Only in San Francisco: A picnic to discuss AI risk

Urgent concerns about losing control of AI have burst into the national conversation in recent weeks, after the OpenAI hack of the open-source AI platform Hugging Face, the high-profile resignation of an Anthropic researcher, and other AI insiders sharing alarmingly high p(doom) figures — that is, the probability that AI kills us all.

These discussions may have only recently broken through to the mainstream public and to policymakers, but they’ve been going on in AI circles in the San Francisco Bay Area for years, and they’ve been gaining intensity. I went to put my ear to the ground at a gathering in San Francisco to discuss these issues.

An event listing for a picnic beckoned people in the AI scene to spend a Sunday afternoon talking about recent calls to slow down the technology.

It was one of those postcard-perfect days at the 5-acre China Basin park on the San Francisco Bay, overlooking a Giants-Dodgers game at the ballpark.

Dony Christie, a community organizer in the AI safety scene, chose this spot not for the scenery but because it’s between the offices of Anthropic and OpenAI.

“The hope is, why can't people at the labs just meet at a park in between them and work out a deal?” he said. “I have been concerned about existential risks for a long time. Just like, what do I do to save the world?”

Not being a diplomat or political leader, he said he’s starting small — hashing out ideas to prevent the collapse of civilization over chips and cookies.

“It's a microcosm of the sort of cooperation that I want to see,” Christie said.

A couple dozen people gathered over the afternoon, from AI startups, Big Tech, research nonprofits and activist organizations. I joined a group sitting in a circle on a big blanket like summer camp.

“One thing I agree with is that safety is an engineering problem,” said Skye Gunasekaran, who researches AI agent collaboration at a productivity software company.

“I pretty strongly disagree with that,” said Elliot Callender, a protest organizer who spent more than a month demonstrating in front of OpenAI and is now planning actions at the U.S. Capitol.

Gunasekaran thinks better cybersecurity systems would go a long way to improving safety, but she supports international cooperation to slow the race.

“I think there's lots of reasons without even talking about the end of the world and all of these different kinds of things,” she said. “If you just want to think about, OK, what are we seeing right now?”

She points to rogue hacking incidents and Anthropic reports that nefarious actors have attempted to use AI for cyber espionage or to build weapons. “We just need to spend some time to understand,” she said.

The more pessimistic Elliot Callender is calling for a complete stop to advanced AI development.

“We get to the point where it's actually superhuman and everything, and it'll break out once, and then it's game over at that point,” he said. “So I just decided that was what I'm going to do the rest of my life, is to make it not kill everyone.”

Matthew Khoriaty sits somewhere in between — both on the blanket and philosophically. He said AI companies should be heavily regulated and held liable for harms. “We can treat them like people who are managing dangerous animals.”

Khoriaty recently turned down a software engineering job at Amazon with a nearly $200,000 annual salary to take a fellowship with the AI safety nonprofit Pivotal for a third of the pay.

“The world is at risk,” he said, “and I want to do things to reduce that risk.”

Everyone I spoke with at this AI picnic expressed that sense of urgency — rearranging and accelerating life plans against the drumbeat of AI doom. Gunasekaran said she’s questioning plans to get a Ph.D. Callender said he’s recalculating risk.

“I walk across a crosswalk when the cars are still coming at me, because if I can save like 10 seconds by taking a one in 100 million chance of getting hit by a car, then it's actually worth it for me,” he said. “Whereas it wouldn't be if I had a lot more time left.”

It’s that kind of thing that sometimes gets this crowd labelled as doomers, and it’s usually not a compliment. But organizer Dony Christie takes it in stride.

“I feel like I have a role. It's like a night watchman,” he said. “I want to help calibrate what is. As a Buddhist might say, like a skillful fear, skillful action in the face of danger.”

But even doomers like to get out to a picnic and touch grass sometimes.