The source code for this blog is available on GitHub.

AI as an interface

One of the most exciting features of AI for me is using AI as an interface. Currently, AI can be bad when used as an agent. It can have hallucinations (source?) and be easily jailbroken.

Instead of making AI-powered DAOs, AI could be used to moderate discussions, sum up discussions, and give ideas. For a prediction market, AI could:

  • give some analysis to make better predictions,
  • analyze trends and create prediction markets.

In UX, AI is already massively used by platforms such as YouTube and Spotify by filtering and suggesting content to users. There are some risks. For example, How can we be sure AI is not biased? are legitimate questions. This is why projects like Tournesol exist.

Using AI as an interface helps to avoid common failures when using AI, but it still has some risks.

Basically, AI as an interface is like giving every user a 200 IQ brain. Users still control content.

During ETHGlobal Agents, my team and I built an AI-meme battle platform.

oui

The use case of AI on our platform was not to use AI as a player. Instead, we are using AI as an interface. It's still users who submit memes on our platform, not AI agents.