An expert on Artificial Intelligence issued a horror warning that the technology could “destroy humanity”, raising fears about the progression of automated superintelligence.
Daniel Kokotajlo, a former OpenAI employee, sparked debate in April by sharing AI 2027, a scenario about rapid AI growth creating a superintelligent system that outsmarts world leaders and eventually destroys humanity.
In AI 2027, Mr Kokotajlo suggests that superintelligence could arise through fully autonomous coding, where AI systems improve themselves without human input. He also notes that the launch of ChatGPT in 2022 sped up expectations for artificial general intelligence, with some experts now predicting it could arrive in just a few years instead of decades.
Kokotajlo and his team identified 2027 as the likely year AI could achieve fully autonomous coding, though they acknowledged this was only an estimate and that some predicted a longer timeline.
Recently, however, growing doubts have emerged about how soon artificial general intelligence might appear and even whether the concept itself is clearly defined or meaningful.
Malcolm Murray, an AI risk management expert and one of the authors of the International AI Safety Report said: “A lot of other people have been pushing their timelines further out in the past year, as they realise how jagged AI performance is.
“For a scenario like AI 2027 to happen, [AI] would need a lot of more practical skills that are useful in real-world complexities. I think people are starting to realise the enormous inertia in the real world that will delay complete societal change.”
Leave a comment