When Will AGI Arrive? What AI Labs and Forecasters Predict

Public expectations for AGI are clustered in three different time bands, mainly because people are using different definitions of “AGI”:

• Some AI lab leaders have been talking as if AGI could arrive in the late 2020s, roughly 2026–2030, but usually with a practical definition like “systems that can do most economically valuable cognitive work.”
• Metaculus forecasters currently tend to place it around the late 2020s to early 2030s:
  • Weakly general AI: around 2027
  • First general AI system publicly announced: around 2031
• Large researcher surveys are still much more conservative:
  • a major AI researcher survey put the 50% date for high-level machine intelligence around 2047, under a demanding “human-level across essentially all tasks” style definition.

What different sources are actually forecasting

The big issue is that “AGI” is not one milestone:

• Practical/economic AGI: a system that can do most useful white-collar or cognitive work well enough to be economically transformative.
• Publicly demonstrated general AI: a system that outsiders would accept as broadly general, not just a private internal claim.
• Strong human-level AGI / high-level machine intelligence: a much tougher bar, closer to “better than humans at essentially all tasks.”

Those different definitions produce very different dates.

Current public forecasts in plain English

• Lab-style forecasts are often the earliest, and tend to imply something AGI-like by the end of the decade.
• Crowd forecasting platforms are also fairly early, but a bit more cautious: around 2027 for weaker definitions, around 2031 for stronger public-demonstration definitions.
• Academic researcher surveys remain later: mid-2040s for a demanding, broad human-level capability threshold.

Best one-sentence summary

If you mean “when will AI systems start being called AGI by their developers?”: many current forecasts point to the late 2020s.
If you mean “when will a broadly accepted, strong AGI exist?”: current forecasts still often point to the early 2030s or later, and some survey-based estimates remain in the 2040s.

My synthesis

A careful reading of the current evidence suggests:

• First systems plausibly marketed or described as AGI: 2026–2028
• First broadly credible public AGI: 2028–2033
• Strong, survey-style human-level machine intelligence: mid-2040s is still a reasonable reference point

So the answer is: AI labs are generally earlier than forecasters and researchers, but the range depends heavily on what “AGI” means.