Bill Gates published a lengthy piece on his Gates Notes website today, offering a glimpse into just how deeply the Microsoft co-founder has been contemplating the societal consequences of artificial intelligence. For the most part, Gates aligns with the Responsible AI movement, suggesting that *Pacing The Frontier*—an open letter from AI staffers calling for a slowdown in development—would be a sensible step, though he doubts whether such a pause could be maintained over the long haul. He also voiced enthusiasm for AI's potential to transform science and medicine, while expressing concern about its impact on employment—themes that will sound familiar to anyone tracking Anthropic's policy discussions. Still, a handful of genuinely novel proposals could shift the debate if they gain traction.

One of his standout ideas is a "robot tax." Currently, when a company hires a person, it pays payroll taxes on that worker's wages. But if it buys a robot instead, the purchase can often be written off immediately as a business expense. That tax structure effectively nudges businesses toward automation over human labor. A tax on robots, Gates argues, would temper the speed at which machines replace people and generate revenue for retraining programs and a more robust social safety net.

He also suggests designating certain jobs as "Human Reserved," essentially prohibiting AI from taking over specific roles. It's an intriguing concept, and one that policymakers could implement with relative ease. As Gates explains, the rationale might be economic—if automating a job would displace a large group of workers who can't easily transition to new fields, that role could be protected. He points out that telling a 55-year-old construction worker to retrain for a job in elder care isn't realistic or fair. Other reasons could be more personal: in healthcare, for instance, delivering a terminal diagnosis is something a robot could technically do, but it shouldn't. The Human Reserved category would likely evolve over time, with some jobs set aside now and AI introduced gradually over years or decades, always with a commitment to preserve certain positions.

It's worth noting that both proposals would take a significant bite out of the profits of major AI labs, which might explain why these ideas haven't been widely discussed until now. There are also plenty of open questions about who would enforce such rules and what exactly they would entail, should they ever be put into practice.