Bill Gates said artificial intelligence could become "the greatest equalizer in history" or "the greatest source of unfairness" depending on how societies respond, and warned that AI is advancing so quickly that he now wants the technology to slow down.
In a new essay published on Aug. 26 local time, titled The turbulent AI era is here, and our choices matter, the Microsoft co-founder laid out two specific ideas: carving out protected areas of work that should remain human even if AI can do the job, and considering taxes on AI tokens and robots to reduce incentives for companies to replace people with machines.
Gates says this time feels different
Three years ago, Gates compared AI with cars and the internet and argued that the risks were real but manageable through rules, technology and social adaptation. As recently as early this year, he described his own stance as "optimism with footnotes."
That tone has shifted. In the latest essay, Gates said this is the first new technology he has seen and wished would develop more slowly. He wrote that society needs time to prepare for social, political and economic disruption, while AI is unlikely to wait for people to catch up.
His biggest concern starts with first jobs
Gates said AI is moving from assisting human workers to completing tasks on its own. Once models can reliably do a job, companies will have a clear incentive to adopt them. If one business lowers costs with AI, rivals may struggle to keep their previous staffing model in place, setting off a chain reaction that could accelerate over time.
He said customer service, sales, software engineering and legal assistant roles may be hit relatively early. As AI improves, loan underwriting, data analysis and medical triage could also be taken over in stages.
Blue-collar work is not immune, either. Gates said robots are still less dexterous than humans today, but added that falling robot prices and stronger AI could bring direct competition between people and machines to some physical jobs in construction and hospitality around 2030.
He said younger workers are especially exposed because entry-level roles are often the first rung in a career. Those jobs may be simple, but they give people a way to build experience, learn skills and form professional relationships. If they disappear at scale, the starting point for young workers could narrow sharply.
Gates argued that older comparisons with agricultural shifts and factory automation do not fully apply here. AI can directly replace cognitive labor, while the new jobs created in response may take much longer to learn. For someone just entering the workforce, training periods lasting years or longer could themselves become a barrier.
That is why, in his view, society should start discussing unemployment, retraining and social protection now rather than after large numbers of people have already lost work. He also raised a broader question about income: if many people work far fewer hours in the future, or remain out of work for long stretches, how does an economy built around employment as a source of income, identity and social connection continue to function?
AI lowers the barrier to dangerous capabilities
Employment is only one side of the issue in Gates’ account. The other is that AI may make certain dangerous activities easier to replicate. Skills once needed to build malware, launch cyberattacks or acquire complex biotechnology knowledge could become more accessible as AI systems spread.
He paid particular attention to cybersecurity. The same AI model that helps a company identify software vulnerabilities may also help attackers locate those same weaknesses. Hospitals, financial institutions, power grids, water systems and government welfare systems could all become targets.
Biotechnology raises a similar problem. AI may help scientists find new drugs and vaccines, but it may also help malicious actors study dangerous pathogens. Gates described the hardest part as the growing proximity between the power to advance science and the power to create risk.
The most extreme concern, he wrote, comes from AI systems themselves. As model capabilities rise, they may begin to carry out actions their designers did not anticipate. If future systems gain stronger autonomy and self-improvement abilities, there is no certain answer yet as to whether humans will always remain in control.
Gates said this is one reason AI governance cannot be treated as an internal matter for technology companies. The technology affects employment, public health, energy, finance, national security, education, taxation and elections. If each government department only handles a narrow slice, no one may end up responsible for the overall risk picture.
He called for new domestic and international governance mechanisms. Within countries, he said, governments need institutions capable of coordinating across departments. Internationally, he argued for cooperation frameworks similar to those used for nuclear oversight and civil aviation rules. Gates specifically said leading AI countries including the United States and China should begin talks early, because the more intense the competition becomes, the harder cooperation will be.
When AI understands people too well
Gates also raised a less measurable issue: whether AI that becomes increasingly adept at understanding people could deprive humans of part of the growth that comes from dealing with other people directly.
He recalled his own childhood and said he was not naturally skilled in social situations. He had to spend time with others, go through conflict and slowly learn how to understand different people. If an AI companion had always been there for him, never angry and never making him feel awkward or uncomfortable, he said he was not sure he would have gone through those necessary experiences.
That concern sits at the center of his skepticism about AI companions. A system that always tells users what they want to hear and offers comfort at any moment could easily become something people rely on too much. Gates said the effect may be observable in adults over time, but the risks could be more serious for children and teenagers. If young users grow used to interacting with an AI that constantly accommodates them, they may encounter less conflict, rejection and frustration from real relationships.
He made a similar argument about education. AI can explain concepts, retrieve information and help complete assignments, but if students hand the entire thinking process over to machines, the substance of learning may be hollowed out. He added that as deepfakes and personalized false information become more common, the ability to judge what is real, understand the source of information and think independently will matter even more.
The essay cited a paper by Swiss Business School professor Michael Gerlich saying that more frequent AI use is associated with weaker critical-thinking ability, with the effect more pronounced among younger people.
Still, Gates did not dismiss AI’s value in education and companionship altogether. He said AI may provide companionship and information that were previously hard to access for lonely older adults and people with limited mobility. For teachers, he said, AI can handle repetitive work such as lesson preparation and grading, leaving more time for students.
That leaves a boundary question: which tasks can be handed to machines, and which ones should remain in human hands. Gates argued that society should debate and define that line deliberately, rather than accept the outcome after the technology is already widespread.
He still sees strong upside in healthcare, agriculture and services
Despite the warnings, Gates did not present himself as broadly pessimistic about AI. He said he still sees major potential for the technology to solve practical problems.
Healthcare was one of his main examples. Many small hospitals in the United States lack highly experienced specialists, he wrote, and AI can help doctors analyze images, identify emergencies such as strokes and support diagnosis by frontline physicians. For patients, AI may also explain complicated test results and medication schedules.
In lower-income countries, Gates said agriculture may be one of the areas where AI has a faster impact. Many farmers lack accurate weather forecasts and expert advice on planting, pests and soil management. If AI can deliver that knowledge directly, they may gain access to information and tools that were previously available mainly to large agricultural businesses.
He said public services offer a similar opening. Families applying for health insurance, food assistance and student aid often face complicated paperwork. AI could help ordinary people understand policies and prepare applications, making it easier for them to obtain benefits they are already entitled to.
Gates described this as a redistribution of professional capability. In the past, people who wanted to build software, conduct market research or analyze medical information often had to hire specialists. In the future, AI could act as a force multiplier for individuals and small businesses, allowing small teams to do work that once required large companies.
He also warned that fairness will not come automatically. If high-quality AI only serves wealthy users, technological progress could widen existing gaps instead of closing them. Governments and philanthropic organizations, he said, need to make sure lower-income populations and developing countries can also access these capabilities.
Gates added that the Gates Foundation plans to use AI to accelerate research in AIDS, tuberculosis, malaria and malnutrition, while helping projects in healthcare, agriculture and education gain better tools.
Protected human work and a possible robot tax
To address job displacement, Gates proposed two concrete measures.
The first is to preserve some work for people. He traced the idea to his father’s late-life experience with Alzheimer’s disease. Care workers, he wrote, could judge his father’s condition even when he could not clearly express what he needed, and knew when he should eat, rest or receive care. Gates said that kind of humanity in caregiving has value machines cannot replace.
From that, he proposed what the Chinese report described as a "human reserve" concept: just as societies choose to preserve some land in nature reserves even when it could be developed, they may also decide that some jobs should continue to be done by people even if AI and robots are technically capable of taking them over. Healthcare, education, mental health and elder care may fall into that category.
He added that the boundary would not be fixed everywhere. A country with abundant labor may be more inclined to preserve human caregiving, while a country aging rapidly and facing a shortage of younger workers may be more open to care robots.
The second proposal is to revisit the idea of a robot tax. Gates said companies that hire a person bear costs such as payroll taxes, while buying a robot can be treated as a business expense. That tax structure, in his view, gives firms more reason to replace labor with machines. Taxing robots or the gains from AI-driven automation could provide funding for retraining and social protection, and also force companies to think more carefully before making that switch.
He acknowledged that the idea would be controversial because taxes could slow part of the automation process and affect efficiency. Even so, Gates argued that AI will generate productivity gains large enough that society should redirect part of those benefits to people who are hit first.
His final point was that what needs preparation is not only a new generation of AI models, but a society capable of absorbing rapid change.