● Advanced RMIT

Artificial Intelligence and Web3

10 minutes a year ago

Today there are two important technologies transforming our economies and societies: Web3 and generative artificial intelligence (AI).

We are continuing to build-out a decentralised stack of web3 infrastructure, underpinned by blockchains and smart contracts. This web3 stack is facilitating new types of digital money, contracts and organisations.

At the same time, we have seen rapid adoption and experimentation with generative artificial intelligence (AI) models such as chatGPT and Stable Diffusion. These models have opened to the public the power of prediction engines.

But how do these two frontier technologies interact? How can decentralised web3 infrastructure support AI. And how can open powerful AI models help build and govern the web3 stack?

The core connection between AI and web3 is through the role of data and governance. We explore how web3 facilitates authenticated data markets, the role of AI in web3 governance, and the potential for smart contracting between AI agents.

Web3 authenticated data

Generative AI models need to be trained to produce effective outputs. But what data was used to build the AI model and produce an output? Today most of this training happens in a black box. For users this means it can be difficult to assess whether the output is credible and trustworthy.

As a verifiable source of truth, blockchains might combat some of these black box issues. Blockchains can facilitate transparent and immutable data tracking not just of money, but of any data. Non-fungible tokens (NFTs), for instance, can represent individual data inputs. Over the medium term it might become possible for users to determine if their work has been used to produce an AI output, subsequently facilitating royalties or other compensation.

More liquid data markets

Not only can Web3 facilitate authenticated data, but can also be the infrastructure for more liquid data markets. Web3 infrastructure enables users to come together to collectively pool their data and sell it.

Markets need both supply and demand. Liquid data markets not only need a supply of data (including through verified NFTs). There also needs to be demand for those data rights.

Many AI models need vast amounts of data for training the models. Through the development of increasingly specific and differentiated AI models and wrappers, we may see demand for more high quality and bespoke data sets. Web3-based infrastructure could facilitate that market.

An example is useful here. A group of researchers could monetise their data by pooling it together and collectively deciding how it is used. They can represent their data rights on blockchain networks, and establish this group as a DAO, which is governed by rules written in code and executed in smart contracts. DAO members can then collectively decide whether to sell, licence or share their data with different entities.

Conversely, a group could establish a DAO with the intention of pooling resources to acquire data sets and its associated IP. This data could then be integrated into AI models, such as for revitalising a video game that never saw a sequel but maintains a cult following. Using generative AI, if a group acquired the appropriate training data, they could generate additional content or alternative endings. Similarly, these possibilities exist for movies, TV shows, books, artists, or musicians, among other areas.

Incentivising training data

Many web3 protocols use token economic incentives to direct behaviours of participants. For instance, protocols distribute tokens to addresses who provide liquidity or security. This provides a direct monetary benefit for desirable behaviours in decentralised ecosystems.

Throughout decentralised finance (defi) summer of 2020, these token incentive models became widely deployed to bootstrap and grow ecosystems.

The same business model logic might be deployed to generative AI models. For instance, the producer of a money may want to incentivise users to provide quality training data (and penalise low quality data) and can do so through web3 token incentive models. Users might simply be rewarded for providing rights to the prompts they have made.

Using AI for web3 governance

Because the world is constantly changing, organisations need to make decisions about what to do next, such as what product to launch or how to manage an incident. For millennia humans have implemented mechanisms (e.g. voting, committees) to make these collective decisions. Yet despite improvements in decision-making processes there is still human error and bias.

Human-based governance can and is augmented through AI-based governance. Of course AI models have their limitations too, including a lack of complete information, and biases of their own. Nevertheless, AI will continue to become an important decision-making tool for businesses, including augmenting human judgements in business processes.

In web3 organisations, including DAOs, the importance of collective governance mechanisms is underappreciated. Making decisions in DAOs is hard, including for users who don’t have the capacity to deeply understand each issue that goes to a vote. For decisions such as treasury diversification and adjusting parameters, we expect AI tooling to be more deeply integrated into Web3.

Contracting AI agents

One popular image of an AI future is that there will be one single AI. But just as web3 is multi-chain, the AI landscape will involve many specific applications of AI. We are already in an environment with many different AI models competing and interacting in an open way. Furthermore, the tools that enable us to customise and build specific instances of AI are emerging today.

It is in this future that we can see a deeper convergence of web3 infrastructure and AI. We can delegate some decision making power to generative AI models. These AI models become our agents. Importantly, we may want AI to execute on some decisions. Our aim is that the way that generative AI agents will interact in the world is through smart contracts and web3 infrastructure. For instance, your AI agent could spot an opportunity to trade an asset, and, with your permission, execute that trade via a smart contract on a blockchain.

Converging technology trends

In this article we’ve explored how two frontier technological trends intersect. At the most fundamental level, web3 infrastructure looks to become a verifiable source of authenticated data in a world of deep fakes and AI hallucinations. Timestamped and verifiable data will be critical for navigating these complex phenomena.

New web3-based data infrastructure also enables new types of data markets. We will observe more buying and selling of quality authenticated data. This not only enables data monetisation for producers but incentivises the production of training data for AI models.

Finally, we explored AI, web3 and governance. In the first instance, AI is likely to be integrated into the decision-making processes of different organisations, including DAOs. More deeply and speculatively, web3 can provide an infrastructure on which different AI agents can contract and organise, more fully realising the decision-making potential that AI presents.

--
Dr Darcy Allen and Dr Aaron Lane are with the RMIT Blockchain Innovation Hub.
The authors thank the research assistance of Tulley Kearney.

Share this article

Further Reading


● Advanced RMIT
Non-Fungible Tokens are Digital Property Rights
An overview of Non-Fungible Tokens (NFTs) and a look at novel ways they can be used.
8 minutes 4 months ago
● Beginner Crypto Basics
Ethereum vs Solana: Battle of The Blockchains
A look at two of the most popular cryptocurrencies, what are their similarities and differences? And why have they become so popular?
5 minutes 3 months ago
● Intermediate Crypto Basics
Volume and Market Cap? Get to know some key terms
A brief description of the terms Volume and Market Cap as they relate to cryptocurrency and why it is important to understand them.
6 minutes 3 years ago

Join 2.5 million other users
and start earning!