iO by Barry Whitehat | Devcon SEA

[Music] okay thank you everybody uh is this good can you hear me okay so today I'm going to talk about iio so this talk starts with like an inciting incident so in 2013 or 2023 I took a trip to Marseilles and I got really bad food poisoning I was I was very sick and I had these kind of like Hall inations that ordinarily you need to pay like a lot of money to get so it was either this or it was this but I don't know which which one was the culprit um but like so people think that like your your stomach is actually like a part of your brain that like the the bacteria in your stomach uh it digests food in a certain way and it communicates with your brain which like affects the way that you actually think and after I had the food poisoning it really kind of changed my life a lot uh here are some of the changes that happened so I started living alone I live I moved to a new city to live with friends I I stopped being a digital Nomad and I had a a home for the first time in like five or six years uh I started to focus on health I found like peace in my life that I didn't know was possible to have I also quit my job um and all of this was like precipitated by me having food poisoning but I think that it was more like another contributing factor is that like I just always want more from life like the things that the things that I have or the things that we have is is like never satisfying for me you know so I I'm always like searching for more um so like so now I'm going to talk about like some of the things that we currently have that I think are not enough uh one of them is the like the whole programmable cryptography tool of like proof of validity like this is a useful tool but like we've been using it for years and it seems that like we've we've we're getting close to the limits of applications like I'm very excited that we have ZK EVMS and we have ZK rollups and that we have this like whole proof of validity uh uh ecosystem that's that's growing but I don't think that this has transformed the way the world in the way that I was hoping that programmable cryptography would change the world another thing that I'm concerned about is like zkid so I spent a lot of time thinking about like well why was the first blockchain thing a currency rather than a reputation system uh because I thought that like well a reputation system is going to be much more like free and open but the reason I think is that like reputation systems are inherently about like excluding people and it seems that like this is a difficult thing to like build as like uh it it seems as less empowering than a currency system which is like like free for anyone to use basically um um so like I'm less excited about doing like private identity stuff and because it's inherently about like excluding people uh so like last year I spent a lot of time thinking about like what else could we have you know like what was the future what was a better future for us so this time last year I gave a talk about it's called 2pc is for Lovers and it talks about how we can use FHA and NPC to have this like multiple party hallucinated server where you have this like encrypted computer that like you do all of the operations inside the computer and nobody knows what's going on inside and like the talk is interesting and it digs into this idea and proposes some sample applications uh but one of the big problems with this whole direction of work is that it required interaction so like with MPC you needed to be online to in order to execute the server and I was I think like I'm currently most excited about like having servers with many many people uh Computing together and if you wanted to have a server with like a million people you just can't do that right because a million people would have to be online at the same time and Computing together at the same time which seems to me just to be completely impractical because like someone is going to lose their password or someone's going to lose their phone and you just won't be able to see the final result or you won't be able to finish the computation but the thing that uh that so so in order to overcome this the tool that we need to use is IO which which is stands for indistinguishable observation which basically takes a away the interactivity uh that we that we need to have for hallucinated servers up until now um so if you want to do something new you have to do things differently so the rest of my talk I'm going to talk about like the the different things that we're doing uh so why don't we have it already so it's hard uh like IND distinguish alisation has been an open problem for a long time recently in 2022 the first practic or the first scheme from standard assumptions was proposed and this was a big deal in the IO world because it was the first time that it was like even theoretically possible to do this but unfortunately the scheme is Not Practical it would take millions of years to like execute to to do a full uh execution of IO an IO circuit um but this is from standard assumptions so in cryptography we're we we always try to base what we're doing upon standard assumptions uh but like actually in the cryptocurrency world we won't really do this so much uh in the cryptocurrency world we we use a bunch of non-standard assumptions that like the regular cryptography Community would be very concerned about and like two examples are like the snack friendly hash functions they use this very like crazy kind of assumption that like we like the rest of cryptography doesn't really think about and also grinding is right is another non-standard assumption so like well the question is like like another question to ask is like why do we even need to assume anything in the first place and the reason for this is is that like when like P equals NP question mark is like this big open problem in computer science it's like this kind of like million Millennium prize if you're able to prove like P is not equal to NP is basically says that like some things are easy and some things are hard and like all of the easy things all of the hard things are not actually just secret secretly easy things uh so because we can't prove that something is hard or something is easy easy we're not able to prove that like no sorry because we're not able to prove that anything is harded and or anything is easy we're not able to prove that like specific things are harded or easy and because of that we're not able to to say that like well this problem is hard it's hard to solve this it's hard to decrypt this unless you have the private key so what are current assumptions we currently use well we currently use a bunch of assumptions that are based upon prime numbers so elliptic elliptic curves is like a prime number based assumption and this assumption has been chosen because it's been like an open problem for hundreds of years for hundreds of years number theory has been trying to figure out to understand more about about prime numbers and about like their distribution or whatever and this is the assumption that like most of our cryptography is based upon because it's just really high like it's been open for a long time but like the problem is that there's a lot of there's a lot of complexity here there's a lot of complexity I want to go back yeah there's a lot of complexity here in Prime not because it's just like this kind of security by obscurity thing it's this problem we've been trying to solve for years and years and years and like no one has made any progress on it so therefore we think it's hard but like I'm not like there's been like recently more people have been trying to come up with like simpler assumptions that are Al that are also like we think hard and like lisis is an example of this uh okay now I want to go to the next slide okay so like the lus assumption is that much simpler than the prime prime number based assumptions and they seem like much more it it it intuitively it seems to me like much harder but it's not like but we can't prove it like we can't prove that anything is hard so we just have to base everything on these assumptions um it seemed yeah so like one exam so what does it take for like an assumption to become like a standard assumption so it took about 10 years for the lce based assumptions to become like standard although like many people in the cryptography Community are still very skeptical about luses for example Daniel Bernstein is just like does not is not confident in in lce based uh cryptography um yeah so like this slide I don't know if I have time to go deeply into it but it's basically claiming that like AI is more dangerous than quantum computers because for cryptography because like if AI seems to be a tool to like teach people and that seems to me like inherently more powerful than a tool that's like able to do compute compute so like I feel that like this like I make a prediction here that like well in the next 20 years AI is going to make break more stuff than quantum computers will um I think this is a good this is a good bet um but like it means that like the the assumptions that we that we previously had uh we need to be more careful about right because like we're we're going into a world where where the like old this has been hired for a very long time might not CH might not be remain being the case as Okay so right so so like basically I've become open to to using like non-standard assumptions to build IO and the hope is that like we will be able to here's a list of papers uh and and and basically the way it works is the top uh from top to bottom is from like standard assumptions to like more exotic assumptions um yeah and here's our implementations of the various schemes but this is in no particular order I should have ordered this better uh yeah so like basically our goal is to like Implement these schemes and to like break them or or try to find and then to build tools to build things on top of the schemes so like we have a we have a bunch so so we're announcing this bounty where we've implemented the first paper or the last paper on the list the most like a uh extreme assumption on the list before it's called local mixing and basically you can take take a circuit with 1,000 Gates and OBS fiscated it which basically means that you add like Randomness to the Circuit until the circuit becomes 230,000 Gates and then basically the challenge is to take the 230,000 gate circuit and reverse it to the 1,4 gates circuit right so be able to remove all of the noise while maintaining the functionality of the circuit and we claim that this is really hard and if this is really hard it's it's basically like indistinguishable ofation we can build a bunch of cool stuff so this is a a boundary that we're just announcing today and so the the price is $10,000 for someone who's able to do this and this is a a collaboration between these three groups phandom Zone ZX and the eum foundation uh yeah so well what would it mean if we were able to do IO um we would basically finish cryptography and the reason like okay so here's how to think about this so basically you can model IO has like a trusted third party that you can actually trust so Z you can Al you can model zero knowledge proofs like this but the party is more limited so if you have a trusted typ party that like does some compute and verifies the compute for you then that's like that's like what a zero knowledge proof is but with with I with with like once you get to IO you have a that's able to do basically everything they're able to verify computation they're able to they're able to like do multiparty computation stuff they're able to maintain secrets and only use the secret under certain conditions um so IO is basically very close to the end of what we of what we would need uh or is close to the end of programmable cryptography uh and like I don't know what I'm like I I feel like this is like the last big Quest it's kind of like you know that in history there was a time when people thought that there was there was just more land that people didn't know like people were like exploring the world and there was just more places to go you could always like go further west but then at one point you just got like to the end of the world and then it was like what do you do like what do those people do like imagine that you're like you're like yeah I want to live a life of adventure and then you get to the west coast of the US and you're just like done like what next I don't know but I feel like no matter what I will not be happy because no wants to finish programmable quotography less than I do okay so uh while you're in Thailand there's all these opportunities to have these like life-changing experiences so go to the night market and and don't be too careful uh so there's two more there's a couple more talks about this um there's 2 p.m. on stage 6 today at J is going to talk more about the Bounty he's going to dig into like how local mixing works and a bunch of and and like a bunch of things about like the the scheme itself like the we don't have any tools yet that people can like build applications on top of um at the moment we're still like we just made the simplest possible circuit but it's reasonably practical the the the thing the thing that takes the most time is actually doing the obfuscation to run the circuit it's just like a 2,000 260,000 gate circuit that you execute so like the performance is pretty good but like that said it is a very extreme assumption that we're building on I think that like our plan is to imp sorry let me go back and show the show the other papers so we already did local mixing but our plan is to sort of work up from like most extreme assumption and most practical or least and most practical to to less extreme and least practical and hopefully somewhere along the way we'll find we'll find something that's that like is in The Sweet Spot uh I'm confident that we'll be able to find something that works uh from standard assumptions but I think that like it's important to start and like our goal is to like have a bounty for a bunch of these different schemes and always have one that's not being broken or like always have two that's not been broken and just like continue this effort because like the the lattice based assumptions uh it took like 10 years to go from being like kind of crazy assumption to more useful or like more more accepted so we're hoping that like we'll be able to like navigate a similar trajectory here okay okay yeah so there's more talks I think on uh Tomorrow there's a bunch of talks about IO that would be interesting for people to check out um yeah I have a little bit extra time I can take some questions or I can talk some more if I remember some other stuff I wanted to say thank you so much Barry thank you this presentation was very complex for my non technical brand but actually the audience here have added very cool questions and the first one here that I wanted to ask was was you know like a definition what is an what's IO and uh pretty much an explanation if you could do like an explain like M five that would be helpful sure uh so iio stands for indistinguishable OBS fisc and it basically means like encrypting a program such that you can execute the program and you can get a an output that you can that is not encrypted so like no one is able to see what the program is doing but what you get out is something something that you can understand so basically it's like FHA but in a non- interactive way like another way to think about it is I can give like the formal definition but maybe we should go to the other the the next question let's just take the next question another question here what does IO replace ZK ah so IO kind of includes ZK so there's ways that you can use IO to to well it like it depends what you mean by ZK because people abuse the term to mean a whole bunch of things and like I used to do that too well I don't know if I did used to do that too but yeah okay so if you mean zero knowledge proofs as just like a proof that a computation was done correctly then then it's like yeah you can do you can you can do that with IO but IO is much more than ZK it's it's more powerful because like another way to think about this is ZK is like a single player thing that I can prove about things that I know I can show people things that I know and other people show me things that they know but there's no way to do like a collaborative proof there's no way to do a multiplayer things right and like iio basically lets you do like multiplayer stuff so you can do like all these interesting Co coordination things well you got here some love from your psse colleagues and I have another question for you so Quantum machine learning will break more things than AI or qu to each alone sorry Quantum machine learning will break more things than AI or Quantum by themsel will break more AI or Break It on its own uh I think that like AI by itself will break more than than than quantum computers by itself does IO not rely on any assumptions yeah yeah yeah it does IO relies on a bunch of of different assumptions like all cryptography lies on assumptions except for one time pad which doesn't rely on any assumptions is there any physical heart limit to how fast cryptography ZK Etc can get uh yes it can like ZK cannot get faster than the than the thing you're trying to prove you at least have to like calculate the witness um or you need to be able to like do the computation um ZK has that limit and so does everything I guess I suppose that like you at least need to understand what you're trying to do I need to think more about like the io case to have like a more coherent answer when will IO happen one year five years how would you define a prediction market for this H yeah that sounds good uh an IO prediction Market I think that like we like it did like I think the key thing to ask in that question is from like what kind of assumptions that like we historically we've had a whole bunch of IO schemes that have been from like standard from non-standard assumptions but they've all been broken um so I think it's just a matter of time on like so so like that's the important thing to include if you're thinking about when you're asking this question so from like nonstandard assumptions I think pretty soon like it's already the case that there's practical IO schemes from non-standard assumptions but we don't know if they'll be broken or not so like that's another interesting thing we could do prediction markets of whether a particular assumption will be broken that could be really fun um yeah do we need eteron if IO works yeah yeah I think I think that there's still these things kind of like add together uh so there's a couple of things that IO is not able to do one of them is like data proof of publication or data availability uh cryptography as far as we know is not able to make sure that you have published something because that's like a it's it's like an objective reality to be confident something has been published is not possible to do with just compute you need to have some consensus so ethereum is still useful for this what are you personally excited to work on almost the end of cryptography research is reached I don't know I don't know yeah I don't know I I feel like I'm just secretly going to try and sabotage our efforts yeah there is a lot of questions here but I do wanted to find one maybe like one last question if you can compare IO with food what would be [Music] um I don't know probably the hardest question you've got in this I see it seems to be some kind of like food that consumes all of the food so like atoms yeah okay yeah yeah okay carbon yeah that's a good that's a good closing for your present bar thank you so much it's very cool to learn more about IO everyone give it up for Barry