AI Isn't Evil and Smart Contract Proves It | Suede Kam - OraProtocol
Ethereum Denver·Wed, Oct 9, 2024, 12:00 AM
Speaker
What problems can blockchain solve for open source, decentralized AI? Why and where smart contract and Ethereum can come in? Plus, dissect a range of use cases that people can relate to such as DeFi, NFTs, deep fakes and protocol level design (for the nerds out there ;)). APPLICATIONS FOR ETHDenver 2024: #YearOfTheSpork ARE OPEN APPLY NOW https://ethdenver.com/apply
Transcript
a i a i a i a i yeah I'm not a bot I'm cting from AA okay can uh let me shck the slide is it working if you want to goad and full screen your present okay yeah should be great so how's your guys today do you guys enjoy the e d 2024 and uh I see a lot of AI project and AI times crypto narrative is going on and uh today I'm going to tell everyone that you don't need to worry about the Skynet because AI is not evil and we can use SM conat to prove that so so although like we have some good Ai and some bad AI how we Define the bad AI first when I go on I want to talk about that CU like the bad AI is the AI that we cannot know what's going on inside the black box it's not transparent nobody know what's going on in the inference nobody know what data they use for the training is it ethical or not so that's why like close Source AI is really easy to be the bad boy AI right and how we resol this problem we open source the AI model then we'll be able to know what is going on inside the AI model what data you use to train the AI model what the logic of the inference when the AI model is executing the inference and most importantly we can control use blockchain to control what kind of you we will know what kind of inflence is going on inside AI model but open source AI model is not enough cuz like yes yesterday if your guys open the website of the hugging phas which supposed to be the biggest open source AI platform in the Earth on the earth and it it's down 4503 error code and nobody can assess their even their own AI model you cannot even open the website so just by open sourcing the AI model is far from enough for us to have like a good AI future so because like hugging face is hold thing on the centralized service provider it would be the choke Pawn for the uh AI decentralization so sometimes we just cannot use it because the server is down so we don't want that happen again so imagine if you are driving a AI driven like a uh Auto auto driving car and all of the sudden the air is offline then what's going on like your car will crash right so you don't want that happen you want the AI Services can provide the inference res 24/7 sta so by at enabling the AI model inference on blockchain the most important part is like you you need to make the AI model verifiable because like blockchain itself it's like a machine of a network of consensus put down the stamp of approval of certain computation res so we this really hard for us to essentially put let every know there's like tons of ethereum know you cannot let every of stand to run the AI inference inside the no so what we need to do is like need to let dedicate the job to the AI inference operator and verify the result on blockchain so the verifiability is the most important feature when we put the AI model into the blockchain so when we have the AI model Des verifiable then we can guarantee the inference happen inside a decentralized environment so that we can have a stable 247 AI services for us even like some centralized server is down so no need to worry about the service is not on when the AI influence is on blockchain the second one the second most important component will be like the tokenization of the AI model you want to incentivize the community to improve the open source AI model because think about this today if you a builder of a AI company then you have no way to profit the AI monetize the AI model beside close Source your AI model nobody can assess the model you just provide a subscription API and people got to pay for it every month and nobody know how this AI function beside yourself right so we will want that open and then but once you open that you people can just copy and download the model and run it by themselves they no longer to pay for the contributor who made the AI model right so that's important for blockchain industry our job is like the reniss on of the open- source software movement then we can begin to think about like why we cannot just like May the AI motor itself become a token then everybody can have the token and once they own the token they own the model and once they own the model they will try to improve the model they will have the motivation and incentive to make the model better that's why we want a community drip one AI model and verifiability is the Spectrum in rollup world we have op optimistic interactive FR proof mechanism to protect the transaction is correct and we have ZK rollup that use the cryptography to ensure the transaction is executing correctly and in AI influence on blockchain we can have this two mechanism again and what's even more amazing is that we canot just have either CK or op we can have them both on the AI inference so um zkm M how many of the audience know about zkl raise your hand if you do yeah it's like quite good amount people know what ckm now and ckl essentially is used like the cryptography to ensure the execution of the AI inference is correctly but the problem for that is a huge overhead normally is like thousand even 10,000 ask the cost to do a Zam out make it verifiable on blockchain so that's why we proposed OPM out a year ago that we use the interactive frog proof to make the merco tree make the inference Trad become a merco tree do a binary research do the interaction game use Game Theory and toonomic to ensure the correctness of the AI inference on blockchain and uh it's really high performance and it's affordable decentralized and uh have many benefit but there's a some problem there is that because this op mechanism so you have to provide a prom and a model to everyone so like but for some important transaction if using defi you use the AI model to execute a defi transaction for you you don't want anyone to front run your transaction so the privacy of the input layer is really in critical so hence we have the OPI optimistic pracy preserve AI on blockchain we just announced the paper released the paper two year two days ago and essentially we used the ZK to preserve the Privacy for the input layer and the rest isq by op so that the whole efficiency of the system is drastically improved it so since we have like opai this end game and the best practice of onchan AI so that we can have like the um provable inference result in smart contract on ethereum and um so the workflow we provide for the developer right now is extremely simple we have the enchant AI Oracle and this m cont chat can just send a request with like what Transaction what no what model you want to use what's the prom is and send the transaction into the oo SM contract then you'll be able to have the reserve and the as the codb and it will return to the smart contract and how oao function is that they will delegate the machine learning inference into all the no is permissionless everybody can execute the no and use the op mechanism to ensure the inference is correct and um so with the on Oracle and opml we have like at least 1,000s cheaper than dkml when you execute Oni inference and this word is like a phenomenal we be able to build any AI power D app today imagine that you have like a compound that has the AI to management managing what asset they should they can be used on the compound what asset you can caliz as a stable Con on makeer down and even like some onchain gaming it's like it's like going to change the whole gaming industry so you'll be able to have like a onchain Pokemon that have a personality that have been Prov it on the blockchain is immutable most importantly and you can trade this Pokemon on blockchain to another player and um having a good AI model is really important and we need to because that we have at least like 10,000 AI model in the world right now we obviously we cannot like have all this AI model supporting on blockchain in oao in onchi Oracle so we will need a selection mechanism to like to select to F to nominate the best AI model we need on blockchain so now's research news research have proved is really correct to use the token to do the selection for the model in the M sub netex of big taner we are able to have like purely the community G one we have the uh the open source language model that is really close the performance of C gbd4 which open night have spent multi-million dollar to train and build and with crypto we just use the com the power of community we will be able to have something similar and it's fully open source and everybody have the credit when they contribute to the AI model so the selection mechanism have been proved it success and work out so when we want the community to contribute to the AI model we will need to get them incentive right the biggest incentive is the ownership so we will need to provide the ownership to the community to let them have the ownership of the AI model that have they create for the community so that's why we proposed a EIP today ERC 7641 en chanic Revenue sharing token so this EIP will ensure the the token share the revenue that AI model have produced to the community to the Token holder the differ ASC 7641 AKA erc20 m m stand for model the difference between the ERC 20m and erc20 is that like yes C20 M will have inic value that provide by a AI model on blockchain so every time the anybody or any smart contract pay the fee in oao then this Revenue will directly go to the uh ERC 20m smart contract and then all the to all the model contributor we have the air drop for this token and then we have the Rand new sharing when people pay for the model uh they build and today I want to tell you guys we want to go to announce the IMO initial model offering we are tokenizing AI model on chain so IMO is is really powerful mechanism that we can use on ethereum to empower the open source AI Community the IMO implementation is based on two important component the first component is the ownership for the model so all the contributor for the model we have a piece of the ERC 20m token on ethereum and the second part is that when you put the is the inference asset when you put the AI model on blockchain when you put the inference on blockchain like you are not just generating one asset the model token you also have the second ass along with that is the inference asset it's like nft it's the ERC 707 uh aigc nft so it's like this nft is purely generated by AI model and it's verifiable which AI model generate this certain piece of content so we'll be able to dedicate the ownership of the AI model and the content decentralized Le on blockchain so with these two major asset we will be able to perform empower the open source AI Community with the power of crypto and ethereum so please check out aura. slimo we have more information there and about Aura we are a verifiable Oracle protocol that en that enable AI on all blockchain not just on ethereum right now we just support evm or those and uh this is a QR code for of information our mission is like May AI transparent open source and democratized yeah thank [Applause] you any question yeah any questions for cotton oh oh good let's give him another round of applause well done ke C that was super interesting thank you very much just going to take a couple of minutes before we bring on the next speaker uh so yeah bear with us
Automatic transcript — names and jargon may be misspelled.