en
Back to the list

'AGI Is Too Important'—Ben Goertzel's Crypto Bet Against OpenAI

source-logo  forbes.com 4 h
image

"I don't want to give up control of what I'm doing to venture capital firms," Ben Goertzel said on the On The Margin podcast. "Because I think AGI is too important for that."

Goertzel helped put the term "artificial general intelligence" on the map with an influential book, led the team behind the Sophia robot, and has spent decades trying to build a machine that can think. His bet now cuts against almost everyone racing toward the same goal: the most important software humans ever write should not be owned by a single company. "I'm pretty adamant that the core AGI code doing the thinking should be free and open source," he said.

That conviction drives SingularityNET, the blockchain project Goertzel runs, and the broader Artificial Superintelligence Alliance, which Fetch.ai, SingularityNET and Ocean Protocol formed in 2024 to merge their tokens into one, FET. (Ocean Protocol left the alliance in late 2025.) While OpenAI and Anthropic raise tens of billions of dollars and keep their best models behind closed doors, Goertzel is wiring his AGI work to a crypto network owned by its users.

His argument starts with a problem he thinks open source alone cannot solve. Publishing the code, he says, does not help much if no one can afford to run it.

"If the code is open but the data takes a server farm to store it and you need a hyperscaler server farm to use it, the fact that the code is open doesn't help that much," he said. "What you want is to be able to roll out the first AGI on a decentralized network spanning fifty different countries and controlled by ten thousand different people."

That, in his telling, is the whole point of putting AI on a blockchain. He concedes it makes some people nervous, because a bad actor could fork the system and build something harmful. He takes the trade anyway.

"I think there are more good people than bad out there who will want to host the AI system," Goertzel said. "We'll be better off with an open and decentralized AI than we will with an AGI arms race where just a few big countries are the only ones who have AGI and they're using it to try to blow each other up."

The labs that started open and closed up

Goertzel is blunt about the companies that once shared his open ethos and then walked away from it. He pointed to the litigation between Elon Musk and Sam Altman as a record of how quickly OpenAI's founding mission shifted.

"If you look in all the transcripts of these court hearings between Elon Musk and Sam Altman, it seems Sam Altman didn't think that for very long," Goertzel said. "He very rapidly shifted and wanted to make it proprietary. The record shows he very quickly pivoted to getting hardware in a way that required them to be closed and proprietary."

He was harsher on Anthropic's origins, and unsparing about Musk's own arc from AI doomsayer to AI builder.

"Dario Amodei, his stuff was closed and proprietary from the very beginning," Goertzel said. "Elon Musk has been the one to have a sort of tortured relationship with openness. At first in 2015 he was saying AI is summoning the demon, nobody should build it. Then he set about probably trying to build it himself. But now he's doing his own closed AI development in xAI anyway."

The closed path, Goertzel allows, is simply easier: raise venture money, lock up the model, and steer toward an acquisition. He insists the open one is not impossible, only harder. "Linux and the internet itself are proof points that something can be open source, globally decentralized, and can be the foundation for people still making a lot of money also," he said.

How a token founder plans to get paid

For now, the business runs on crypto. "I have SingularityNet, which is a utility token, so it's a blockchain project," Goertzel said. "We have people running nodes of SingularityNet, hosting AI processes. We basically make money from the tokenomics of the token on this network."

He expects that to change. The plan is to keep the AGI code open while selling polished products on top of it, with the blockchain hidden from view. "We'll probably shift a little bit into web two from web three in my projects and start offering just more products for regular money," he said. "You'll sort of have our crypto network on the back end, but from the end user it'll just be using an AI service."

Goertzel says SingularityNET will release a paid tier sometime next year aimed at businesses and power users. "We'll launch something comparable to a Claude Pro or ChatGPT Pro but smarter," he said. "Runs on a decentralized blockchain based back end, but has more reasoning and more creativity than you get from chatbots now." What he will not do is chase a mass-market chatbot. "I'm not gonna try to launch a retail thing like ChatGPT, because those guys are just losing money," he said.

The agent economy he is building toward

Goertzel's pitch to ordinary users is that the next wave of advantage goes to the people who command fleets of AI agents, not to the labs.

"The next batch of victories will probably come to small groups of people orchestrating larger teams of AI agents to get their things done," Goertzel said. "It's gonna be then like how effectively did you educate and orchestrate your beneficial agent army."

Other builders in crypto are converging on the same picture, where agents do not just answer questions but spend money. "The next step, which is already starting, is that the AI agents start transacting on your behalf," Varun Kabra, of the identity chain Concordium, said on On The Margin. "They pay for things, they sign up for services, they probably handle your financial transactions now."

That hands software a job that comes loaded with risk. "Agents are fundamentally about outsourcing a purchase, and anyone who has ever outsourced a purchase knows that this comes with trade-offs," Nitya Subramanian, of crypto wallet firm Para, said on the same podcast. Goertzel's answer is to run that economy on an open network rather than a corporate cloud.

Why it matters

Goertzel still thinks human-level AGI arrives soon. "I think we can get there by 2029," he said. "If it happened in 2027, I won't be shocked. If it was 2030, I wouldn't be shocked." His worry is less about what the machines do than about who gets left behind by them.

"If there's an understanding gap between how well people at the top and people at the bottom understand what's happening with AGI, that understanding gap will exacerbate the rate of growth of inequality even worse," he said.

Whether a crypto network can out-build companies sitting on tens of billions of dollars is still unproven. Goertzel's first test ships within weeks. "We're launching within a few weeks the first downloadable version of a new agent called Omega Claw," he said. "We're gonna have the opportunity to teach our own personal agents to help us manage our lives and help us make money."

forbes.com