The Paperwork Was the Point

This is good. Genuinely, unambiguously good. Now the part nobody wants to sit with.

Somewhere north of 100,000 looted objects have never gone home. Not because they vanished. Many of them hang on walls with placards, in buildings with donor names on them, in cities that have apologized in general terms. They didn't go home because going home required provenance research, and provenance research was hard, slow, expensive, and ambiguous.

All of which is true. All of which was also extremely convenient.

Here's the machinery. An institution that wants to refuse something publicly indefensible does not refuse it. Refusing creates a decision, and a decision creates a person who made it. What it does instead is cite process. The archives are scattered. The records are in four languages. The chain of title has gaps. We take these claims very seriously and are reviewing them carefully. That sentence has been load-bearing since 1945.

Notice what the difficulty was actually doing. It wasn't preventing restitution. It was preventing anyone from having to say out loud that they didn't want to do it. The friction was a service. It let a museum board be sympathetic and immobile at the same time, which is the most comfortable position available to anyone who benefits from a wrong they didn't personally commit.

You have your own version of this. The conversation you haven't had because the timing's never right. The thing you'd deal with if you had the bandwidth. You are not lying, exactly. The bandwidth really is short. But you've also never tested it, because testing it would tell you something about what you want, and the not-knowing is doing work. The mechanism is the same at both scales. The stakes are not — a delayed phone call is not a seventy-year refusal, and nothing here collapses the two. It's the machinery they share, and the machinery is what's about to break.

So the interesting thing about the provenance bot isn't the research. It's the removal of the excuse.

If the archive work drops from years to weeks, the cost stops being an answer. What's left when the cost is gone is the actual position, and we are about to find out what that is. Watch for the reasons that move up to take its place. They won't be new. They'll just be promoted. Legal finality. Statutes of repose. The good-faith purchaser. The argument that the object belongs where the public can see it, made by institutions who did not ask the public anything when they acquired it. The 1998 Washington Principles committed forty-four nations to just and fair solutions and were nonbinding on purpose. The 2016 HEAR Act reopened American courthouse doors and museums responded, in several documented cases, by arguing technical defenses the Act was written to close. Those defenses have been running the whole time, in parallel, right alongside the paperwork.

So what dies isn't the refusal. It's the sympathetic refusal. Research cost was the softest alibi on the shelf — the only one you could offer while still looking like you were trying. Take it away and what remains is a set of positions that are legally strong and morally naked, which is a much harder place to stand in public with your name attached.

There's something worth naming in what the tool is made of, and in what it's fed. A model like this runs on the pooled human record — everyone's archives, everyone's cataloging, everyone's tedious clerical labor, aggregated. And the particular record it reads in Paris exists because the people doing the looting were meticulous about it. They kept lists. The theft documented itself, and eighty years on the documentation is the instrument of return. That's not redemption, and it isn't irony either. Efficiency has no direction of its own; someone chooses the direction. What it shows is that the commons can be turned toward return instead of accumulation when a person decides to aim it there.

Which is where the real constraint shows up. The tool reads the Jeu de Paume because the Jeu de Paume's records are open. It does not read the archives of institutions that would be sued if it did — and those institutions are both the defendant and the archive, holding the object and the evidence about the object in the same building. The scatter was never purely an accident of war. It's a distribution of custody. So the excuse doesn't vanish so much as migrate: from "the research is too hard" to "our records aren't digitized, aren't complete, aren't public." Same alibi, better hygiene, and this version is controlled entirely by the party with the most to lose. An AI that reads archives is only ever as good as its access, and access is granted by the defendant. The pooled record can be pointed at return — but only across the portion someone has agreed to pool. Open access is the restitution question now.

The bot won't give anything back. It has no hands and no standing. What it does is strip away the last reason anyone could offer sympathetically, and hand the question back to the people who were always going to have to answer it.

They can still say no. They just have to say it now.