Ethereum Founder Vitalik Buterin's article: The prospects and challenges of integrating AI and blockchain

Ethereum founder Vitalik Buterin believes that there are promising applications in integrating blockchain ecosystem, cryptography, and AI. He categorizes this into four major types and discusses the prospects and challenges of each category.

How is AI integrated in the field of Cryptocurrency?

Since the advent of AI, the cryptocurrency field has been exploring the most effective integration between blockchain and AI. What is it?

Vitalik believes:

The decentralization of blockchain can balance the centralization of AI. AI is opaque, while blockchain brings transparency; AI needs data, and blockchain is suitable for storing and tracking data. When asked to delve deeper and discuss specific applications, my answer has always been disappointing: "There are some, but not too many integrations."

However, Vitalik has seen changes over the past three years. There are indeed some promising applications in the integration of the blockchain ecosystem, cryptography, and AI. He categorizes this into four major categories and discusses the prospects and challenges of each category.

Category One: AI as a Participant Highly Feasible

This is actually a category that has been around for nearly a decade. Vitalik uses arbitrage bots as an example. Since the widespread use of decentralized exchanges (DEXes), there are trading opportunities whenever there is a trade, and the arbitrage ability of bots far exceeds that of humans. He expects AI to quickly touch on many other applications, such as prediction markets like AIOmen as a case where AI becomes a participant.

He points out:

Prediction markets have not been very successful in practice for a long time. The main reason is that the biggest participants are usually irrational, and those with the correct knowledge are unwilling to invest time and money to participate. However, AI's involvement may change this situation. AI can work for less than $1 per hour and has encyclopedic knowledge, and can even combine real-time internet search capabilities.

This is a powerful basic function because once AI can predict these small things, it can also be applied to many types of questions:

• What will happen to X's Twitter stock price?

• Is the message I received really from Elon Musk's account?

• Is 0x1b54....98c3 the address of the ERC20 token "Casinu Inu"?

Category Two: AI Front-end Potential and Risks Coexist

The idea of AI as a software front-end is that it can clearly identify and explain users' cybersecurity concerns in the online world. In the blockchain field, AI can provide richer and more humanized explanations to help users understand the types of Dapps they are participating in and the consequences of signing complex transactions. Vitalik believes that Metamask's scam detection feature is one of the existing examples.

Another example is the simulation feature of Rabby Wallet, which can show users the expected results after signing a transaction.

In the image below, Rabby explains to Vitalik the scenario where after signing a transaction, all "bitcoins" are exchanged for some ERC20 scam coins, not real BTC.

Vitalik believes that some projects will start developing in this direction. For example, LangChain wallet uses AI as a front-end. However, relying entirely on AI as a front-end interface may also pose significant risks, and Vitalik will discuss this in the third category, "AI as a rule."

Category Three: AI as a Rule High-Risk

Vitalik emphasizes that this is an exciting but high-risk application. Taking blockchain as an example, if a smart contract or DAO needs to make subjective decisions, can AI become part of the smart contract or DAO to assist in decision-making?

In this scenario, smart contracts are self-executing code, and the logic and rules designed for these contracts are AI. This will make the rules more flexible and intelligent, able to automatically adapt to different situations and user needs. For example, a DeFi project might use AI to design smart contracts to dynamically adjust loan rates and evaluate based on current market conditions and risks.

However, this poses risks and challenges. Vitalik points out:

If an AI model that plays a critical role in the mechanism is closed, you cannot verify how it operates internally, so it is no better than centralized applications. If the AI is open-source, attackers can download and simulate it, deceive the model through highly optimized attacks, and even replay the attack process in real-time on the network.

Category Four: Let AI Achieve Goals Long-term but Interesting

Similar to the third point, but not only letting AI design mechanisms such as blockchain and DAO, but also aiming to create AI projects with potential broad application value.

Vitalik specifically mentioned that the NEAR Protocol team has continuously viewed this as a core goal. He pointed out that such AI projects could benefit both blockchain applications and extend beyond the blockchain field.

There are two reasons for doing this.

First, since the advent of AI, many people have expressed a desire to govern AI, mainly due to concerns about biased systems, etc. If a "safe AI black box" that allows the public to trust AI could be created through blockchain and multi-party computation (MPC), many applications would benefit. AI based on blockchain technology could be a way to achieve this goal, ensuring that the training process of AI is fair and transparent.

Secondly, from the perspective of AI cybersecurity, blockchain can give AI a secure termination switch. Once AI is detected to be at risk of abuse or threat, it can protect the entire system from risks. In addition, the unique economic incentives of cryptocurrencies also help build better AI projects. Vitalik believes that the decentralized AI project BitTensor falls into this category.

Vitalik concludes that the integration of AI and blockchain in the future is exciting, but the adoption of AI also needs to be cautious, especially when it involves applications with subjective decisions. This will be a continuously developing field, and he looks forward to seeing more interesting applications emerge, as well as unexpected challenges that may arise.