Compass Investments

The head of Eliza Lab, Shaw Walters, states that current artificial intelligence systems already meet his AGI criteria.

He raises concerns that autonomous agents pose significant security threats, including rapid adoption and the potential to compromise cryptocurrency wallets.

According to Walters, fully decentralized AI does not yet exist, and the closest thing to it is local execution.

Universal artificial intelligence (AGI) may have arrived.

That’s the opinion of Eliza Labs founder Shaw Walters, who Decrypt spoke with last week during ETHDenver. Walters believes that advanced modern models already meet his definition of AGI.

I think we’ve reached a tipping point where we have AGI, he said. I am absolutely convinced that this is a universal intelligence. It’s very different from us. It learns in a very different way, but it is intelligent nonetheless, and in a very broad sense.

Walters, who originally went by the nickname ai16z, founded Eliza Labs in 2024 to develop the open-source ElizaOS, one of the first tools for creating autonomous AI agents for the blockchain.

The term “general-purpose artificial intelligence” first appeared in 1997 and has since been popularized by researchers including SingularityNET founder Ben Hertzel. AGI refers to a theoretical form of AI that can match or exceed human cognitive capabilities in a large number of tasks.

Although prominent AI developers, including Sam Altman of OpenAI and Dario Amodei of Anthropic, predict the emergence of AGI within the next decade, Walters rejects the idea of its emergence as the sole dominant entity.

I just don’t see it as some kind of AI god, he commented.

“There will never be one, because life prefers variety. Vitalik Buterin is calling for a different vector of AI development that eschews the blind race to AGI and instead relies on decentralization, verification, and Ethereum-like privacy as defense mechanisms for the coming AI era.

“There is a fundamental flaw in the very concept of working toward AGI, Ethereum co-founder Buterin wrote in a post on X on Monday, pointing out that the goal is often perceived as an undifferentiated competition where the only difference is who comes out on top. He likened it to a fi…

Walters said he first started working on AI agents during the GPT-3 era, when getting structured results was not always reliable.

I felt like most of my work was just putting wheels on a baby, he shared.

Just keeping it working, getting it to respond in the structure that I needed to process the actions. That was a major challenge.

Progress became apparent with the release of GPT-4 in 2023, which Walters said began producing more predictable results. “It was exceptionally good at providing structured answers, and now I could actually initiate action, he said. “It was at that point that we went from an agent who was barely functioning to an agent who could do something, even though he was still very limited.