AI timelines
Ask a room of experts 'when will we build AI as generally capable as a human?' and you will hear answers ranging from a few years to many decades to never. AI timelines is the shorthand for this question, and for the forecasts people make about it: when major milestones like artificial general intelligence or transformative AI might arrive.
In practice, timelines are estimates, usually expressed as probabilities over dates rather than single guesses. A forecaster might say something like 'a 50 percent chance of human-level AI by some year, with wide error bars.' These estimates come from several methods: surveying AI researchers, extrapolating trends in compute and performance, reasoning about what capabilities are still missing, and aggregating forecasts on prediction platforms. The methods disagree, individual experts disagree, and even the definitions disagree, since 'human-level' or 'AGI' means different things to different people, which is part of why the spread is so wide.
Timelines matter because they shape urgency and strategy: if transformative AI is decades away, there is more room for slow, careful institution-building; if it is near, the same safety and governance work becomes pressing. But forecasts here have a poor track record and rest on shaky foundations, so treat any specific date with humility. The honest summary is not a number but a range with deep uncertainty, and beware anyone, optimist or alarmist, who states a confident year as if it were known.
Large surveys of AI researchers have produced median guesses for human-level machine intelligence that span decades and that shift noticeably from one survey to the next, a clear sign of how unsettled the question is.
Timelines are credences, not measurements; the wide, shifting spread is the honest answer, not noise to be averaged away.
Beware precise-sounding dates. Timeline forecasts have a weak track record, depend heavily on how you define the milestone, and are better read as 'wide uncertainty' than as a credible single year.