Decentralized AI firm Yuma bets on the Bittensor blockchain.

Barry Silbert isn’t done yet. The billionaire entrepreneur first made his mark in finance when, at age 17, he became the youngest person in America to receive a stockbroker’s license. The Maryland native became a Wall Street trader before starting alternative asset platform Second Market, which he sold to NASDAQ. Silbert then hit it big with bitcoin, buying a stake in 2012 when the price was $11, and building a crypto conglomerate known as Digital Currency Group. On Wednesday, it announced its next big project: Yuma, a subsidiary that aspires to compete with the likes of Google and OpenAI in the field of artificial intelligence.

The twist is that Yuma is working on a decentralized version of AI — the idea of distributing powerful technology across a loose network of independent contributors, rather than relying on a giant tech company to provide the service.

Talking with good luckSilbert likened decentralized AI to the World Wide Web, replacing the “walled garden” version of the Internet run by a handful of tech firms in the 1990s. It’s unclear whether a decentralized model of AI can hold its own in an industry where leading firms rely on massive amounts of data, high-cost chips and computing power. But Silbert says he’s convinced the permissionless version of AI is better — so much so that he’s tapping the CEO to lead Yuma for the first time in four years.

AI and Blockchain

Yuma’s decentralized AI ambitions revolve around a blockchain project called Bittensor, which launched in 2021 and offers tokens as incentives for people to contribute to a network of AI services. Founded in 2019 by a former Google engineer, Bitner is not widely known but has the backing of wealthy investors, including Silbert and venture capitalist Olaf Carlson Wei, known as TAO. Buying tokens.

Acknowledging that both tokens and AI have been popular fodder for hucksters, Silbert says Yuma is underestimating crypto angles because “blockchain scares people.” Yuma will instead focus on helping build networks of decentralized intelligence and computing services in the form of what Bitner calls “subnets,” he says. Applications are similar, and Yuma currently supports about 60 of them, but Silbert believes there will soon be thousands.

Still, crypto is very much part of the equation as Bittensor and Yuma count on TAO tokens to provide incentives that convince people to contribute to a decentralized AI network.

Like Bitcoin, TAO tokens are mined using electricity and will be mined overtime with a total supply of 21 million. Currently, TAO has a market cap of around $3.5 billion, making it the 34th most popular cryptocurrency, far behind the likes of Ethereum, which is 100 times the size.

For now, the Bittensor network is still in the early stages of development so there is little in the way of everyday AI applications for mainstream users. There is also the question of whether, when these applications arrive, they will be able to overcome the complexity and complex user interfaces that have characterized both crypto and decentralized projects.

Silbert says he’s confident that it won’t take long for developers to push these complexities to the background and create interfaces where users don’t know they’re even using BitSor service in the first place.

Meanwhile, Michael Casey, an author and journalist who heads a group called the Decentralized AI Society, says the solution to the user’s design challenge will be provided by AI itself. He points to the burgeoning world of AI agents and predicts that it will soon be possible for users to rely on these agents to handle all kinds of nefarious applications — including decentralized AI.

Technical Challenges Facing Decentralized AI

Jeff Wilser, who hosts the podcast AI-Curious, says he’s interested in decentralized AI, and the potential to create access to a form of artificial intelligence that isn’t controlled by big tech companies. But he also points out some clear challenges: OpenAI and Google have the massive capital to develop the computing power that a successful AI project needs, and it’s not clear that a decentralized version would be the same. Will be able to pool resources.

Among the challenges is not just buying custom chips, but building centralized data centers where processing facilities are close together—a concept known as colocation that is key to AI performance. This is something that a decentralized competitor would struggle to replicate, although over time clusters of services may become closer together. At the same time, there is plenty of redundant computing power and so decentralized AI may be able to gain a foothold in areas of the industry, such as training datasets, where speed is not of the essence.

Yuma’s vision of a decentralized AI competitor powered by a hidden crypto layer may seem far-fetched to some. But skeptics might want to consider another decentralized project that has been a resounding success: Bitcoin, which is distributed around the world and has grown to become the largest of all but a handful of major companies in 15 years.

“Like the early days of Bitcoin, which promoted a new form of transparent, borderless money, we are moving from digital ownership of assets to decentralized ownership of intelligence,” Silbert said.