Anthropic whistleblower warns AI disaster could happen 'before close of decade'
Mr Coxon said he quit Anthropic after a gradual and “scary” realisation that his own intellectual contribution was being overtaken by AI.
A former artificial intelligence researcher at Anthropic has warned that some experts fear AI could trigger a catastrophe before the end of the decade.
Listen to this article
Jacob Coxon told LBC's Andrew Marr that a near-term disaster - within the next three years - would most likely involve a biological weapon or a novel virus, rather than robots taking control of society.
He said: “Certainly some researchers believe it could happen before this decade is out.”
Mr Coxon said the risks would grow as AI became more embedded in the economy, with automated systems potentially controlling key infrastructure.
He warned: “A sufficiently smart intelligence will be intrinsically scary, irrespective of the precise scenario.”
Read more: Nearly a quarter of Brits cool AI use after 'doomsday' warnings
Read more: Burnham says AI safety should be an issue for governments and not left to chance
Asked whether an attack on banking systems or food supplies could prompt social disorder, he said: “I imagine, yeah.”
Mr Coxon said he quit Anthropic after a gradual and “scary” realisation that his own intellectual contribution was being overtaken by AI.
“There was no single moment,” he said.
“It was a gradual realisation that my own intellectual contributions were becoming less relevant when compared to what the AI could contribute.”
He called for tougher international action to prevent an AI arms race between the US and China, including potential limits on the computing power used to train models.
“If we don’t do something radical, we risk sleepwalking into a race,” he said.
Mr Coxon also suggested Donald Trump could be well placed to secure a rapid agreement if persuaded of the need to intervene.
“He does have a track record of taking bold, quick action,” he said.
“If he comes round to the position that we need to do something about the AI race, then I think he’s the right guy […] to make a quick deal.”
He added that if governments fail to act, responsibility may also fall on technology firms continuing to race ahead despite knowing the risks.
“You may also want to start putting blame on the people that are knowingly taking huge risks,” he said.