For most of modern economic history, technology policy has focused on helping machines do more. Bill Gates is now asking governments to consider the opposite: identify some work that society will deliberately keep in human hands, even when artificial intelligence or robots become capable of performing it.
Gates calls this proposed domain “Human Reserved”. It is not a law, an international agreement or a finished policy framework. It is an argument that the market should not make every decision about automation—and that efficiency may not be the only value worth protecting.
In a wide-ranging essay published on 26 August, the Microsoft co-founder warned that AI could displace cognitive work much faster than earlier technologies transformed employment. His proposals include permanently reserving certain responsibilities for people, temporarily shielding workers who cannot realistically retrain, and taxing AI usage or robots to fund a stronger safety net. The primary essay also addresses cybersecurity, biological risks, education and AI companions, but its most concrete economic idea is a new boundary between work that can be automated and work society chooses not to automate.
What are Human Reserved jobs?
Human Reserved jobs are occupations or individual tasks that would remain assigned to people for social, ethical or economic reasons, regardless of whether a machine could technically perform them.
The distinction between an occupation and a task is important. Gates is not proposing that AI be excluded completely from healthcare or education. A doctor could use an AI system to review scans, for example, while a human retains responsibility for diagnosis, consent and communicating life-changing news. A teacher could use AI to create exercises while remaining accountable for a child’s development and welfare.
In an interview reported by Axios, Gates pointed to childcare and jury service as examples where human participation has value beyond measurable productivity. Caregiving, mental-health support and sensitive medical communication also illustrate the principle: the outcome matters, but so do empathy, trust, legitimacy and responsibility.
Two forms of protection: permanent and temporary
The proposal has two different layers.
- Permanent reservation: selected responsibilities would always require meaningful human control or delivery because society considers the human relationship intrinsically valuable.
- Temporary protection: automation could be phased in more slowly where rapid replacement would devastate workers or communities that have limited opportunities to retrain.
The second category turns Human Reserved from an ethical principle into a labour-market intervention. A worker approaching retirement after decades in construction cannot necessarily move into a new profession simply because an online course exists. Slowing automation in a particular role could give workers, employers and public institutions time to adapt.
That protection would not be uniform worldwide. A country with a young workforce and high unemployment may make different choices from Japan, where an ageing population and labour shortages could make robotic care more attractive. The value assigned to human work—and the economic cost of preserving it—will vary across societies.
Why Gates believes this AI transition is different
Past waves of mechanisation usually replaced physical tasks while creating demand for new forms of human judgement. Gates’s central concern is that generative AI and robotics could automate both cognition and physical execution. Natural-language systems also run through devices and software companies already use, reducing the adoption friction that slowed earlier technological transitions.
That does not establish that mass unemployment is inevitable. AI models remain unreliable in important settings, adoption is uneven, and companies frequently redesign jobs rather than eliminate them entirely. New occupations may also emerge. Gates’s forecast should therefore be read as a warning from an influential technologist, not a settled labour-market fact.
However, the speed of change makes advance planning rational even if the eventual disruption is smaller than he expects. Waiting for unemployment to rise sharply would leave governments attempting to design retraining, income support and regulation after the damage had already become visible.
How would a tax on AI tokens or robots work?
Gates connects job protection with taxation. Employers pay payroll-related taxes when they hire workers, while spending on software or machinery may receive more favourable treatment as a business investment. He argues that this difference can make replacing labour financially attractive even before the wider social cost is considered.
His proposed response is to tax AI tokens—the units used to measure input and output processed by many generative-AI systems—or robots used to replace labour. Revenue could support retraining, wage assistance and communities affected by concentrated job losses. Gates also argues that a levy could slow the transition enough to make it more manageable.
The simple formulation hides difficult design choices. A token used to discover a medicine has a different social effect from a token used to automate a call-centre interaction. Efficient software may consume fewer tokens while replacing more work. Open-source models can run privately, making usage harder to measure. A robot may assist a worker, replace one task, or eliminate an entire role. Taxing the equipment cannot by itself distinguish among these outcomes.
A workable system might therefore need to consider business outcomes, workforce reductions or defined high-risk uses—not merely count computing units. Gates acknowledges that beneficial applications in areas such as medicine and education should not be unnecessarily slowed.
The four unanswered policy questions
Who decides what remains human?
Governments could create protected categories, professional regulators could define tasks requiring human accountability, or collective-bargaining agreements could establish limits within industries. Each route risks lobbying and inconsistency. A decision that protects dignity in one sector could preserve an inefficient monopoly in another.
How would the rules be enforced?
Companies could keep a nominal human in the process while allowing an automated system to make the substantive decision. Regulation would need to define meaningful human control, auditability and legal responsibility rather than simply require a person to click an approval button.
What happens to international competitiveness?
A country that taxes automation or reserves jobs for people could face higher costs than a country that allows unrestricted deployment. Imported automated services would make national rules harder to enforce. Gates has separately argued that managing advanced AI risks will require cooperation between major powers, including the United States and China, according to Reuters.
How much work could realistically be protected?
Gates told Axios that, in an extreme scenario, he could imagine initially reserving up to 40% of jobs for people. That is not a forecast, a policy target or a calculation supported by a published methodology. It shows the scale of his thought experiment, but governments would need far more detailed occupational evidence before using any percentage as a basis for policy.
What Human Reserved could mean for India
The question is especially important for India because technology services, business-process operations and a large young workforce meet on the same fault line. AI could improve productivity and expand access to education, healthcare and public services. It could also reduce demand for entry-level coding, support and administrative work that has historically provided a path into the formal economy.
India would therefore need a more targeted approach than a blanket ban on automation. High-value human accountability could be protected in healthcare, education, public administration and legal decisions, while businesses continue using AI for research, translation, documentation and routine processing. Temporary transition support may be more practical than permanently reserving large occupational categories.
The larger lesson is that AI policy cannot stop at model safety or data protection. It must also address who gains from productivity, who carries the cost of displacement, and which decisions society considers too consequential to delegate entirely to machines.
A proposal, not a prediction
Human Reserved is valuable because it reframes automation as a choice rather than an unstoppable natural process. Yet it remains an early proposal with no agreed job list, enforcement mechanism, tax design or international framework.
The strongest version of the idea is unlikely to be a wall around entire professions. It is more likely to become a set of specific rights and responsibilities: a right to human review, a human accountable for high-stakes decisions, protected person-to-person care, and time-limited support where automation would otherwise impose sudden social costs.
AI may eventually prove less destructive to employment than Gates fears. But the policy question survives either way: if a machine can do a job, does that automatically mean it should? Human Reserved begins with the argument that the answer, in some cases, should be no.
