longer there so we're in a very new political reality and we must prepare ourselves for a world of abundant competition now another computation is showing us something that is a very scary and is that the cheaper information gets the more truth becomes manipulated and I think an extraordinary example this is big Bates he takes all of these pictures of these guys that you see here don't exist they can be generated by what is called a generative a parcel network which ISA is to machine learning algorithms or to neural networks that reinforce each other and learn from each other and once the network is able to understand the properties of human faces I guess the final a chibi is in the room [Applause] actually I don't mind I can talk louder among friends here there might be some spice but that's okay so the weird thing about this technique of Gans is that usually in technology there's this thing called the uncanny valley the uncanny valley and in CGI computer graphics the uncanny valley is that moment where you can see the thing tries to be so realistic that you see it start seeing the glitches rather than their realistic attempts of the graphic or the 3d animation that you're looking at in these pictures there's no more antenna valleys extremely hard to tell if these pictures have been generated or not my algorithm there are some tweaks that you can look very closely that might tell you that this has been generated by others but otherwise we have to prepare for a world where this can be done with any place in video format and also with our voices and the better Gans get the weirder reality gets now this is not necessarily all bad this is this actually can be weaponized and to break existing political reality I mean testing example and effects is the under medium campaign which is in Casa pal called the yang Tao which I would encourage you to check it out and become a member you have to pay 40 light to put to become a member of the yang now one of the first experiments that came out of that is doing deep fake videos about Andrew yes I think this is brilliant because you know in a McLuhan when Lupin said the medium is the message Andrew Young is this guy talking about universal basic income because artificial intelligence is going to take all our jobs and we get to communicate this display exactly what the power of the artificial intelligence is doing with reality itself so the main message defects let's use them to break existing political reality but let's be aware that this can be used this is the space that the world that were gonna be living in the next couple of years another big problem for a decentralized identity were verifying humans on Cheney's serials Sybil's that might impersonate multiple identities usually a technique to prevent this is using reputation algorithms PageRank a very well-known reputation algorithm that gave us Google but the problem with reputation algorithms is that they are really in mathematical terms algorithms of centrality they identify they start making more notes more central than other notes so it starts centralizing all over again well when we've what we want to achieve is a decentralized protocol then there's how we store identities or how we index miss identities if we narrow down identity to a one dimensional piece of information whether it's a user name or an address or or you know any kind of one dimensional identifier whoever has no knowledge of that the list of addresses can become you know can start exploiting this information or anyway so we maybe would have to also think about how we store the information related to identity or how we refer to identity in in digital space so because I'm very confused what I thought I'm gonna do here at Def Con is just show you everything that I've been looking around that's happening there strength to do a proof of human approach and these are all prototypes or very interesting demos and and experiments that are trying to take different angles of how we verify humans over decentralized networks so a very very nice experiment since humanity now humanity now is a token curated registry you've staked tokens and you nominate candidates and this get approve or disapprove or you can challenge existing candidates by using a simple token economy humanity now is based on Twitter so you evaluate the candidates by looking at their Twitter profiles and if they're due to proper looks like it's a unique human then you can vote in favor of that profile it has some box I was able to generate five identities with it simply changing the username of the of my Twitter account but first it is a nice first step where we can start playing well how can t see ours up with this runner and I encourage you to check out humanity play with it yourself it's already like on the main an evolution of this approach and this is still under the wraps I think it's you know it's still being worked by the team of klaris we are democracy art we are collaborating with them is to do a web of trust with using TCR with video groups the interesting thing about clips is that it introduces an element which i think is very important for evaluating these groups which is randomness if you know if we look at page rank in page rank you will an define validators that are gained more reputation and other validators and as you gain more reputation that means that maybe in the future you can bribe those highly reputable validators and those highly reputable validators can corrupt the system so we need to do a system where Random randomness is an important element and the jurors or those who are evaluating the proofs you know cannot be predicted because they are the consequence of sortition which is one of the properties of the original Greek democracy sortition and claryce is definitely the one technology in the space that is really playing with with randomness and electing jurors to evaluate evidence in very interesting ways and they are you know working in the direction of the helping figure out how to do proof of human another approach which i think is one of the most approaches in this space is doing synchronous tuning tests this is a project from Russia called Idina and what idea has identified is tuning tests our machine learning resistant or our very hard Peters to do but very easy for humans to and here is a very simple example they call these flips so you have two flips the same four images on each strip and in one of the images the correct is the right order and the order is the is correct and the other image the order is not correct so you have to decide which one is right usually you can do this right now yourselves usually humans get this right 95% of the times over 90% of the times computers get this right 70% of the times why is that because we humans we have the cultural background to understand what might be the narrative or the story of the cat and the flowers and what might be going on their computers might be able to recognize some of the patterns inside the pictures but computers don't have the cultura background to understand how these pictures go room together so this is a very interesting exercise and this has let's do the guise of idea to be able to actually a research and here's another one if you want to do it to actually research machine learning resistance tuning tests and they have found out that these types of tests they need to have certain properties to be machine learning persistent one is these programs need to belong to them what is considered AI card programs AI heart problems are problems that do not consist of pattern recognition anything that is based on pattern recognition can eventually be explained it with anomaly these are problems that will abstract thought or some kind of insight that they know it's color for a machine to grasp also this means that these problems must not be generated by an algorithm because anything that is generated by an algorithm can be reverse-engineer so this need to be products that are created by humans and not by algorithms so what I dinna does what I needed us is they put this during tests in the network and participants of the network have to solve this during test at the same time so this is how they prevent civil attacks they have a party where let's say actually the next one is in two days you can go to the EDA website thanks yeah I guess for the API outside the so you get people solving a problem that is easy for humans - machines at the same time because this during tests partly solving problem a solar project party is happening in simultaneously in a synchronous event you might have a genius that is able to solve three or four or five of these superfast but that's why it's not exactly a proof of human protocol more of a proof of personhood protocol you might get some signals but because it's the resolution of these problems happens at the same time it prevents the civil attack problem assuming that these during tests are still hard for machines to figure I think that we need to look deeper into this time creativity or how we think about this but this is a very interesting approach of some of these ideas is your research in your social graph there is a very good paper by a while we'll define identity as a social intersection persist on evaluating or your social graph assuming that identity is a consequence of the different social intersections or the different groups we are a part of and it's interesting because for example usually cities a lot of people will belong to a lot of groups so you have many many narrow intersections and then each one of these intersections will describe a part of your identity whereas in rural areas usually a lot of people below same group so you have populated few populated intersections whereas in cities you have many intersections with few people in them and that's the difference between living in a city and living in the countryside intersectional identity is a pretty formal way of identity and the claim of this paper is that we can formalize some aspects of this identity model and start basing identity or creating even a new political subject that is not making a link to the individual but it's also linked to its communities or its multiple communities which is an interesting approach I will recommend checking on this paper there are some implementation of some of these ideas this protocol called brain ID right ID you you have score which begins with 0 and you start making hot chicks with trusted contacts and if those contacts belong to a group and you can you 50% of the group you have already established a connection with and you become a member of that group that will beat your score in a civil right score proprieties is an implementation of graph analysis and blue graph analysis establishes a score that tries to estimate your your the lightness of you being a unique human so they have a working version on you can play with it there's you know something that we ask ourselves is you know can we actually piggyback piggyback government infrastructure what about the piece of information you provided by governments which have already done a lot of the proof of work to this in mind know that you're a unique human in a way that we don't reveal information about the government so maybe our technique is your show is that you want a government-issued ID and you take the typical picture unit a will see of your face and showing your government issued ID but we have technology that is able to automatically blur from that government ID any piece of personal information you know the same technology that can automatically blur bits of information in your government ID is the same technology that can actually paste a new information into the internal government ID what we want to verify is you know not necessarily who you are or what specific information that is personal about me that I be but we will do verifies that you hold a government issued by the ID we don't care about your name username your surname or your address or anything like that we only care that you are a legitimate older a government-issued ID but you know this technology can also be used for other other purposes the other is sticking with hard work not only thinking about this with software using tracking devices that never ask for any kind of personal information technology like this thing that I have here that is constantly attacking my movement around the planet or my movement are on the streets estimator actually moving around this world so make sure that either I'm a living being might be a way of trying to to use a bit of information than to mind some kind of proof of human assuming that you know this this hard work is antara which is a big assumption these are some of the ideas of what's happening with with proof of human so look at all this looking at everything that's happening and all of these implementations have trade-offs some are easier than others some are cheaper than others some are more is specific with be workable let's not just focus on one specific approach let's try to find out how no how we can work with all of these different approaches and kinda come up with a common denominator or a common score that comes all of us wait is different ways of abandoning identities so an idea that has been injected to me by Albert Wagner actually is going to the rearmost pin you know that entity in a probabilistic way not fidelity either you know you are a human or not human but begin an identity or human is a spectrum and the chances of the landing code that you're a human based on different certificates or different schools so the problem of myung basically when and it is very hard is because it's this thing that lives in these two universes no one is the subjective universe that consists mostly of our attention and the other is the objective universe and is it is something that lives in between this so considering the proof of human things I showed you before we can see that in the object inside of the spectrum basically we're doing planes as strings of bits any string of bits we have pictures fingerprint video you know these districts of pills come from nowhere entropy entropy yes they are and we have different verification mechanisms that rely on computing some running more on computing and the balance rule a more on a conical attention like a be Einstein stands intersections machine learning during tests and TC arts so can we waste all this in a score that can help us make sense out of this something that no even any medium address some kind of heuristic that will tell me okay this is this address is likely to be 92 percent human and then people can write small contracts using them input in order to evaluate a creditor or ever with any kind of a human based service but the bigamy classes were doing an algorithm like this is who watches the watchman who is determining how this algorithm works so this algorithm must not be simply had a blank machine must be something that can take into consideration you know how we want to be observed how we want to decide how we want to be observed how we know already is the source of something that comes from a process that is legitimate so to understand legitimacy underneath that you know it's interesting and we have been researching under that democracy is quality building how many of you are familiar with quality building okay I would encourage you to the Riverina high quality coding basically you can both agree on Miami PD boats but the more boats you put on the same issue the cost of that increases nonlinearly by quadratically so you put two votes on something it would cost you four tokens we put people's it will cost you nine talk ins and this means that with vanity building you are not only measuring the Preferences of the boaters you're also mentioned the intensity of those preferences and a consequence of quality boating is that it tends to generate the normal distributions of preferences in tests we have done using Likert scales the Likert scale follows this typical service where you have strongly disagree strongly agree and balance in between when it without quality holding people tend to go to the extremes to strongly disagree strongly agree but when it's week was a debugging we have this Gaussian burst dismissed normal distributions and people tend to go closer to the center so there's less polarization and the data is a much better organic distribution which helps to miss birnley actually perfect balance for machine learning algorithms we have done an implementation in quality coding the first implementation the u.s. for the state of the United States and we can exactly that we got an organic distribution of 41 state legislators to resign over a hundred and seven bills and we have a distribution of the how to rank these bills throughout time so when we compare this with participatory budgeting with participatory budgeting this illegal distribution of tokens or post with participatory budgeting legislators found the 50 60 % of the bills have the same amount of tokens so they were not able to prioritize among the long tail of bills with one big body we were able to prioritize a very long tail of bills so we're working on bringing some of these ideas together we're building a democracy down below so I'm gonna put my shares of all of the subject to this democracy doll experiments and we you know there's more to this like zero knowledge proof and year 675 and LED standard come look for me on Twitter and ask Santa City on Twitter and there's a group of humanity meet up if you want to come is tomorrow at 6 p.m. take a picture of that and we'll talk more about this thank you very much [Applause]