#AI
Bill Gates changes the tone of the AI conversation: this time is different (Rubric #AI)
In April 2023, I examined Bill Gates’s essay “The Age of AI Has Begun.” His position then was measured: the risks were real but manageable, and AI’s benefits should ultimately outweigh them. The story now has an important second act.
In a new Radio Atlantic interview, Gates does not abandon his AI optimism. He revises the expected speed and scale of the transition: model capabilities are advancing faster than institutions can devise rules.
| Gates’s position | 2023 | 2026 |
|---|---|---|
| Analogy | The risks resemble the problems created by cars, PCs, and the internet: society adapts | The old analogies are misleading: AI can replace cognitive labor |
| Work | Primarily productivity growth and a manageable transition | Gates’s forecast: office work within two years; office and physical work within four |
| Regulation | Adapt laws and develop technical safeguards | Corporate self-regulation is insufficient; binding rules are needed |
| Economy | The technology will create new roles, and people will need retraining | Some occupations may disappear; certain tasks should deliberately be reserved for people |
Key points from the conversation
- This time is different. AI spreads through familiar devices, understands natural language, and can automate not one operation but a broad range of cognitive tasks. The next layer is connecting models to robots: moving from the screen into the physical world.
- The leap in programming was the signal. Gates says the progress of models at the end of 2025, especially Claude at first, made him raise his estimate of the pace of change. The claim that models are already at or above human level remains his assessment: the interview names neither a benchmark nor a specialist level for comparison.
- Bioterrorism and cyberattacks require control before an incident. In his view, powerful models lower the knowledge barrier to finding vulnerabilities and developing dangerous biological agents. Hence the proposals to monitor the most capable systems and license dangerous capabilities. The criteria for such controls, and their applicability to open models, remain undefined.
- The labor market is no longer only a productivity story. Gates forecasts substantial consequences for white-collar jobs within two years, and for cognitive and physical labor together within four. The interview publishes no methodology for that forecast, but the mechanism is clear: if replacing people lowers one company’s costs, competitors gain an incentive to follow.
- Human Reserved. Gates proposes leaving some tasks to people even when a machine can perform them more cheaply. His direct examples are caregiving and delivering a grave diagnosis; in education and psychological care, a person should remain in charge, using AI as a tool. In his view, work provides not only income but dignity, meaning, and social connection. Who draws this boundary, and by what criteria, remains an open question.
- Taxes and rules will have to be rebuilt. Gates proposes shifting some of the burden from labor to capital, robots, and AI tokens, directing the money to retraining and social protection. He considers voluntary industry audits insufficient; countering bioterrorism and cyberattacks requires coordination among governments, including the United States and China.
- Gates nevertheless continues to see enormous potential for AI in medicine, education, agriculture, and science. This is not a reversal from “AI will help” to “AI must be stopped.” It is a move from faith in ordinary adaptation toward recognizing that beneficial and dangerous effects will arrive together, while the market, in his logic, will not choose a socially acceptable pace on its own.
In 2023, the central question was: what useful things will AI be able to do for people? Now it sits beside a more uncomfortable one: what will we want to reserve for people even when AI can do it better and more cheaply?
#AI #Governance #Economics #Security #Robotics #Engineering