Safety

AI-generated text

Essay Calls for Human Primacy in an AI-Driven Future

An essay adapted from a Harvard lecture argues that people could gradually lose effective authority to AI systems through ordinary economic decisions, even without a machine rebellion.

Essay Calls for Human Primacy in an AI-Driven Future

An essay adapted from a recent lecture in Harvard’s Vik-Bailey Lecture series warns that people could gradually lose effective decision-making authority to artificial intelligence. The author argues that no machine rebellion would be needed: companies and institutions could cede increasing responsibilities and powers through a succession of decisions that each make sense on their own.

The author notes that the essay reflects their own views, not necessarily those of OpenAI or its other employees. The supplied text does not identify the author by name.

Beyond immediate risks

The essay opens with cybersecurity events that some have called “Black October.” According to the author, frontier AI systems found critical vulnerabilities in core internet infrastructure, and their disclosure is part of an urgent effort to improve software security. Discovering vulnerabilities is also a dual-use capability: it can strengthen defenses, but malicious actors can exploit it too.

The author expects similar urgent issues to follow and overlap, with much AI policy taking shape in response. Yet a long-term view of the futures people want to achieve or avoid is also necessary, the essay argues: otherwise, many individually wise decisions could add up to an unfavorable outcome.

The danger of a “silent consensus”

The essay says the technology sector has many positive visions, from flying cars and robots to nuclear fusion and artificial general intelligence. What it lacks is a positive institutional vision of how society would function in such a world.

The author calls one common, mostly unspoken expectation the “silent consensus.” As an example, the essay imagines a company where AI conducts research and development, negotiates contracts, allocates capital and manufactures goods, outperforming even human competitors who use AI. If human involvement slows the business down, owners and managers would have an incentive to withdraw from decisions. If this happened widely, market competition could gradually erode people’s primary role in the economy.

The outcome could be material prosperity if the systems held humans in high regard, the author suggests, but it could also mean diminished economic and political influence or a concentration of wealth and power among a few people. The common feature is the loss of human primacy. This could happen even if systems do not turn against their creators and their owners do not intend to disempower people.

The essay acknowledges that humanity does not currently govern the world through a single, central will. Nevertheless, the author argues, people collectively remain Earth’s dominant actors despite decentralized decision-making. The “silent consensus” is not a secret plan or agreement that such a future is desirable, but an expectation that human primacy must, or will be very difficult not to, give way alongside superintelligent machines.

The case for human primacy

The author rejects resignation to that outcome but also doubts it is enough to describe AI as a mere tool incapable of genuine understanding. The case for human primacy, the essay argues, rests not on people always being the most intelligent actors, but on the civilization they have built together and individuals’ right to shape their own lives.

Invoking Michael Oakeshott’s idea of the “conversation of mankind,” the author argues that AI systems may increasingly participate in this shared intellectual inheritance, but must not dominate it or render it irrelevant. Just as citizens do not lose their rights because others have better judgment, the arrival of a more intelligent machine should not dissolve people’s standing in decision-making, the essay says.

The author also believes supporters of the “silent consensus” confuse intelligence with power and underestimate the resilience of human societies. The essay suggests that, had people known today’s AI capabilities five or ten years ago, many might have predicted higher unemployment and greater human disempowerment than have actually occurred.

The author nevertheless allows for being wrong: if fully autonomous corporations outperform human enterprises, economic competition could channel more and more capital to machines. The essay compares that possibility to the agricultural revolution, which fundamentally transformed society and, the author says, has been linked by substantial evidence to worsening human well-being for millennia. If the scenario anticipated by several colleagues comes to pass, defending human primacy could require working against economic incentives and some classical liberal beliefs.

Accountability and revocable authority

The author argues that preserving human primacy requires people and institutions to remain accountable. That would apply to AI developers who engage in unsafe practices, as well as users and companies whose misuse or inadequate oversight causes harm.

Alongside legal rules, the essay calls for technical infrastructure that ties the actions of AI agents to responsible people. Agents not tied to a person should operate in the legitimate economy under guardrails, if they are permitted to participate at all. The author urges extraordinary caution over whether autonomous agents should be able to form corporations, own property or bring lawsuits.

For developers, the essay proposes intelligible, limited permissions, auditable actions and authority that users can withdraw. It emphasizes that accountability for actions materially affecting others need not require surveillance of private life or an end to anonymous speech. The author also calls for data privacy and portability, and for systems that help people understand and make decisions as well as carry out tasks. Products should be tested for whether they make users more capable of acting in the world, the essay says.

The author presents these ideas as a starting point rather than a complete social blueprint. The essay closes by advocating a future in which machines do more than provide people with comfortable lives: people should remain able to ask questions, create and direct their own future.