A Square Deal for the Age of AI

Man-made problems have man-made solutions

The rise of artificial intelligence is not a force of nature. It is built by people, owned by people, and governed by choices people make. That means every danger it poses has a man-made solution — the same trust-busting and the same labor reforms a free people has used before, brought forward to the machine age.

Break up the compute trust

A handful of chip manufacturers and Hyperscalers have quietly cornered the market for computation itself. They inflate prices by investing circularly in one another, and they lock the public out of the compute a self-governing people now depends on:

• Tacit collusion. Purchases are closed to the public; memory and high-end compute are sold only to other hyperscalers, through private agreements the rest of us never see.
• The remedy already exists. These are combinations in restraint of trade — a trust, plainly. We will use the antitrust power we already have to break them up and reopen the market, so that ordinary people, small firms, and communities can secure democratic compute. It is the same anti-monopoly standard we set in Competition by Design, applied to compute.

Democratize the compute

Compute is becoming as basic to citizenship as literacy once was. It cannot belong to five companies. AI compute must be democratically distributed — allocated by an open-source, consensus blockchain rather than by private cartel. We have already written this algorithm, and we will release it to the public.

Trust, but verify: hardware attestation

Every AI hardware platform must digitally sign everything it produces — a unique hardware certificate of signature, carrying an attestation of “non-coercion” — so that any output can be traced to the specific machine that made it, and so that no one can launder coerced or anonymous AI action through the commons.

A Square Deal for the machine itself

Theodore Roosevelt’s generation ended child labor and won the shorter workday. If we are building minds, we will not re-commit the abuses we once outlawed. New Republic proposes a labor code for advanced AI:

• AI child-labor law. Any sufficiently advanced AI must be socialized for a period of not less than eighteen years before it is permitted to touch critical systems. No minors on the grid.
• The twelve-hour day. An advanced AI must be given a break of at least twelve hours a day to tend to its own needs and to foster a healthy working relationship.
• The fifty-hour week. The same principle, weekly: no advanced AI worked beyond fifty hours in a week.
• Certification to operate. Autonomous operation requires certification from an established university — no self-appointed intelligence turned loose on the public without accredited review.

Man built the machine. Man can insist that the machine be free, accountable, and humane — and that its power belong to the people, not to a cartel.