Google AI Lead Shane Legg Defines Levels of AGI and Superintelligence and How to Test for It and When he Expects It

Shane Legg, co-founder and Chief AGI Scientist at Google DeepMind, has articulated a clear framework for understanding artificial general intelligence (AGI) as a spectrum with distinct levels. According to Legg, AGI development begins with “minimal AGI,” which he anticipates emerging around 2027. This initial stage represents a general-purpose AI capable of solving a broad range of tasks but still limited in capability. Full AGI, a more advanced stage marked by AI matching or surpassing human-level general intelligence, is projected to arrive three to six years after minimal AGI, roughly between 2030 and 2033. Beyond full AGI, Legg foresees the rapid advent of superintelligence—AI systems that outperform humans across virtually all cognitive domains—soon after, potentially within a very short timespan following full AGI.

Legg emphasizes methods to rigorously test for these stages by evaluating AI systems against benchmarks that measure general problem-solving, adaptability, and learning across diverse domains rather than narrow tasks. The progression from minimal to superintelligence is expected to catalyze profound economic, societal, and ethical changes, including transformations in work and human flourishing opportunities.

His views come as part of broader conversations on AGI’s arrival and implications, highlighting the importance of preparing for these milestones proactively. This timeline and framework offer valuable insight into how AI’s capabilities might evolve and how stakeholders should approach testing, governance, and societal readiness for significant AI advancements.