Timelines to AGI

How long do people expect it will take for AGI or superintelligent AI to be developed?

• Metaculus Weak AGI estimate
  : It predicts a 50% chance of a weak AGI being publicly known by 2027.
• Yoshua Bengio (Turing award winner): 2028 to 2043, 90% confidence interval
  . Also asks himself: “And what if it was, indeed, just a few years? ”
• Geoffrey Hinton (Turing award winner): 2028 to 2053
  (he first thought it would take “at least” till 2070)

Our take

Prediction is very difficult, especially if it’s about the future. Past expert AI predictions turned out to have been overly optimistic.

In some ways AI models are already superhuman, so we could be very close to models with dangerous capabilities. Current state of the art AI models have an IQ of >130, speak 20 languages, write code in most programming languages, and can explain quantum mechanics in 18th-century German poetry. Innovation is hard to predict, and important innovations could happen at any time.

Since we’re risking human extinction, we should err on the side of caution and act as if AGI could be developed in the next few months.