Christie's Augmented Intelligence, the first auction devoted to work made with artificial intelligence, closed on 5 March at $728,784 against a pre-sale expectation of about $600,000. Thirty-four lots, 82% sold.
The top lot was Refik Anadol's Machine Hallucinations – ISS Dreams – A at $277,200 on a $200,000 estimate. Holly Herndon and Mat Dryhurst's diptych Embedding Study 1 & 2 took $94,500 against $70,000 to $90,000.
Nearly 4,000 people signed an open letter asking Christie's to cancel the sale, on the grounds that the models involved were trained on copyrighted work without consent or payment.
Read the bidder data, not the total
$728,784 is a rounding error. A single lot in an evening sale beats the entire auction, and anyone presenting the total as evidence of a market is overselling.
The composition is the interesting number. 48% of bidders identified as Millennial or Gen Z, and 37% were bidding at Christie's for the first time.
An auction house's structural problem is that its client base ages and does not replace itself. A sale that brings in a third of its bidders new, at an average lot value low enough to be a first purchase, is doing recruitment. Whether the work is any good is almost beside the point commercially, and the house knows it. This is the same instinct that put digital work on the fair floor at Art Basel rather than in a side room.
The protest was about training data, and it was not answered
The objection was specific: that the models producing this work were built on copyrighted images taken without permission or compensation. That is a claim about how the work was made, not about whether it is art, and the sale proceeded without addressing it.
It will not stay unaddressed, because it is a provenance question, and provenance is the one thing this trade cannot be relaxed about. A market that will not buy an antiquity without an export licence, and that has spent two decades tightening restitution diligence, cannot indefinitely sell work whose inputs are undocumented by design.
The likely resolution is the boring one: disclosure. Catalogue notes stating which model, trained on what, licensed how. The houses will resist it, then adopt it, then market it as a standard.
Where it sits against the rest of the digital record
This is the third attempt to give digital work an auction identity, and the first two are instructive. The speculative wave of 2021 collapsed. Online art sales have since fallen to their lowest level since 2019.
What is different here is that the work is being sold as art with a medium, in a themed sale, with estimates in the tens of thousands rather than the millions. Sold quietly and cheaply to new buyers is how a category actually establishes itself. Sold loudly and expensively to speculators is how the last one did not.
What to watch
The second sale. First-of-its-kind auctions attract curiosity bidders and beat estimates almost by construction. A second one, without the novelty and without the open letter, is the one that tells you whether there are collectors here or whether there was an event.
Sale results, lot prices and bidder demographics as reported by ARTnews and Artsy. The open letter and its signatory count per ARTnews.
