DBC Deep Dive: The promises and challenges of AI and crypto applications
This article is part of the #DBCDeepDives series about Digital Assets and Web3 in collaboration with Descryptor. Special thanks to Zeki Erkin.
Vitalik Buterin highlights ways in which AI and crypto can strengthen each other in his recent article. In this article, he goes a step further than other experts and commentators who have pointed out that there are synergies between AI and decentralized technology. We published a Deep Dive article about this a few months ago. Following Buterin's publication, we asked Zeki Erkin, associate professor of the Cyber Security Group at TU Delft, to reflect on the article and share his ideas about how AI and blockchain/crypto can be combined.
In his article, Buterin highlights four specific categories of combinations of AI and crypto.
- AI as a player in the game (highest feasibility):
AI can participate in crypto ecosystems by using mechanisms driven by human inputs within protocols. This includes AIs that trade, invest or even participate in governance within blockchain ecosystems. The idea is that these AIs can add value through efficiency and new insights, as long as the incentives are well designed. - AI as an interface for the game (high potential, but with risks)
This is about using AI to help users navigate the complex world of crypto. AI can help filter information, provide advice, and protect users from fraud and scams. However, this requires careful development to prevent deception and manipulation by AI. - AI as the rules of the game (approach very cautiously):
In this category, AI is integrated directly into blockchains or DAOs, for example as 'AI judges' who make decisions within the ecosystem. This is a challenging area because it is difficult to make AI completely reliable and transparent. The risks of bias and error are significant and require careful consideration. - AI as the goal of the game (long term but intriguing):
This concept is about designing crypto structures for the explicit purpose of building and maintaining AI. These AIs could then be used for various purposes outside the crypto ecosystem. Crypto mechanisms can help to better reward the training of AIs or ensure that AIs do not leak or misuse private data.
Need for transparency and regulation
Erkin agrees with the potential benefits outlined by Buterin, but underlines that before we can combine AI and blockchain, there must be a clear consensus on definitions and goals. For example, what do we mean by 'privacy' and 'transparency'?
He also advocates clear rules and responsibilities for AI developers and auditors. Regulation is essential to guide innovation without harming society and to ensure that technologies are used ethically and equitably. These clear rules should mitigate the risks of autonomous AI and minimize the risks of misclassification and discrimination by AI.
He sees blockchain as a means to level the playing field by providing transparency about the datasets used to train AI models, which allows other parties to compete or collaborate on a more level playing field. In the long term, it can therefore provide checks and balances for the large tech companies.
In general, Erkin is positive about the combination of AI and blockchain, but he emphasizes that without adequate regulation and a clear understanding of the technologies, serious ethical and operational problems can arise. He proposes that European regulations, such as the GDPR, can steer tech companies towards more ethical and transparent use of AI. Erkin sees the current time as a transition period in which companies learn to deal with new technologies within the limits of these regulations.
Opportunities: yes, but under conditions
Erkin recognizes the potential of the combination of crypto and AI, but warns that much work still needs to be done to implement such applications in a socially responsible manner. If we look at Buterin's four categories, Erkin sees the greatest social added value in category 4 (designing crypto structures with the explicit goal of building and maintaining AI). For category 3, he indicates that autonomous AI can be risky and that human supervision is needed. Erkin emphasizes the importance of human involvement and accountability when AI is deployed for critical functions. The first category (AI as a player in the game) also seems to Erkin to be the most feasible in the short term, but it requires a lot of human work to set it up properly.