Forecasts vary by more than the technology does. Some lab leaders put AGI within a few years. Large surveys of researchers cluster in the 2030s to 2050s with wide uncertainty. A substantial minority expect it will not happen with current methods at all. Anyone offering a confident date is describing a belief, not a result.
Why the range is so wide
Three reasons. First, definitions differ, so people are forecasting different finish lines. Second, extrapolation from recent progress assumes the curve continues, while the history of AI is full of plateaus. Third, incentives shape public estimates, and labs raising capital have reasons to sound close.
What would move the date
A demonstration of continual learning after deployment, reliable long-horizon autonomy, or a system that makes a genuine research contribution in an unfamiliar field. Each would be more informative than another benchmark record.
How to think about it
Plan for uncertainty rather than a date. The decisions that make an organisation resilient to AGI in five years are largely the same ones that help if it takes twenty: instrument your processes, keep humans in the loop where errors are expensive, and build the data foundations now.