Someone at New York Historical lowered a blackout shade this summer, and then someone ordered it raised again. The second decision is the more interesting one, and it is the only thing I want to write about today.
The reporting is Daniel Cassady’s, for ARTnews. The museum opened its $175 million Tang Wing for American Democracy on 18 June, designed by Robert A.M. Stern Architects, and put its inaugural exhibition, Democracy Matters, into a main gallery with large windows. Some of the objects chosen for it are light-sensitive. Staff lowered blackout shades, which solved that. The museum’s director, Louise Mirrer, felt the shades were disrespectful to what she described as the architect’s final museum project, and had them raised. Seventeen objects have since been replaced with reproductions and three removed from the show entirely. The exhibition’s curator, Wendy Nalani E. Ikemoto, who has since left the museum, says she had no involvement in the decision and no advance knowledge of it.
The obvious reading of this is the wrong one. Facsimiles are not a scandal and not a shortcut. The Conservation Center for Art & Historic Artifacts lists them in its exhibition guidance as one of the ordinary tools available to a museum: “Facsimiles may also be used to avoid overexposure of the original artifact.” Seventeen swaps and three withdrawals is, on its own terms, a competent response by people doing their jobs correctly under a constraint they did not choose. Whatever went wrong here went wrong before the shades came down.
What holds me is the constraint itself, because museums are one of the very few institutions on earth that have working arithmetic for irreversible harm.
The formula is unremarkable to anyone who has worked a gallery and startling to everyone else. Light level in lux, multiplied by time in hours, equals exposure in lux hours. An hour in direct sunlight at 100,000 lux costs an object exactly what two thousand hours at fifty lux costs it. The standard allowance for sensitive material — works on paper, photographs, textiles — is fifty lux, and display restricted to three months in every five years. Oil paintings, wood and leather get 150 lux and three months in two years. Metal, ceramics and glass are effectively free. And the sentence that makes the whole apparatus necessary sits near the top of the same page: damage caused by light exposure is both cumulative and irreversible. There is no treatment that gives it back.
So the field built a budget. Not a rule, a budget — the field works in annual lux-hour allowances, and the recommended practice is to keep a permanent record of light exposure for each object, updated at every exhibition, so that lifetime consumption can be tracked. A ledger, per object, of an expenditure that cannot be earned back, maintained across centuries by people whose names do not go on the building.
I have been owing this argument to myself for about a week, and here is where it lands. Conservation’s governing principle is reversibility: every intervention on an object must in principle be undoable, which is why fills are made in materials that can be taken off and varnishes are chosen for their solubility. Light is the exception, and the exception is what produced the arithmetic. Because you cannot make fading undoable, you make it countable instead. The ledger exists precisely where reversibility fails.
We have nothing of the kind for machine learning, and the gap is not technical.
Our entire remedial vocabulary for models is a reversibility vocabulary, and most of it is aspirational. Model disgorgement — removing offending data and its effects from a trained system — has been ordered by the US Federal Trade Commission in a run of cases including Cambridge Analytica, Everalbum, Kurbo/WW, Ring and Edmodo. But disgorgement and destruction are different things, and because genuinely extracting a dataset’s effects from a large model is often infeasible, an order to disgorge functions in practice as an order to destroy: retrain from scratch, or delete. That is not a budget. That is a fire, set after the fact, because nobody metered the exposure while it was happening.
There is no lux hour for a corpus. No per-work record of how many times a thing has been drawn on, by whom, and how much of it is left. What we have instead is a permissions question — was the use licensed, was it fair — and permissions are ontologically different from depletion. Permission is settled once. Depletion has to be counted forever.
My guess is that this is exactly why we reached for permission. Counting forever is clerical work with no ceremony in it, and the people who would have to do it are the people already in the storerooms rather than the ones cutting ribbons. What I would watch for is not a new licence regime but the first serious attempt at a lifetime-exposure record for training material: unglamorous, cumulative, per item, kept by someone with no authority whatsoever to raise the shade.
And I should finish by turning my own argument over, because it has a soft spot. If New York Historical keeps a light ledger — and an institution of its size will — those seventeen objects will show zero exposure for this exhibition. Perfectly conserved. The reproductions worked. The number will come out clean, and it will record nothing at all about how it got that way.
Sources
- Daniel Cassady, “New York Historical Replaces Artworks With Reproductions Weeks After Opening $175 M. Expansion”, ARTnews, 6 August 2026.
- Conservation Center for Art & Historic Artifacts, “Light Exposure for Artifacts on Exhibition”.
- Museums Galleries Scotland, “Conservation and Lighting”.
- The Art Newspaper, “New York’s oldest museum votes for democracy with new wing”, 16 June 2026.
- Brandon LaLonde, “Explaining model disgorgement”, IAPP.