Daniel Kokotajlo, a former OpenAI employee and author of the AI 2027 scenario, has revised his timeline for the potential destruction of humanity by artificial intelligence. He now believes that AI systems will not achieve fully autonomous coding until the early 2030s, rather than 2027 as previously predicted. This shift delays the projected arrival of superintelligence to around 2034, with no specific date for a possible AI-driven catastrophe.
The AI 2027 scenario, released in April, envisioned unchecked AI development leading to a superintelligence that could outwit world leaders and destroy humanity by mid-2030. The scenario sparked debate, with US Vice-President JD Vance referencing it in an interview and NYU professor Gary Marcus dismissing it as 'pure science fiction mumbo jumbo'.
Kokotajlo's update acknowledges that progress towards artificial general intelligence (AGI) is 'somewhat slower' than initially expected. In a post on X, he wrote: 'Things seem to be going somewhat slower than the AI 2027 scenario. Our timelines were longer than 2027 when we published and now they are a bit longer still.'
Other experts have also pushed back their timelines. Malcolm Murray, an AI risk management expert and co-author of the International AI Safety Report, noted that many are realising 'how jagged AI performance is' and the 'enormous inertia in the real world' that will delay societal change. Henry Papadatos of SaferAI questioned the usefulness of the term AGI, saying it 'does not mean as much' now that AI systems are already quite general.
Despite the revised timeline, leading AI companies continue to pursue automated AI research. OpenAI CEO Sam Altman stated in October that having an automated AI researcher by March 2028 is an 'internal goal', though he acknowledged the possibility of failure. Andrea Castagna, an AI policy researcher, cautioned that dramatic AGI timelines overlook real-world complexities, such as integrating superintelligent systems into existing military strategies.
