AI x Web3: Pioneering Decentralized Intelligence - TOKEN2049 Dubai 2024

you know I think everyone is here to to hear a little bit about decentralized AI which is sort of a a topic of you know a lot of uh a lot of memes recently uh Ilia being the the center of one of the the most powerful memes um but everyone on on the stage has a sort of different lens a different take a different sort of view on decentralized AI overall and it'd be great if you you could each introduce in a real tweet size bite what your project does and what you know kind of your background sure uh hi everyone I'm Sean co-founder and CEO of Sahara AI um reallyy My ultimate goal is to democratize scouting law for AI which I mean covers a lot of things data model and compute but Sahara AI is basically building a AI utility blockchain platform uh for everyone to really have the true ownership Providence and attribution of AI access including data and model and really withis high performance user privacy and security I'm IIA co-founder of near my background is in AI I was uh doing AI research for about 10 years and I'm one of the co-authors of the attention all you need which introduce Transformers and with near our vision always been to give control of assets data and Power of Choice to everyone in the world and AI is a a major part of that it is how we going to be interacting as Computing going forward and uh providing user owned Ai and the platform for that is what we do it near nice hey everyone my name is Ahmed I'm this CEO and founder of IET IET is a compute AI compute Network that has compute power 20 times more than open AI so far we're growing really fast um uh we are building like a an AI compute Network that could facilitate the Next Generation like we of AI models on chain we believe uh we believe in democratizing access to Ai and we believe that um the world is does not have Fair access to AI compute power and it's limited with few and uh we managed to like group in all the gpus from across the globe into a single Network that acts like a single processing chip so powerful that could solve like real Humanity problems and challenges so yeah looking forward to be here I'm happy awesome um yeah I think this is probably the only panel you'll see today where TPS doesn't mean transactions per second so hopefully uh hopefully people can disambiguate in their answers um so you know the first question that I think a lot of people have when they hear you know think about decentralized AI is well how do they act how does how does how does decentralization actually mix with AI in a lot of ways AI is sort of the ultimate centralizing force uh it's very clear that centralized AI is very effective at because it can you can aggregate data you can kind of have these economies of scale and in model training um on the other hand open source AI in the last probably 3 to six months has had sort of this insane growth rate um and and you know anecdotally you know a lot of my friends who work at either open a eye or deep mind or places like that a lot of them effectively have said you know I don't think there's a real growth market for the closed Source world because it's too expensive to compete and the open source models are actually getting good enough to catch up so you know a natural question for all three of you is what is the role of decentralization within the open source AI world and do you think that decentralization is sort of a necessary condition for open source AI to be successful or not so maybe start with you Ilia and then we'll rotate around yeah I mean obviously it's a loaded question so I'll I'll try to unpack it first of all the kind of from values perspective right uh I personally am extremely scared of a few companies that are controlling what will become you know the most sophisticated tools uh in the world that have you know power of intelligence of you know designing new drugs defining new strategies and being able to pretty much shape our society and the problem is this company have a fundamental uh kind of structure where they need to continue growing their revenue and if you're a company like Google for example which already achieved its target audience the only way you can grow revenue is by extracting more value from your users you need to grow average revenue per user and this is not because the company is evil or any individual person is evil but when you're making decisions day-to-day as engineers and product managers you're running AB tests and you're maximizing what model will extract more revenue and generate more revenue for the company and so we are in this Fork on the road at the at the core where we have for profit companies that will be trying to maximize pretty much extractive value from the from its users or we need an alternative which is actually the this models this AI tools are working for you for you individually and they trying to maximize your kind of Happiness your well-being your economic State and so so that's kind of at the core for me where where this difference lie now it's true that the as a centralized company you have such a massive flywheel you have data data leads to better models better models mean you can generate more Revenue more Revenue means you can build more compute and so actually VB to open- Source AI startups don't have much of a legs and we see this now in the market we have startup after startup now that actually folds and being acquired by big companies because they're not able to generate enough Revenue to justify their valuation at which they need to raise to build those models and so we need an alternative we need an alternative that actually is truly uh like has economic incentives and it's aligned with its users and crypto is that incentive structure it's that tool that enables that the open marketplaces is a way to aggregate resources and assets the way to con collaborate and contribute data together is way to reward creators which right now you know alternative is a lot of lawsuits with open Ai and so we have a tools to build a better model it uses cryptographic incentives it uses distributed systems and it in result achieves this vision of self- sovereignty so for me decentralization is just a tool to achieve the self sovereignty and so that's kind of how I see this all the pieces working together Sean yeah um great question I think I mean working on AI for over 15 years so I think the really interesting um transitional moment for the AI Community starting apparently from chat gbd but for me it's really like gbd3 I've been working on that for over a year and I think the inflection point we having here is the AI is no longer a toy right it's not just like a task bath for research Community to work on but really start impacting different parts of human life and start impacting people's job opportunity and monetization opportunity and really the pressing problem for us like also IIA mentioned that is um how we actually um Embrace every single participant in the AI economy and give them the right sort of uh share of the whole AI econ economic Grows Right for example um a great example here is like think about open AI I've been using all these open resources uh like Wikipedia GitHub and rits to do like do paining and some of them use for post training of the model and create chat gbt but all the money's been making by chat gbt hasn't got really back traded to all of the contributors of these resources and how can we actually solve that problem that's a burning problem for Sahara AI right now and really going back to the question about where we see the uh organic Synergy between decentralization and AI I think there's big part of that is about proprietary data um I personally think that open source is not the best way to really grow the AI economy even though if I put on my researcher sort of a head I would really Embrace open source AI because it just really help accelerate the model developments and just make the information more transparent but think about if you are uh a business that have very propor information about financial domain about medical domain you wouldn't really want to just like open source everything but you want to think about how could you become part of the development flow to turn that data into into valuable model and access and that access be further used by other people to build Webers of applications and monetize them but you still become a shareholder of the entire tra and you can earn that by just contributing to part of that process but you don't have to do the whole endtoend procedure right so I think that's like really a sort of a decentralized way to think about how to leverage data sets to build model and play a sort of a a healthy ecosystem of AI economy that's already big problems for whole society well I think decentralization is a decision like we we have the chance to shape what is decentralized looks like going to look like in the next 10 years or five years um it's it's a decision that like like uber say like we we don't want more taxes like we want you know sharing economy want people to come in bring in their cars and be part of the taxi industry in a way so same same thing here is like we have to take the decision as of think of it this way when we started IET we thinking like there's so much money being made in the AI computer industry all the GPU owners and so on they're making so much money like AWS making massive amount of Revenue because of GPU bookings and clusters and training machine learning models and so on so they GPU earnings is massive and then a data center or facility owner of like 5,000 GPU is like how can I earn like AWS like I can't I it's too big it's too huge I need a sales team I need marketing I need so much for me with small data center competing with AWS with few hundred thousand gpus so how there's no chance for me to compete alone uh so everyone would wish to be part of AWS like imagine a crowdsourced AWS so then they yeah what if multiple data centers all of us gather together and make this Alliance like let's all of us make a new AWS together because we don't want one person to control it you own a share that own a share each one of us on a share of of what of this massive money that's flowing in from uh the AI computer industry it's massive it's a trillion dollar industry and when we say decentralized AI I think it's multiple layers that we we can reshape everything like we can shape what is the AI compute layer going to look like the the main one and then above that and that AI compute layer basically the gpus and all the compute compute jobs that have to be executed we could uh like how can we facilitate payments like using near using aptus using salana using like building all this infrastructure piece by piece together we're just literally at the beginning Inception of the AI in in the crypto industry so whoever Starts Now really have a big Advantage because he has the chance to actually shape how the industry going to look like now you go above that you have Sahar and others like they have the models they they give royalty owners to royalty fees to owners of the models model stores and so on that is another layer on top of compute that layer on top of it you have app Builders who going to build AI products inferencing whatever type of products so there it's multiple layers that all of us have the chance to shape how it looks like in a way where like let's share data Gathering let's share building models together uh if someone built a model dropped it in the model store I want when someone to pick my model and play with it or put it in his own app and he becomes successful I'm as an owner of a model I also want to earn something so there so you know model owner royalty fee data Gathering there's so much so much we can think of products that actually can really make an advantage I mean and the benefit of that is like who in the world can by himself go and buy 500 700,000 gpus no one except Ma and the big guys like only Facebook has 700,000 gpus on the whole planet there's no one has more GPS than these guys so it but because of the power of decentralization because we all of us come together ion now right now has 980,000 gpus who who can even think of this if I'm alone I can't do it if each one of us is alone no but when when it opens when we open the economy come in let's bring data we will bring data for model to train more than what openi has and more than anyone has because you have the power of hundreds of thousands of people in the community to work with you and that's really the advantage like I think decentralization is a decision and all of us working together is like we all of us in one team here like it's not somewh we're not competing here the industry is outside crypto it's web to that's where actually the competition has to be we're competing with the big players so yeah there's so much happening in this world I'm so excited to be a part of it cool so I I think uh you know from these three views one of the clear things that um you know all all of you have sort of mentioned is a there needs to be some form of a Marketplace for different components within the stack whether it's um a Marketplace for the inputs in the data whether it's a Marketplace for um some form of provenance or whether it's a Marketplace for agents to use money in right but you know even though know AI has been a very hot theme in crypto and realistically the overall economy uh there's really nothing that has found really great product Market fit in terms of things at the intersection of the two um other than what you're working on what are the things that you're most excited about that people are working on in the kind of intersection between between cryp you sir yeah um just want to quickly follow up on Ton's question about marketplace right I think Marketplace is nothing new I mean it's been in web to for decades but really I I think the fundamental problems all of us trying to tackle here is how to have a layer underneath the marketplace web her that really have the sort of ownership Providence and attribution functionalities and this is where really blockchain and all these privacy security technologies will be able to power right if if we put the marketplace on top of a centralized Cloud let's say AWS has great Marketplace they have Amazon Mechanical Turk for data they have all of the other API Marketplace but you know that once you start placing things on top of the marketplace it's pretty much you have to comply to all of the policy and regulations by Amazon which is a centralized provider and you wouldn't be able to have the true ownership of all of the future monetization of things right so so going back to the problem here uh one I'm seeing really exciting in the intersection of AI and cryptos space is not just merely to think about how crypto provide a really high liquidity or sort of a way to expose AI capability or utility to broader Community by designing economics layers or things but really think about how to protect the user on their valuable access right uh apparently there's an interesting premise I'm taking here is like not every single one has the equal valuable sort of uh information or knowledge or data data uh at their current position but they would have time they can use the time in exchange to U contributing to some public data resources like a sort of a rather an open source version of Wikipedia maybe some sort of a propriatary version of domain knowledge bases which can then further be used to create um proprietory domain specific models by a community rather by a single uh organizations or company and then they be able to monetize through those models and um and data sets down the road and really providing infrastructures with blockchains and security uh um privacy preserving technique I think this is the really the right way and the big direction to go for the whole Community yeah yeah I think I mean just to mention on product Market fit I think the data labeling platform built on blockchain and for example near has uh near crowd is probably at product marketting fit because it has way lower cost it's like benchmarks at like one10 if not 12th of the cost of centralized platform and uh it's able to pretty much get Workers across the globe really easily and uh pretty much scale its Workforce as well as scale the expertise required as as data requirements shift uh as models mature now I'm really interested in looking forward into two aspects one is is uh how do we go beyond current Creator economy where we reward for eyeballs where in the future these models will be uh really serving as your medium where they consume the news they consume the you know new Tik Tok videos and they provide you a personalized feed of you know everything that's happening in the world uh and uh you know whatever your favorite entertainment is and so the reality is the current platform current advertisement current copyright are pretty much not applicable into generative AI world and I think there's a really interesting kind of um opportunity here to not just you know kind of traditional let's do nfts for Creator economy but actually reimagine how Creator is going to be rewarded in a generative AI world and there's like few startups working on it but I think that's like an amazing opportunity that uh blockchain really sits in the right place to serve and then looking forward uh I expect more and more of the you know onchain governance as well as you know some of the other offchain decisions to be made by AI where the population the token holder shareholders or the kind of population of the unit of governance are really serve as a veto and input uh but allow model to pretty much summarize and make decisions based on the different criteria and so you can kind of expect a lot of interesting both research to come out but like in blockchain right now it's a really good place to test this out because you can set up a DA as a million dollars and let it lose on like you know set of criterias that even potentially you cannot do as a human like for example as a human you cannot maximize the uh token price of the Dow uh but as a AI model that governs this Dow you are actually able to do that right so we have a you know decentralized for example inference uh reporting on how to make action for a dow and so you can have really interesting experiments right now uh to facilitate that again there's like few companies exploring this but I think this will be future of governance because people suck at governance and uh you know we've seen with Dallas that like coordinating a number of stakeholders is extremely complex and it requires pretty much a superhuman ability of summarizing information providing context and so I think that's a really interesting use case that I'm really excited about I I think um basically if you look at it well there's there's a big problem I see in the market which is we build an AI product for the AI not we I mean like as a crypto not IET as a crypto person the crypto Community builds AI mostly builds AI products for AI commity in web 3 and they limit themselves with this very small Market size that's is just yet started and the use cases are extremely complex for like next Generation use cases that is still too early for it and what happens is they would build for this Niche AI web 3 community and they will forget the 99.95% of the community that's actually web 2 and far away from uh web 3 so what I believe the move is to move in to it's two B it's two fronts battle you you make a product that's doesn't feel like it's web 3 you target Web Two like the people in the web two you let them use your product you let them flow the money in your product while that is happening you're building something more complex Next Generation what while adoption is slowly growing in the AI community in the web 3 so building an AI product in the web 3 you should definitely not Target anyone in the web 3 you should definitely build a product that actually not for another chain or for another uh something AI in the crypto for some use case that's extremely limited and has no yet enough users but rather build an app that is served for you know millions of users outside and then slowly if that market in in the web 3 grows you know you you start targeting that web 3 but mainly you should be targeting web to and the user experience should be extremely simple like any AI app you don't want to add more complexity AI have already reshaped a lot and it's moving so fast that you just can't keep up with the new technology involved you can't keep up learning new papers every day coming up that's you just lose control there too much happening already there so when you add even one more layer of complexity to the engineers or to the end user you're just like making it harder and harder while we we want to to make it easier for people so we can bring in the adoption not harder well that uh that that that kind of answers another question I was going to bring up which I'm kind of curious about everyone else's views on which is you know there's there's always this talk about Crypton native consumption of of decentralized AI so you know like the examples you gave with governance or the examples of people doing onchain inference to provide uh kind of some calculations to contracts you know on the other hand there's hey maybe maybe it's actually that the payments Associated to the usage of the model is more important and then you have this kind of less Crypton native web 3 type of usage how do you think the market will evolve do you think the first uh successful application uh will be something that's Crypt native and and really focused on on existing crypto users or do you think it will be something that escapes the the Box yeah I can start first um I think um I I can first give a very examp a good example of what happens with in Sahara AI or particularly speaking Sahara data which we launched like last October so we have initially only um been outreaching to web to AI native uh data labelers which we have a network about 200,000 labelers m jely in APAC and africar area and these people initially when we do the survey they only have a crypto wallet adoption rate lower than 15% and gradually we start educating them about the the good things or advantage of using crypto water instead of going for PayPal or Bank WS these kind of options apparently you have lower uh transaction fee and then you have instant almost instant type of a speed for the transaction and this is really smooth pickup process for those 200,000 laborers to embrace crypto wallet and I think this is a great example about um what things are relatively easy or like low hand infs to really educate uh web tool AI native people to use right and I think crypto payment Gateway is is is a really strong um sort of uh um example for me to talk to the AI people and convince them to leverage it and then going back to the broader question about uh where I see this like uh uh crypto native apps to come out first versus like cryp uh crypto for AI I personally believe we will first see a bunch a bunch of really strong utility uh crypto for AI applications really soon and I think part of the reason is that uh if you're looking at the current um um Capital Market for AI projects it's pretty tough basically uh and I think the the shape is start pretty much settling down where you see every single sector let's say video Generations text language models image generation they start having the top number one player Which is far um surpassing the rest of the players and they have the user the traction the accumulated data the flying rle effect for the apps they have the capital the rest 80% of the AI projects they've been struggling to find Capital to further growth or find a way out most of them start just switching to very customized very specialized use cases is to really like survive in a in a sort of Enterprise facing way and I really think the whole crypto Market provide them a strong utility and liquidity sort of a space for them to come to play really I think the Gap right now is this communication Channel between the AI native community and crypto native Community are pretty much not there yet um for many reasons historically and I think if we can really think about how to close the gap between these two communities there's a lot of great things going to happen happen in terms of using blockchain to provide security Providence attribution monetization for AI models and provide them a sort of a more ways to access right for their great projects of different valuable AI models and data sets yeah yeah I'll second that I think um just to mention I think right now everybody because of how exciting and hot this area is of a web 3 there's probably every single idea being funded for somebody to build right so I think like either way we'll figure out what works in next like six to 12 months capitalism has a way I guess yeah cap yeah pretty much like Evolution capitalism making it making its uh uh decision right now so we'll we'll we'll pretty quickly see uh what are things working but I definitely agree that like everything from payments to Providence to kind of coordination of resources of of infrastructure like all of those things are like pretty clear and uh provide you know value not just for kind of web 3 but actual like web 2 uh consumers I think the other side is like web 2 AI startups indeed are you know going to start struggling because they raised that massive valuations and they need to justify those and they don't have as much users and so this is where you know some of the advancement in crypto technology like points are providing a pretty pretty extra interesting way for people to uh pretty much bootstrap their growth of their web 2 solution like I've talked with some of the kind of AI gaming like not web 3 like proper AI gaming folks and they struggling with growth right they're struggling with figuring out how to put this in front of massive audiences and so for us actually we we have massive audience on near we have distribution Channels with like tens of millions of users now and we can actually offer that as a platform from some web 2 companies now to offer this and then you know turn it into you know an asset and make it more entertaining and and exciting for people to continue so I think like there is an interesting also Direction Where web 3 also because of the audiences we did build which have money which have knowledge and understand the values which are interested in ownership which are interested in kind of like in a way those of early adopters as well right even though it's like now in in tens of millions it's still early adopters who are excited about new technology so we have an interesting opportunity as well to now offer that as a platform for this incoming new startups which look for for this audiences all right since we got two minutes unless you want to I think um I think honestly the the first problem when I started IET was like there have some been some people who tried to build compute networks before um but it never picked up and it was very momental and then it didn't really pick up in the normal industry outside crypto so I think iio is in my opinion the first AI compute Network that actually kind of unlocked that and accessed the web to and the AI industry right now we have more than 400,000 users at least 100,000 of them are AI Engineers who are totally have no clue about crypto and we have more than4 billion dollar of GPU infrastructure in the network that's and that's like really flowing in the AI industry like fueling the AI industry outside the the user experience is native you don't even feel you are in crypto massive amount of startups running hundreds of gpus um on our what we call it the internet of gpus so I think in my opinion I've uh I somehow like me and the team have I feel we have really done the First Co like the first AI coin that goes to the real AI industry and matches and connects Both Worlds together uh from web 3 to AI industry and I'm really proud of it and then yeah that's it thank you very much so for last uh 30 seconds you know I was I had dinner last week with one of my friends who's uh you know worked at Fair Deep Mind and open AI you know has probably 100,00 citations very kind of prominent AI researcher and I told him I was I was going to host a decentralized AI panel and he was like how is that not a scam how is how is like any of this actually useful so what what is your you know one one rort to someone like that who's like a prominent AI researcher yeah I mean I think AI is a collaborative effort right it's a collaborative intelligence in terms of data in terms of computing in terms of how to put together thoughts about a model configuration and S so we should just properly give people the share of the AI give people the ownership of AI make the right attribution and monetization works and I think this is just good thing for both communities yeah yeah I would second that it's user owned AI requires a different structure different incentives op like open source web to company AI companies don't work they're going to fail and like if the only alternative is Google and Microsoft that's also not a good reality which we should live in so like if they want an alternative they want a way to fund open source they want to publish papers and not you know sit on their research uh for years they should join open kind of and user owned AI initiative so send the resume well I just say watch and see wait and see that's I guess it's always Us and Them and we don't have to convince him he's like we're doing it our job in the ecosystem is to defend the ecosystem from uh anyone who's just trying to ride the hype but not really doing any AI product making sure the the AI web 3 image is very clean and pure uh you know pushing and funding and helping all these real projects that are building together so we can like show the best of what we have because we still just started like really the AI crypto industri is just this is the first time we see AI in the crypto industry so so it will take us few Cycles but then eventually AI I believe would be at least 30% of the market cap so it's it's worth that's that's a big that's a big claim yeah all right well I think we're out of time so uh thanks everyone for for being on the panel thank you thank you