'Probability of doom from superintelligent AI still a real concern'

'Developing tools to accelerate AI chips not in line with Jesus'

Calls for slowing down AI development and establishing related regulations are gaining renewed attention as the technology advances at a pace many researchers find alarming. [Reuters-Yonhap]
Calls for slowing down AI development and establishing related regulations are gaining renewed attention as the technology advances at a pace many researchers find alarming. [Reuters-Yonhap]

Calls for slowing down AI development and putting regulations in place are gaining renewed attention, following the resignation of a Google researcher who cited concerns about the pace of AI advancement — the latest in a string of departures from major AI firms.

Robert O'Callahan, a researcher at Google DeepMind, Google's AI division, revealed on Sept. 24 via X, formerly Twitter, that he had left the company. "I quit Google today," he wrote, adding that he felt he had no choice but to leave because he believed AI was advancing too quickly.

O'Callahan, who said he lives in New Zealand, explained his reasons in a resignation letter published on his personal blog. "The ultimate goal of our team was to make AI much cheaper and reduce latency," he wrote. "At this point, I don't think that's a good thing for people."

He acknowledged that his departure would have little practical impact given that millions of people are contributing to the acceleration of AI development, but added that his skills are rare enough that his exit is not entirely without effect.

He also said he had tried to simply look away from AI development but concluded that doing so was not consistent with following Jesus — revealing that his religious faith played a role in his decision to resign.

O'Callahan said his work was not directly focused on AI capabilities but on improving tools for hardware chip design. Still, he said, God had led him to confront the reality that those tools would ultimately accelerate AI chip development.

"I'm not certain the probability of doom from artificial superintelligence — AI that surpasses human capabilities across most domains — is 100 percent," he wrote. "But I see the risk as real, if uncertain, and that alone is deeply troubling."

He also said that many of the phenomena pessimists had long predicted, including "reward hacking" and "misalignment," have in practice already occurred.

He added that most of the people he spoke with at DeepMind were genuinely and seriously concerned about AI.

However, he said his decision to leave Google at this particular time was driven by his personal travel plans and was unrelated to the recent wave of AI slowdown advocacy or other resignations within the AI industry.

Earlier, Jacob Coxon, a researcher who had worked at OpenAI before joining Anthropic, drew widespread attention when he announced his resignation with a warning that AI could wipe out humanity within 10 years.


greg@heraldcorp.com