KorvaJoin the waitlist
Tech

AGI Is Already Here, Databricks CEO Argues

He says current systems meet his old Berkeley lab's standard, while superintelligence would compress a millennium of thought into a minute.

AI image
All-In with Chamath, Jason, Sacks & Friedberg · Databricks CEO: AGI Is Already Here, Why Companies Are Behind, and the Superintelligence Debate

When the talk turned to the culture around artificial intelligence in the San Francisco Bay Area, Ali Ghodsi, the chief executive of data company Databricks, made a striking claim.

"I actually think we already have artificial general intelligence," he said. By that term, he meant AI that can handle much of the intellectual work humans do.

Ghodsi based that judgment on the standard he remembered from his time as a researcher in a large Berkeley AI lab around 2009 and 2010. He said most of his colleagues there shared a definition of AGI, and he believes systems in 2025 met it.

Unsure whether memory was playing tricks, he said he checked with people who had been in the lab. They agreed, he said, but told him the goalpost had moved.

His argument is not that AI is finished. He contended that even if progress stopped now, the world already has enough machine intelligence to completely transform the planet, once organizations learn how to use it.

What leading AI labs are chasing now, he said, is something different and far more extreme. He pointed to philosopher Nick Bostrom's 2014 book Superintelligence, which describes a system that could do, in Ghodsi's words, "a thousand years of human thought" in a minute, many orders of magnitude faster than any person and able to write an original doctoral thesis instantaneously.

Ghodsi noted that today's AI is relatively slow. Reaching that vision, he said, would mean running many AI systems in parallel and getting them to collaborate, which is why labs are so focused on hard math problems. His own priority, he said, is making the intelligence that already exists useful inside companies.

Questions people ask

Do other AI leaders agree that AGI has arrived?

There is no single agreed definition. OpenAI has described it as a highly autonomous system that outperforms humans at most economically valuable work, Google DeepMind has proposed levels ranging from competent to superhuman, and others emphasize adaptability across unfamiliar tasks. Ghodsi's point is that earlier academic expectations were lower than today's shifting standards.

What does superintelligence mean in Bostrom's sense?

Bostrom imagined an AI vastly smarter than all of humanity combined, capable of reshaping the world through speed, memory, and planning ability. He warned that such a system could pursue goals misaligned with human interests if not carefully controlled. Ghodsi uses that definition to separate current tools from that speculative future.

If AGI is here, why hasn't work changed more dramatically?

Ghodsi argues the bottleneck is application, not raw intelligence. Companies often lack the organized records of decisions, meetings, and work that AI would need to act on, and they face concerns about security, cost, and control of their data. Building that plumbing, he estimates, will take many years.

More in Tech