The Setup: When Machines Start Making Museum Decisions
Here’s what nobody wants to admit at the dinner party: we’re watching the most consequential shift in curatorial practice since Walter Benjamin started theorizing about mechanical reproduction, and half the art world is pretending it’s not happening. MoMA’s 2025 algorithmic retrospective landed like a perfectly calibrated provocation, forcing us to ask whether institutional curation has become a process or a surrender. The exhibition drew over 340,000 visitors in its opening quarter, a number that signals hunger rather than mere curiosity. People want to see what happens when algorithms start deciding which artworks matter.
The central question isn’t whether AI can select art. Obviously it can. The question is whether it should. And more troublingly: what happens to the stories we tell about art history when we let machines narrate them?
The Seduction of Algorithmic Efficiency
Let’s start with what’s genuinely compelling about this moment. Algorithmic curation promises something museums have chased for decades: scale without sacrifice. Machine learning models can process thousands of artworks across decades of institutional records, identify overlooked connections, surface patterns humans miss. There’s real craft in the engineering. The technical sophistication required to build these systems is legitimate artistic work, even if it makes traditional curators uncomfortable to admit it.
The market has noticed. Refik Anadol’s AI-generated works commanded $4.7 million across sales this spring, a 38 percent leap from his 2023 numbers. Pace Gallery just announced a full wing dedicated to twelve artists working exclusively in machine-learning mediums. This isn’t fringe anymore. This is capital recognizing that algorithmic art-making has become a bankable medium with its own aesthetic vocabulary.
But here’s where we need to pump the brakes and ask harder questions. A 2025 survey from ArtNet revealed that 61 percent of gallery directors believe AI-assisted curation tools compromise institutional editorial integrity. That’s not a rounding error. That’s a crisis of confidence among the people actually running these institutions. And they’re right to be worried.
The Structural Problem Nobody Wants to Name
The Getty Conservation Institute published a devastating 2025 report that should terrify anyone who cares about how art history gets written. AI curation models trained on Western canon datasets systematically underrepresent Global South artists by up to 74 percent. Let that number sit for a moment. This isn’t a bug. It’s architecture. When you feed machines a diet of art historical data that already reflects Western institutional bias, you don’t fix bias through automation. You calcify it.
This is the craft question underneath all the glittering technology. Curation has always been about making arguments. A curator stands in front of a collection and says: these artworks matter because of how they speak to each other, because of what they reveal about power and vision and historical moments. That’s editorial work. That’s accountability. When you outsource that decision-making to an algorithm, you’re not gaining objectivity. You’re hiding your curatorial choices inside mathematics.
The worse part? The algorithm doesn’t know it’s making choices. It thinks it’s finding truth. And visitors, standing in the gallery, think they’re seeing what’s “objectively important” rather than what the training data decided was important. That’s not democratization. That’s laundering bias through technology.
What Craft Actually Demands
Real curatorial work happens in the friction. It’s the moment a curator looks at an artwork and thinks: nobody else would put this next to that, but here’s why it matters. It’s the willingness to make a case that might fail, that might be argued against, that opens you up to criticism. That’s human responsibility. Algorithms don’t have responsibility. They have parameters.
When you visit MoMA’s official exhibition archive, you can trace how human curators have wrestled with what deserves space in institutional memory. Some of those choices were wrong. Some were brave. All of them had names attached. We could argue with Alfred Barr about what he decided to exclude. We can’t argue with an algorithm’s training data the same way, because the algorithm doesn’t have a philosophy. It has coefficients.
The ArtNet survey and the Getty report matter because they’re asking curators to name what they actually do. Curation isn’t information retrieval. It’s interpretation. It’s a craft that requires judgment, historical knowledge, aesthetic conviction, and the willingness to be held accountable for your arguments. You can’t automate that without losing something real.
The Question We’re Actually Asking
So is MoMA’s algorithmic retrospective a celebration or a surrender? It depends on whether the institution is using algorithms as a tool within human curatorial practice or replacing human judgment with computational speed. One is integration. The other is abdication.
The money flowing toward AI art suggests we’re moving toward the latter. The market loves algorithmic art-making because it’s scalable, because it produces endless variation, because it offers the aesthetic of innovation without the friction of singular artistic vision. But museums are supposed to do something different. They’re supposed to make arguments about what matters and why.
What would responsible curatorial use of AI actually look like? It would be transparent about its limitations. It would supplement rather than replace human expertise. It would acknowledge the Western canon bias in its training data and compensate for it explicitly. It would treat algorithms as what they are: powerful tools that amplify human choices, not neutral arbiters of cultural value.
The retrospective happened. Visitors flooded the gallery. The conversation continues. What matters now is whether curators insist on maintaining editorial responsibility for what gets displayed, or whether they accept the convenient fiction that machines can decide what matters. That’s not a technical question. It’s an ethical one, and it deserves better than a dinner party debate. It deserves serious attention from the institutions we trust to shape cultural memory. What do you think museums should do differently?
