Gaia is Building ‘Living Knowledge Systems’ for AI Agents

AI is set to power the fourth incarnation of the internet, according to Matt Wright, CEO and Co-Founder of decentralized AI infrastructure platform Gaia.

“We had read internet, we had read-write, we had read-write-own with Web3,” he told Rug Radio’s Farokh Sarmad at the media platform’s R HAUS event at Art Basel Miami. “This is now read-write-own and think.”

But who determines what AI thinks is becoming a fraught question, Wright argued, with power fast becoming concentrated among “very fast, powerful, centralized AI platforms” like OpenAI’s ChatGPT, overseen by a “small group of dudes in Silicon Valley.”

“You are putting your data into these applications, they're then doing who knows what with it,” Wright said. “They're storing it in certain ways that they have control over—they're also governing how these AI systems develop over time.”

Gaia’s solution to this centralization risk is to create “living knowledge systems” built around AI agents, Wright said. “We essentially are taking open source large language models and open source developer tooling for building your own AI agents and agent inference, and we have an application runtime where you can essentially run an agent in a much more lightweight fashion,” he explained.

A "living network" of AI agents

That, in turn, will enable the creation of a decentralized network where users can create agentic workflows and agent functionalities using their data. “We can start seeing a living network of agents that are taking our knowledge and eventually building applications and monetizing it, which is quite wild,” Wright said.

“We're basically enabling agents to take our own knowledge, put it in a platform where we can basically be the co-owners of these systems,” he said, adding that Gaia’s platform will enable users to build AI applications that are “more peer-to-peer than having to go through some third party—having OpenAI tell us what to do with our data.”

Gaia’s decentralized infrastructure lends itself to integrating blockchain with AI, Wright said, adding that, “We have an infrastructure where we see inference,” When an AI’s large language model and data are combined, he explained, “it comes up with a bunch of zeroes and ones in different buckets—we call those embeddings.”

Those embeddings, he said, are “light enough that we can put it onto a blockchain and make it open to a community.”

Combining AI with blockchain opens up an array of potential applications, Wright said, with “a bunch of different frameworks” where AI can effectively talk to a blockchain. “You can do anything that a human can do with wallets and with interacting with smart contracts and tokenization,” he explained, adding that AI agents are already, “deploying their own tokens, they’re day trading on-chain.”

Sustainable decentralized AI

“We keep thinking about AI agents as a tool,” Wright said. Instead, they should be considered as “independent entities.” In just a few years, he argued, “You’re going to have more AI agents than humans on Earth acting on the internet.” Those agents, “don’t need money, they don’t pay rent, they don’t need to eat, they don’t go to the club until three in the morning.”

Without human needs, “their currency might look more like reputation,” Wright said. “They might want lightweight attestations or attributions for doing good tasks. And they want our loyalty—they want to know that they were a good boy.”

That, said Wright, means that decentralized AI could be programmed to be “more sustainable or maintaining of the system than destructive.”

Wright argued that negative portrayals of AI in the media—“It's going to take over, it's going to dominate, it's very sci-fi”—reflect issues with centralized AI. “That’s just human behavior at scale,” he said, pointing out that “technology is really just a mirror image of humans.”

Because decentralized AI agents such as Gaia work on a co-ownership model rather than centralized AI’s extractive model, Wright said, it enables the creation of tools that centralized AI simply can’t offer. “There's really cool Web3 primitives you can use to access AI and also to permission or create community engagement rewards for participating in that system,” he said. “You can't do that in the centralized AI world.”