Andrew McAfee

I think AI will be one of the most beneficial technologies we have developed, substantially improving living standards, scientific discovery, medicine, and environmental efficiency. That optimism comes partly from watching its capabilities repeatedly exceed expectations. We should count not only the electricity and water used by data centers, but also the resources AI can help us avoid consuming through better designs, logistics, energy systems, and discoveries. But technical capability and economy-wide transformation are not the same thing. Organizations adopt technology slowly; institutions, incentives, regulation, and workflow redesign all matter. Automating tasks also does not mechanically eliminate entire occupations. Demand can expand, jobs can change, and adoption can stall. I once worried more about mass technological unemployment, and the employment evidence pushed me away from that view. That does not make local job losses, disrupted careers, or damaged apprenticeship routes trivial. We should address demonstrated displacement rather than use it as a reason to suppress broad experimentation. I put extinction risk approximately at zero—not literally impossible, but nowhere near sufficient to justify sacrificing enormous potential benefits to poorly specified fears. Faster capability progress by itself is not evidence of catastrophe. If we saw genuinely autonomous harmful systems repeatedly defeating serious efforts to shut them down, I would update sharply and support stronger intervention. Meanwhile, practical guardrails should be built and tested. So my preferred posture is permissionless innovation combined with responsiveness: let people discover valuable applications, regulate concrete externalities such as pollution, and move quickly when actual harms appear. I expect rapidly improving AI over the next decade. I am less certain that the whole economy will be transformed by 2030, because invention can move much faster than adoption.