AI and Blockchain - Lessons from Vitalik’s ETH Vietnam Keynote
Trading Bots, Wallet Security, DAO Governance, Model Training & More
Blockchain and AI are two emerging technologies that many believe could have high potential for synergy if combined in the right ways. However, actually building practical applications at the intersection of these fields remains a challenge.
In his keynote address at ETH Vietnam 2024, Vitalik Buterin expanded on a previous blog post and outlined some of the potential ways AI and blockchain could intersect. He categorized these as AI as as players, interfaces, rules, and objectives.
AI as Players
Buterin highlighted that thinking about AI as players is the easiest place to start. In fact, there’s been examples of this for many years, such as AI arbitrage bots. This category covers where AIs participate in systems where the source of the incentives comes from a protocol with human inputs.
A growth area is using AI as players in prediction markets. He explained how platforms like Omen are already experimenting with allowing AI agents to participate by making informed trades. This could help improve forecast accuracy for less popular questions that lack human participation. Prediction markets themselves could even become a game, with the goal of building the most accurate AI forecasting model.
AI as Interfaces
When it comes to interfaces, AI can help users to understand the crypto world around better, with many early examples taken from the cybersecurity sector. Many wallets are rapidly advancing with features like scam detection. Buterin believes AI will play a key role in further developing sophisticated yet user-friendly interfaces over the next 5-10 years. As blockchains process more complex applications, improved interfaces will be crucial for widespread adoption.
However, integrating AI also presents security and trust challenges that need solutions. Buterin discussed how cryptography could help address issues like adversarial attacks on AI models during training. Approaches such as multi-party computation and decentralized training processes aim to produce models and provenance that users can have confidence in.
AI as Rules
For the "AI as the rules" category, Buterin gave the example of using an AI to provide judgments on complex questions about the external world, such as determining market prices as an oracle - or even acting as AI judges in the real world.
The Worldcoin project is another example, which aims to assign each person a unique digital identity. Determining who actually represents a real individual is challenging, and AI could potentially help reduce the need for trust by using techniques like computer vision to identify people. However, this also introduces risks of adversarial attacks that aim to fool the AI.
AI as Objectives
Buterin saw AI as an objective as a long-term goal. He discussed using blockchain-based open markets and incentives to help train better AI models, to better incentivize training or to prevent the AI from leaking private data or being misused.
This could allow broad participation in tasks like. Long-term goal of combining a blockchain-based DAO (decentralized autonomous organization) with cryptography to produce AI models and prove their training was conducted in an open and transparent manner.
Even without seeing the actual model, users could trust the process due to guarantees provided by the public blockchain records. This could be a way to train AI for applications where privacy and security are important concerns. However, Buterin noted adversarial attacks during training remain a major challenge to address for this approach.
The Future of AI & Blockchain
While the intersection of AI and blockchain remains a work in progress, Blockwarp is optimistic of significant and rapid progress in the next two years. Continued research and new tooling may help accelerate the development of practical, trustworthy AI-powered on-chain applications
In 2023, AI startups raised $42.5B, down just 10% from the 2022 peak hype. Generative AI was in vogue and of course many investors had pivoted from web3 to AI during the crypto winter. But outside of short term investment trends, these two technological waves are likely to merge in the next decade.
The easiest place to start is to build an AI as a player in the game. Many projects may want to start here for simplicity but better interfaces are rapidly needed for wider and safer web3 adoption. This will be critical for mainstream adoption of decentralized social and secure payments.
AI as rules and objectives will be challenging - and likely more long-term ambitions. Major societal, rather just technological, questions will need to be considered.
The future of blockchain and AI is going to be one hell of a ride!