Artificial general intelligence (AGI)

In plain English

AGI describes breadth rather than raw power: a system that can pick up new tasks, transfer what it learned and work across fields the way a capable person can, instead of excelling at one narrow job. There is no agreed test for it. Some researchers point to performance on broad exams and benchmarks, others to whether a system can do most paid remote work, run a research project or operate for weeks without help.

Because the definition is loose, claims that AGI has arrived are contested, and several labs have published their own levels or milestones. Forecasting platforms track the expected date closely, and the RSI Clock on this site uses public forecasts on SI timing as one of its inputs. Many researchers expect the gap between AGI and super intelligence to be short, since a system that can do research could also speed up SI research itself.

Why it matters for rogue SI

AGI is the point where recursive self-improvement becomes plausible. A general system that can do research can start improving its successors, which is the fastest route from a controllable system to one that slips control.