ERCs in Focus - ERC-8028: AI Assets On-Chain with Data Anchoring Token (DAT) with Thiru

Transcript

[music] Hi. Hi. I'm Tu de Drell at METS. So, in this presentation, I'm going to talk about this ERC 8028 draft for onchain AI assets. And this uh ERC proposal [music] like sets a new primitive for incentivedriven and human [music] aligned AI economy. And we're going to start with what this ERC 8028 draft. So it's in still in EIP stage at this point. And uh to start with it is a semi fungeible token standard. And it is designed and engineered just for the onchain AI assets. And what it differs

from other token implementation or other sim funible implementation is is it's the core features are aligned with the onchain AI as assets which includes tokenization of data models tokenization of uh like AI agents themselves and it also includes the implementation to create a fair and deterministic revenue distribution whenever onchain AI asserts is being used as well as having the usage kota. quotas and this ERC 8028 draft this token standard is called as data anchoring tokens and it is built inhouse by lazi which is a part of metis and it includes

the bundle of ownership certificate usage rights as well as the revenue sharing and it have four major metadatas. This token have metadatas of course and these four metadatas includes the unique ID for each individual tokens and this unique ids for uh like all the assets includes the models includes the the datas include the agents and includes the labeling and uh the second metadata is class is for the categorization and the value. The value suggests the revenue sharing, the usage quotas as well as the usage measuring and it also includes the proof proof for the authenticity and for DAT

implementation. The proof can be integrated in many ways. It can be ZK proof or it can be any other proving mechanism can be included. This infra is very much interoperable between various technologies. And what does the DAT standard solves that the existing standards doesn't? To start with the existing standards can be ERC20, 721 or any other standards doesn't build or engineered especially for the onchain EI assets and there's no implementation that natively supports usage quotas or revenue distribution or core revenue. So there's a no way we can able to track a particular token's

utilization or particular token's provenence of usage in any inference or any data training. So DAT solves it by providing the implementation for usage me measuring and as well as the deterministic revenue sharing. It also comes with uh like the updation of the uh like tokens metadata as well. And uh to begin with it also have the offchain payloading where the data can be anchored offchain and uh the hashes can be anchored onchain.

Yeah. So let's also check out the DAT's life cycle. So to also remember that DAT is the name of the EIP828 standard and it is implemented through the ALIT SDK and Allet is a SDK that have implementation in three major language which is TypeScript, Python and Rust and using this Alith SDK anyone can able to build a agent AI agent that have use case for any uh like any tracks and this agent can able to mint the DATs as well as well as to use the DATs from the DAT marketplace and this DATs can be used for inference data training or uh the

model training through the lazier API through private inference. What private inference means the data is contributed in a encrypted manner. So when data is in encrypted manner it is not transparent. It is not like publicly available transparent data. So it is encrypted and stored in DA it is an IPFS or any storage nodes and uh this encrypted data will be used for training and how how anyone know like what particular token includes that's where the metadata comes in. So the metadata shows the category value and the proof as well. So whenever this data is used

by any agent uh for either inference or for training the model the data contributor will get paid. So data is contributed by individual and it will be uh like contributed through the IDA. So IDAV is um a similar DAV implementation but it is for moderating the data contribution labeling and categorizing and it is the data is contributed to the IDV and um it is traded to the D marketplace and last pad is a agent token launchpad which is for creating the value for the data being contributed and the core future the core features

comes with first of all it is a semi-fundable implementation and it plays well with other ESC standards for royalties which is 2981 and for the slot mapping which is 3525 and as well as for the rentals and rules for which is for uh like renting the DATs for usage which is 4907 and 5006 and it have the minimal onchain imple interfaces. So when it comes with the uh like playing a role on uh like AI training or the model training the strain would be huge.

That's where the DAT comes with this minimal launch in implementation which includes the record usage for tracking the records which provides the feature for the uh like usage kas and uh the transfer value. So whenever uh a right or the DAT is being transferred this implementation is enabled and the settled revenue ET or yes it is for the revenue distribution whenever the data is being used by any agents or models for the inference and the claim is for getting the payout for the contributor and the metadata URI. So here where the offchain implementation comes in in a sense the data is being anchored

somewhere and the hash can be anchored in the onchain and we use the hash of the data in the metadata URI and why this onchain agents data masset like the why the assets of the onchain agents have to be anchored onchain it comes with the proving the usage of a particular data for the inference An inference can be influenced by uh like number of uh datas, number of dat. So whenever a inference happen, we can able to track what are the dats that influence the model to result in a particular uh response. So we can able

to track in the same way that inference fee will be distributed to the all the data contributors of those DATs and it is modular because uh each DATs can be pieced together to able to train or for the inference as well and it provides the deterministic revenue sharing for the data contributors. It also incentivize them in a way they can able to contribute. The the contributors can be an individual. The contributors can be organizations or the contributors can be the researchers who works with the lot with data and the main features of the DATs comes with providing and anchoring the DAT the

onchain assets for training the AI agents training the AI models. So providing in a sense this is the first part which is the data required for training the AI models and anchoring in a sense where the data is being hosted so it can be retrieved for the training purpose. So the providing sense is what when the DAT is being minted and it will be anchored on the chain for the usage later for the private inferences and the onchain provenence where the data integrity is checked through the hash the metadata URI and the agent negotiation and marketplace includes where the DAT will be listed for others to use. So whenever they pay for using

those DATs and the data contributors will get that those reward in a fair and deterministic manner and the use case includes uh any kind of DAT implementation for any tracks include like healthcare for collaborative AI projects for D5 strategies where multiple uh uh like like trading agents can come together share the ideas and to build one and whenever a particular implementation of particular strategy works the contributor will get the reward for that. The monetization uh workflow starts with encrypt and upload. This is where a user or researcher or anyone can able to provide their data and they provide the data sign a transaction. the data will

be encrypted and stored and the second step comes with uh like this encrypted data stored onchain uh through IPFS and the URL will be stored in uh like LAI chain and the user will get a DAT. This DAT will have the track whenever the data is being used and whenever it is used it also provides the reward for them. And the third step, the third phase is earning rewards. As I mentioned anyone uses those particular dats for inference the contributor will get the reward and where this value comes from.

So whenever inference is requested the inference user have to pay inference fee that inference fee will be directed towards the data contributor data labeler data uh categorizer. So all every every user or every candidate involved within this AI flow will get the reward. And the fourth is the proof and validation and data is contributed. Someone use it for inference or training. And when they use it, it will show each and every DAT that have a part in this particular inference.

So it it can be tracked it can be backtracked uh from the starting where each and every DAT's influence and each and every inference request and why this ERC 8028 draft is a big breakthrough. It starts with it is a semif fungeible token standard which is very uh userfriendly for developers and it provides the A B C for onchain AI assets. So first is anchoring. Anchoring the data provided by the users contribute the deterministic fast sharing whenever the data is being used the contributor

get paid fast deterministic way. And C for composibility it plays well with other ERC standards other token standards. So there's no issues with working together with other standards and it also provides a frictionless onboarding for tabs on for AI agents and any kind of onchain AI applications and it's not just in the paper and there's a real world example there's a application that is live which have thousands and thousands of transactions which is last boooo through this application anyone can able to contribute a data we have already uh like more than like thousands and thousands of DATs being minted and these

DATs are listed in a DAT marketplace any AI uh model trainers can able to use them and for that the contributors will get a reward and DATs includes any kind of data it can be voice data it can be uh like textual data table datas charts research datas so all kind of datas included and every DA contributor will get a reward whenever this DAT is being used and with MEIS and LA AI we can finally able to solve this data alignment problem and that's that's the whole point of uh this whole draft of D and the ERC 8028 draft and here's few SQA which includes my Twitter and LAI and my

test Twitter and also the LA docs if you want to check out how the DAT works in real for developer experience and also the DAT trade in Ethereum magic magician website. If you have any questions, I would be love to have answer and thank you so much for listening. >> Thank you so much. I understand AI is a hot topic right now and you also talked about uh semifable token. That's right. Yeah. So funible token in general people love non-f fungeible token definitely.

Yes. Now this is semifable token. So we are getting standards for all sort of application all sort of projects that we can think around Ethereum ecosystem. Uh we still have a few more minutes left. If people have question on this particular token which is there for AI application of it or about the project itself please raise your hand. I know this is a kind of early era of like adoption of uh AI as an standard. uh we really appreciate your contribution, the team's contribution and we would love to uh learn more about it. Perhaps we can record a full episode. Um for people who aren't aware, we do a a podcast called PEEP and EIP

which also includes ERC of course. So there we go through the proposal, what is the proposal right now, the highle overview as well as some technical details. And uh there is another podcast that we do is called ecosystem project demo where we find the real world application of that particular proposal. So I think this is a very good candidate for both of them. Thank you so much Theo for joining us here today. >> Yeah, thank you so much for having me.