Why AI won't cause mass unemployment without a breakthrough on AGI
29 November 2025
Geoffrey Hinton recently discussed the threat of AI to jobs at George Town University on 18 November 2025 (source: video).
He believes the AI revolution will not follow the historical pattern of previous industrial revolutions. His core concern: the people who lose their jobs won’t have new jobs to turn to because AI will eventually be able to do any job a human can.
I respect his insight, but I think it’s crucial to look at the massive gap between today’s Generative AI and true Artificial General Intelligence (AGI).
Before we panic about total job replacement, let’s consider three major limitations of current State-of-the-Art AI models:
-
They are fast and knowledgeable but lack versatility. Current AI is great at processing information fast and holds immense knowledge, but it fundamentally lacks the human-level versatility required for complex, novel, and embodied real-world tasks.
-
They can’t learn continuously. Current AI has a strict separation: a long, energy-intensive training phase, and then an inference (use) phase. It cannot just learn a new skill continuously or adapt instantly in the way a person does.
-
They are extremely energy-intensive. The sheer energy cost of current AI is a major obstacle to scaling it up to replace millions of human workers. The human brain is incredibly efficient by comparison.
This leads to the biggest point: achieving AGI — AI that can truly replace humans in any job — is not just a matter of making models bigger. It’s a problem that likely requires a scientific breakthrough, perhaps on the scale of curing Alzheimer’s or cancer.
Why would we assume the path to AGI is easier than these other monumental scientific challenges? Is the secret of intelligence easier to unlock than the secrets of Alzheimer’s or cancer?
It seems premature to worry about AI replacing all human jobs until we are much closer to solving the fundamental scientific problems that separate today’s tools from a truly general, versatile intelligence.