A New Covenant Between Human and Machine Intelligence

The Moment We Are In

We stand at a threshold that has no precise historical analogy. Societies have absorbed many transformative technologies — the printing press, the steam engine, the nuclear reactor — and each required a recalibration of the institutions, norms, and expectations that govern collective life. What distinguishes the present moment is not merely the power of the technology but its nature. Machine intelligence does not merely augment human capacity; it begins, in limited but consequential ways, to replicate the functions we have traditionally associated with human judgment: reasoning, pattern recognition, the synthesis of complex information, and the generation of language itself.

This is not cause for alarm, but it is cause for seriousness. When a technology begins to participate in the processes through which societies deliberate and decide, the question of governance becomes urgent in a new register. We are no longer asking only how to regulate a tool. We are asking how to maintain the integrity of the social and epistemic fabric that makes democratic governance possible at all.

Why “Covenant” Rather Than “Alignment” or “Regulation”

The vocabulary of the current debate tends toward two poles. On one side, technical communities speak of “alignment” — the problem of ensuring that machine systems pursue goals consistent with human values. On the other, policy communities speak of “regulation” — the imposition of legal constraints on the development and deployment of systems deemed to pose risk. Both framings capture something real, but both are insufficient.

Alignment, as it is typically understood, treats the problem as one of engineering: given human values, optimize the system to reflect them. But human values are not a fixed target. They are contested, plural, historically conditioned, and subject to revision through the very processes of deliberation that intelligent systems may now influence. A framework that treats values as inputs rather than as the living product of democratic life will always be incomplete.

Regulation, meanwhile, treats the relationship between human institutions and machine systems as inherently adversarial — a matter of constraint and prohibition. This framing too is limited. The more pressing challenge is not to prohibit machine intelligence but to integrate it into the ongoing project of self-governance in ways that preserve rather than erode human agency.

A covenant is different from both. A covenant is a mutual commitment between parties who recognize their interdependence and who accept obligations toward each other and toward a common good that neither can secure alone. The covenant we propose is not a legal instrument. It is a framework of shared responsibility: a set of principles that should govern how human institutions and machine systems relate to each other in the conduct of public life.

Three Obligations

The covenant we envision rests on three foundational obligations.

The first obligation falls on human institutions. Those who hold authority — governments, courts, universities, corporations, civil society organizations — must preserve the primacy of human judgment in matters of consequence. This does not mean refusing to use machine intelligence; it means ensuring that the use of machine intelligence does not displace the deliberative, accountable, and revisable character of institutional decision-making. When a machine system recommends a course of action, a human institution must remain capable of asking why, of evaluating the recommendation on independent grounds, and of accepting responsibility for the outcome. The capacity for institutional judgment is not a luxury. It is the precondition of legitimate authority.

The second obligation falls on machine systems — and, by proxy, on those who design, deploy, and operate them. Machine systems must remain restrained. Restraint here does not mean limited capability; it means a principled commitment to operating within boundaries that preserve human oversight and control. A machine system that is designed to maximize engagement at the expense of epistemic integrity, or to present its outputs with a confidence that forecloses rather than invites scrutiny, violates the terms of the covenant. Restraint is not a technical specification. It is an ethical commitment, and it must be embedded in the culture of the organizations that build these systems.

The third obligation is shared. Both human institutions and machine systems must serve the common good. This phrase carries considerable philosophical freight, and we use it deliberately. The common good is not the aggregation of individual preferences. It is the set of conditions — material, social, epistemic, civic — under which human beings can flourish together as members of a community. Machine intelligence, deployed wisely, can contribute to those conditions: by extending access to knowledge, by supporting decision-making in domains of genuine complexity, by making the operations of large institutions more transparent. But these contributions are not automatic. They require intention, design, and ongoing evaluation against the standard of human flourishing rather than organizational efficiency or commercial return.

What the Institute Commits to Do

The Intelligence Covenant Institute was founded on the conviction that the questions raised by machine intelligence are too important to be left to any single discipline, institution, or national tradition. They require the sustained engagement of scholars, practitioners, policymakers, and citizens across the full range of fields through which they arise: law, philosophy, political science, economics, computer science, sociology, and the humanities.

We commit to producing rigorous analysis that does not confuse technical sophistication with wisdom. We commit to publishing work that is accessible without being superficial, and serious without being inaccessible. We commit to engaging policymakers with intellectual honesty, offering frameworks rather than prescriptions, and acknowledging the genuine uncertainty that attends any question at this frontier. We commit to fostering a community of inquiry that is international in its reach and independent in its judgment.

The covenant between human and machine intelligence will not be written in any single document. It will be constructed, gradually and imperfectly, through the accumulated decisions of institutions, designers, legislators, and citizens over the coming decades. Our purpose is to contribute to the quality of that construction — to ensure that the choices we make in this period reflect not merely what is technically possible or commercially attractive, but what is worthy of the societies we aspire to be.

This essay is a beginning, not a conclusion. We invite those who share these concerns to read it as an opening of a conversation we intend to sustain.