AI Researchers Sound Existential Warnings in New Interview Series

A series of video interviews featuring current and former artificial intelligence researchers from OpenAI, Google, and Anthropic warns that advanced AI systems pose catastrophic risks to humanity, as reported by The Verge (https://www.theverge.com/ai-artificial-intelligence/1002238/openai-google-anthropic-ai-researchers-safety-interviews).

The interview project was compiled and released on frominside.ai by Palisade Research, a nonprofit organization that studies AI capabilities and motivations. The site hosts roughly a dozen interviews alongside curated topic clips exploring whether catastrophic warnings represent industry hype and why researchers choose to build systems they consider dangerous.

Geoffrey Irving, a former employee at OpenAI and Google DeepMind, assessed the probability of catastrophic outcomes in stark terms. 'The chance of human extinction is about a coin flip, in my view,' Irving said in his interview.

Neel Nanda, a research scientist at Google DeepMind, similarly estimated 'at least a 10 percent chance that it causes human extinction,' calling that probability 'ridiculously high.' Addressing why he continues to work in the field, Nanda explained that he believes his work directly reduces existential risks, adding that companies will continue developing these systems regardless of whether individual researchers resign.

Former OpenAI researcher Daniel Kokotajlo argued that superintelligent AI 'would basically be god-like powerful,' warning that researchers currently do not know how to control such systems. 'And so probably nobody would control them,' Kokotajlo said. 'This is exactly as dangerous as it sounds and must not be allowed to happen.'

The interview subjects also addressed public skepticism surrounding AI safety claims. Google researcher Mary Phuong acknowledged the conflict, stating, 'I think you absolutely should be suspicious of what I'm saying because I am being paid by the lab.'

Despite the shared concern among participants regarding potential existential threats, the published interviews offer no single consensus framework or cohesive technical solution for managing the risks posed by superintelligent systems.