One of the co-founders of DeepMind, AI pioneer Shane Legg, has reaffirmed his long-standing view that there is roughly a 50 % chance of achieving “minimal AGI” by 2028 — a milestone where artificial intelligence systems could handle a broad range of human-level cognitive tasks. His comments have reignited debate about how soon artificial general intelligence (AGI) might become reality
🧠 What Legg Means by “Minimal AGI”
Legg’s prediction focuses on minimal AGI — an AI system capable of performing most cognitive tasks typical of humans, even if not at superhuman or entirely general levels. He explains that if AI systems can reliably complete a comprehensive battery of human cognitive tasks without obvious weak points, that could qualify as reaching AGI
This perspective places the odds of minimal AGI by 2028 at about 50 %, reflecting both optimism about rapid AI progress and acknowledgment of substantial uncertainty in research.
📍 A Long-Standing Prediction From a Founding Voice
Legg has maintained this timeline for years, originally laying out a similar forecast over a decade ago based on trends in computation, data scale, and algorithmic advancement. He first articulated his 2028 probability estimate many years ago and has reiterated it in recent discussions with AI experts and public commentators.
His position stems from the belief that exponential growth in computing power and data, combined with the continued development of scalable algorithms, could unlock capabilities necessary for AGI within this decade.
🧩 How This Fits With Broader AGI Forecasts
Legg’s 50 % chance estimate is notably on the more optimistic end of expert predictions. While surveys of machine-learning researchers often place median expectations for fully general AI much later — frequently around the 2040s or beyond — voices like Legg’s and other tech leaders underscore the wide range of views on AGI timelines.
Experts emphasise that defining AGI remains subjective, and achieving full AGI — capable of exceptional accomplishments across all domains — could take substantially longer than early benchmarks like minimal human-level capability
📊 Industry Reaction and Implications
Legg’s projection has reverberated through the AI community, where ongoing advances in large-language models, multimodal systems, and generative tools fuel both excitement and caution about the pace of progress. His remarks serve as a reminder that AGI timelines are highly speculative and deeply debated, with major technical and safety challenges still unresolved.
If minimal AGI were achieved by 2028, it could have far-reaching impacts on technology, economy, labour, ethics, and regulation — intensifying the global conversation about how best to manage increasingly powerful AI systems.
