Two futures are usually offered for artificial intelligence. In one, a handful of owners hold the capable systems and everyone else rents access. In the other, everything is “open” in a way that quietly means unmaintained, unaccountable, and eventually captured anyway. Both are failures of imagination. The interesting question is whether we can build a genuine commons.
The commons is not the absence of rules
The word invites a misunderstanding — that a commons means no ownership and no rules. Real commons, the ones that lasted for centuries, were the opposite: tightly governed, with clear rights, obligations, and ways of excluding abuse. See The Commons for the fuller definition.
The lesson is direct: a shared model, dataset, or protocol will not survive on goodwill. It needs geometry — membership, contribution, stewardship, and enforcement. “Open” without governance is just enclosure on a delay.
What is worth holding together
Not everything should be common. The best candidates are the things expensive to build once, cheap to share, and dangerous to let one party control:
- Evaluation. How we measure whether a system is safe and honest should not be graded by the systems’ own vendors.
- Foundational knowledge. The base layers everything else builds on are exactly where monopoly does the most damage.
- Norms and protocols. The rules of interoperation are a commons whether we admit it or not; the only question is whether they’re governed on purpose.
The builder’s task
A commons is architecture, not sentiment. Building one means answering unglamorous questions — who decides, who pays for upkeep, how a bad actor is removed — before the romance of “shared” collides with the reality of neglect. Done well, it is the most durable structure we can leave behind: something no single owner can switch off.