When Bill Gates Reconsiders AI: An Essay and the Debate It Sparked

On 26 August 2026, Bill Gates published a long essay on Gates Notes titled “The turbulent AI era is here. The choices we make now are critical.” It examines the pressure AI may place on employment, security, education, human relationships, public governance and global inequality, while also considering the opportunities it may create in health care, education, agriculture and energy. Its central judgment is straightforward: AI could become the greatest equalising force humanity has invented, or the gravest source of injustice, and governments currently have no plan adequate to the transition.

Original essay: https://www.gatesnotes.com/a-turbulent-ai-era-and-critical-choices-to-make

The essay matters less because every individual claim is new than because of who is making them. Gates was one of the defining technology entrepreneurs of the personal-computer era and has since spent decades working through his foundation on global health, education and development. This is therefore not a detached commentary on AI. It is a systematic assessment by someone who helped drive the spread of an earlier general-purpose technology and later became closely involved in global public problems.

Gates argues that this technological change differs from those that came before it. The Industrial Revolution largely replaced physical labour, while the computer revolution expanded information processing. Generative AI is now entering language, analysis, design, research and judgment—activities once treated as distinctly cognitive and human. It may raise productivity, but it can also substitute for people across a wide range of tasks. Because it is accessible through natural language and works through devices that are already ubiquitous, adoption may be faster than in previous large-scale technological transitions.

Employment is therefore the essay's most prominent subject. Gates rejects the easy analogy that enough new jobs will always appear once old ones vanish. AI may affect white-collar and blue-collar work at the same time, with entry-level and mid-skilled roles exposed first. If the gains from productivity flow mainly to owners of capital while labour income, career pathways and opportunities for social participation contract, the problem will extend well beyond unemployment figures. It will affect distribution, dignity, identity and social stability.

The second category of risk is the expansion of the capacity to cause harm. AI can lower the barriers to cyberattacks, fraud, deepfakes, opinion manipulation and mass surveillance, while also increasing biosecurity risks. Gates is not concerned only with one malicious use. He is concerned that capabilities once available mainly to states or large organisations will spread to many more actors. Existing regulators are usually divided by industry and national border, whereas AI risks cut across technology, security, finance, health and international relations. No single agency can manage them alone.

The third category concerns education, psychological development and relationships. AI assistants and companion systems can offer help at any time, but they may also give children fewer opportunities to practise independent thought and shift processes that once required contact, conflict and negotiation with real people to machines. If such systems are designed to maximise engagement, they may reproduce the addictive dynamics of social media. The issue is not merely whether students will cheat. It is whether judgment, patience and the capacity to form genuine relationships will gradually erode through convenience.

Gates has not, however, adopted an anti-technology position. He continues to regard AI as a powerful tool for improving global health, education, agriculture and energy. High-quality medical and educational resources remain unequally distributed. AI may deliver expert knowledge at relatively low cost to underserved places and offer better support to health workers, teachers and farmers. For Gates, the real question is not whether AI should advance, but how its benefits can enter the public realm without leaving vulnerable groups to bear most of the risks and costs.

The essay therefore proposes several directions for governance. First, countries need coordinating mechanisms that place employment, education, security, mental health and industrial policy within a common framework. Second, a new international organisation is needed to address cross-border AI risks, with cooperation between the United States and China especially important. Gates compares this task with the governance of nuclear weapons, climate and global public health: competition will persist, but some risks require shared limits.

Third, Gates proposes “Human Reserved”: explicitly preserving a human role in certain fields. Care is his central example. Even if machines can perform some technical tasks, people facing illness, ageing or death may still need a real person to offer compassion, exercise judgment and accept responsibility. This is not simply a ban on automation. It is a demand that societies decide which activities should not be governed by efficiency alone.

Fourth, he proposes taxing robots or AI systems that replace labour, both to slow the transition and to fund retraining, social protection and public services. The proposal continues an argument Gates has made before, and it immediately became one of the most strongly opposed elements of the essay. Critics say it is difficult to define labour displaced by AI and that such a tax could reduce innovation and productivity. Supporters reply that if the labour tax base contracts while returns to capital become more concentrated, the tax system itself will have to change.

Mainstream media initially responded with intense coverage that highlighted three phrases: “no plan”, “Human Reserved” and “robot tax”. This quickly made the essay a media event, but much of the coverage was summary rather than engagement with Gates's argument. The more substantial responses came from technology leaders, AI researchers and philosophers. They exposed clearer disagreements over whether this transition really is different, whether it should be slowed, and which activities should remain human.

Gates's position has also been treated as a marked change of emphasis. In 2023, his discussions of AI still focused mainly on productivity, personalised education and medical progress. This essay shifts the question from what AI can do to whether society can absorb the changes it will cause. It is not a reversal of his view of AI's capabilities. It is a reassessment of institutional preparedness. He still expects major benefits, but no longer assumes that those benefits will emerge naturally or evenly.

Nvidia chief executive Jensen Huang represents a more conventional technological optimism. He accepts that AI will change work, but argues that earlier technological revolutions ultimately created more jobs and that AI will also be a net job creator. The disagreement is not over whether change is coming, but over how to read the historical analogy. Huang places AI within the familiar long-run path of productivity and demand expansion. Gates worries that the breadth and speed of substitution may outrun the mechanisms of adjustment.

Oren Etzioni, founding chief executive of the Allen Institute for AI, offers a position closer to “the diagnosis is broadly right, but parts of the prescription are wrong”. He agrees that AI risks and weak institutional preparation deserve serious attention, but is sceptical of slowing research, taxing robots or reserving broad classes of work for humans. This view recognises the social disruption while arguing that policy should concentrate on specific harms, redistribution and adaptive capacity rather than broadly constraining technological progress.

Pedro Domingos is more direct. He dismisses the slowing of research, a robot tax and Human Reserved as bad ideas, asking whether there are any bad AI ideas Gates does not like. He does not provide a complete argument, but his reaction represents the broad rejection found among technological optimists and anti-regulation advocates. In their view, these constraints would delay gains in productivity, medicine and science, while governments are poorly placed to decide which technologies should advance or which jobs must remain human.

Domingos's response: https://x.com/pmddomingos/status/2092791092949533108

The philosophical response moves the debate further. John Tasioulas, Professor of Ethics and Legal Philosophy at Oxford and the inaugural Director of Oxford's Institute for Ethics in AI, broadly supports Gates's call for a social transition plan and explicitly endorses Human Reserved. He nevertheless criticises the framing of AI as either the greatest equaliser or the worst source of injustice. Such heaven-or-hell binaries, he argues, do not help societies decide, because real decisions usually require trade-offs among several competing values.

Tasioulas's response: https://substack.com/@jtasioulas/note/c-323574389

His example from justice is especially important. AI may speed up the processing of cases and make some decisions more consistent, yet it may sacrifice a process in which a real person deliberates, gives reasons and accepts responsibility. Human Reserved thus becomes more than a slogan for protecting jobs. It becomes a claim about the structure of responsibility. Some matters may need to remain human not because machines perform them poorly, but because society requires agents who can judge, explain and answer for consequences. Machine efficiency and human responsibility are not performance measures on the same scale.

The debate can now be described in three positions. Gates favours new constraints that actively moderate structural change, together with domestic and international governance. Tasioulas agrees that constraints are needed but asks us to abandon simple moral binaries and confront conflicts among values and forms of responsibility. Domingos and other technological optimists argue that the constraints themselves would impede progress. Their disagreement concerns more than predictions about AI. It concerns who should determine the speed of transition, who should bear its costs, and whether public institutions may set boundaries around efficiency.

The first philosophical significance of the debate is its distinction between what machines can do and what they should do. Much technological discussion assumes that adoption is justified whenever a machine completes a task faster and more cheaply. Yet care, justice, education and public decision-making do more than deliver outcomes. They also embody relationships, procedures, reasons and responsibility. Even if a decision is statistically more accurate, questions remain: who is entitled to make it, who explains it to those affected, and who is accountable when it is wrong?

Second, Gates sometimes places task performance, cognition, understanding and subjectivity on a single continuum. Superior performance across many tasks does not by itself show that AI understands in a human way, nor that society can treat the human role entirely as a replaceable function. Conversely, an appeal to human uniqueness cannot justify rejecting AI. The difficult task is to acknowledge AI's real capacity to reorganise work and institutions without exaggerating machine subjectivity.

From a more structural perspective, Sustenesis draws attention not to whether any one risk will occur, but to whether technological capability, economic incentives, legal responsibility, education systems and the pace of human adaptation are becoming persistently misaligned. AI can expand very quickly, firms have incentives to adopt it rapidly, public institutions revise themselves more slowly, and individuals need time to rebuild skills, identities and ways of life. If these systems continue changing at different speeds, the whole may lose coherence even when every local decision appears rational. Gates's claim that there is “no plan” points to this structural gap.

A supplementary interview in The Washington Post confirms that Gates recognises his regulatory ideas are not yet specific enough to attract formal support from AI-company executives. He nevertheless plans to devote much of his future time to AI risk and social transition. This makes the essay better understood as an agenda-setting text than as a mature policy package.

The Washington Post interview: https://www.washingtonpost.com/technology/2026/08/26/bill-gates-says-he-worried-ai-will-harm-workers-kids-society/

The most accurate judgment at this stage is therefore that the essay is beginning to move beyond a media event and become an intellectual event. Human Reserved, in particular, is being discussed independently of the original essay. It has brought technological optimism, social governance and the ethics of responsibility into direct contact, while taking the employment question beyond the number of jobs to participation, dignity and responsibility.

It is not yet a policy event. So far, no government, international institution or major political party has introduced a concrete proposal in response to it. Gates himself concedes that robot taxes, regulatory institutions and international cooperation still lack executable detail. What deserves continued attention is not simply how often the essay is cited, but whether Human Reserved develops into a stable concept for thinking about human responsibility, social participation and institutional boundaries—and whether those discussions enter formal agendas for taxation, education, labour protection and AI regulation.