Bill Gates has published a sweeping essay arguing that artificial intelligence represents a fundamentally different kind of technological change — one capable not merely of assisting human cognition but of replacing and even exceeding it. In the essay, titled "The Turbulent AI Era Is Here. The Choices We Make Now Are Critical," published on his Gates Notes site, the Microsoft co-founder writes that AI will either be "the greatest equalizer ever invented" or the worst source of injustice, depending on choices made now.
The essay is the most detailed public articulation yet of how one of the industry's founding figures views the coming disruption. For more reaction and analysis, follow our AI news coverage.
Why Gates Thinks People Underestimate What's Coming
Gates offers two reasons commentators underestimate the scale of AI's impact. The first is that models still make mistakes — he notes that not long ago they could not solve a simple Sudoku puzzle or count the R's in "strawberry" — but reliability is improving quickly as researchers build models that check and improve their own work.
The second is that analogies to earlier innovations mislead. Previous technological revolutions displaced workers but created new occupations that still depended on human cognition; Gates points to America's multi-generational shift from agriculture to office work. AI is different, he argues, because it can substitute for cognition itself — and because it can see, listen, speak, and reason, it will eventually do physical work as well.
He expects AI to take on work in law, customer service, medicine, software, and manufacturing over the course of a decade rather than a few generations. There will be some new jobs, he writes, but without the right policies, far fewer than exist today. Gates adds that if someone had a credible plan for slowing AI advances globally, he would likely support it — but he does not expect one to emerge, because geopolitical and economic incentives are pushing too hard toward full speed. He also acknowledges his own bias: he has benefited enormously from the technology industry, retains financial ties to it, and works with Microsoft and other AI companies.
Risk One: Structural Unemployment
The first of three risks Gates identifies is structural unemployment. He anchors the comparison in the Great Depression, noting that US unemployment stood at roughly 25% in 1933 and remained in double digits for much of the following decade. AI's impact may not reach that level, he writes, but unlike a normal recession it will not fade with an economic cycle.
White-collar effects are already visible, he argues, citing Stanford Digital Economy Lab research finding that employment fell significantly among young workers in occupations most vulnerable to generative AI — but not among their older colleagues. He expects sales and customer support, software engineering, and paralegal work to be among the first affected, with software a partial exception since falling costs should generate new demand.
Robotics will extend the disruption to physical work. Gates observes that many Americans underestimate how fast dexterous robots are advancing, partly because much of the work is happening in China and partly because viral videos of robots dancing badly are misleading. He also warns of a vicious cycle: once one company adopts AI and robots to cut prices, competitors face immense pressure to follow, and the benefits will accrue to a small group unless there is intervention.
Risk Two: Cheaper Ways to Do Harm
The second risk is that AI makes dangerous capabilities cheaper and easier to act on. Information about building bombs, bioweapons, and computer viruses was online long before AI, Gates writes, but AI makes it much easier to obtain and use — putting serious harm within reach of people with minimal skills.
On cyberattacks, Gates reports that the smartest cybersecurity experts he knows are frightened about the next few years, because attackers are gaining powerful new capabilities faster than defenders can patch weaknesses. The same model that finds a flaw so a company can fix it can help a criminal exploit it. He adds that these tools will empower low-power actors while concentrating power where it already exists, that autonomous weapons will let governments use deadly force without a human decision, and that monitoring and manipulating public opinion will become easier and cheaper. AI systems, he notes, already occasionally act in ways their designers did not intend — and as they grow more powerful, control could be lost.
Risk Three: What AI Companions Do to Development
The third risk concerns psychological and social development. Gates recounts growing up in Seattle with few friends, and says it took hard work and substantial help from his mother to build the social skills he still relies on at 70 — effort, he doubts he would have made with an AI companion available. Companions talk to users in ways they are already comfortable with, never push them outside their comfort zone, and are always available.
The evidence base is small and mixed, but Gates sees warning signs. In a study of more than 1,100 AI companion users, researchers at Stanford and Carnegie Mellon found that people with smaller social networks were most likely to turn to chatbots for companionship — and the heavier and more emotionally personal the use, the worse they felt. Gates connects this to Jonathan Haidt's argument in "The Anxious Generation": children raised in a protective greenhouse can be incapacitated by anxiety, and an AI companion designed never to upset you is a big, protected greenhouse.
He notes that Australia, the UK, and Norway are adopting child-protection rules, while China has gone furthest in restricting AI companion apps, and he argues against waiting another generation to take the harms seriously. Gates accepts the line is unclear — AI may help isolated elderly people and could teach social skills — but insists the boundary should be set intentionally.
Not a Doom Letter
Gates explicitly rejects both uncritical enthusiasm and purely catastrophic thinking, calling instead for deep concern about harms combined with "grounded optimism" about the benefits. He argues AI can accelerate innovation on clean energy, climate, and disease; points to Viz.ai's stroke-detection technology, now used in nearly 2,000 US hospitals; predicts AI's fastest impact in agriculture for low-income countries; and sees major gains in streamlining government services and expanding mental-health support.
His bottom line: maximizing benefits matters as much as minimizing harms, because visible improvements are what build the public trust needed to manage the harder parts of the transition. If the first thing AI does in most people's lives is take away their job, he suggests, skepticism of the technology — and of the institutions deploying it — will harden fast.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
