Here’s How AI Is Already Disrupting the Art Business

Lately, it feels impossible to flip through a culture section without reading about the algorithmic invasion dismantling creative industries. We have watched artificial intelligence rewrite the rules of music, screenwriting, and animation—so much so that TIME published an essay detailing how human artists might soon be forced to make their work deliberately “worse”—less polished and more flawed—just to prove that a machine didn’t make it. For a long time, those of us in the art market looked on with comfortable detachment, believing that our industry was simply too emotional and too human for an algorithm to ever map out. That turned out to be a delusion. Today, advanced data infrastructure is flooding our traditionally closed rooms with light. It brings up an unsettling question: what happens when AI stops old-school secrets from working?

Because let’s be honest: our industry has always run on a strict diet of insider information. Fortunes are made here simply because one person gets a phone call about a major gallery signing or a private sale price five minutes before the public finds out. We thrive on this asymmetry. But when a machine can instantly cross-reference millions of data points to evaluate an object, that leverage starts to collapse. Any time a business becomes transparent, the people who build their careers in the dark are bound to lose out (remember what happened to Old Master dealers when auction databases went online?). 

I wanted to discuss this conundrum with Sam Glatman, the co-founder and CEO of Artsignal. He doesn’t come from the typical Silicon Valley mold; he originally studied classical music, pivoted to econometrics, and built a startup in Myanmar that landed him on Forbes’s 30 Under 30 list. When he returned to the U.K., he walked into the gallery world armed with a background in machine learning and self-admitted overconfidence—the belief that he could actually solve the art market’s eccentricities. His platform, which recently launched with serious backing from Christie’s Ventures, essentially gives subscribers the kind of data that used to require a team of private advisors.

In our conversation, Sam and I get into the reality of what happens when this information becomes accessible. We talk about who loses their edge when the playing field levels out, why algorithms still totally strike out when it comes to taste, and why a more transparent future might actually force us to look up from our screens and lean back into the human connections that make this whole world spin.

Your background spans international business, and you come from a family of art collectors, which gives you a unique lens on these ecosystems. How did you develop the idea for Artsignal?

Growing up around art collecting, I always had an interest in the art scene. While working at a tech venture builder focusing heavily on AI, I spent my free time being mentored by art world veteran Keir McGuinness, who taught me how to identify underappreciated art. With a bit of hubris, I thought I’d cracked the art world and realized that combining LLMs and art evaluation put me into a unique position. I designed a spreadsheet methodology to assess artists and artworks primed for reevaluation, teamed up with my co-founder [Ivan Bestvina], who is a brilliant engineer that collects contemporary Chinese ceramics, and began building Artsignal.

To help people understand what Artsignal does, can you lay out two or three different possible clients and their use cases?

For the financial use case, art-backed lending is a fast-growth area. We build the valuation evidence and logic at scale across thousands of objects simultaneously—art, jewelry, watches, and luxury collectibles—so a lender can get a portfolio assessment within minutes. All the logic is auditable, and they can go a step further to assess collection risks, like high concentration in a particular artist or brand in a default sale scenario. The other use case is for specialists within auction houses, galleries, and museums to do commercial research, look at market trends, and stress-test different pricing scenarios. Finally, we work with institutional collectors who use the platform to build custom indexes of any market segment they want, whether that’s David Hockney works on paper from a specific three-year period or Patek Philippe watches, allowing anyone participating in the market to build an understanding.

Looking at an artwork requires us to balance historical data with highly subjective nuances of an individual object. Anyone in the art world knows that two pieces by the exact same artist, created in the same year, can see a drastic difference in pricing. When building a technical framework for a market that has historically been quite gatekept, your formulas have to confront these idiosyncrasies. What were some of the most challenging decisions you had to make in determining how to weigh or value certain information when building an algorithm to determine an artwork’s value?

You’ve hit the nail on the head. We are dealing with unique objects that have condition issues and provenance questions. Take a pair of [Claude] Lalanne Ginkgo chairs: from an important design collection, they might sell for $1 million, but from a normal collection, $500,000. That’s the level of nuance our system is designed to assess using visual search and recognition across a database of about 15 million historic sales. The system first filters to find logical comparables, and then it analyzes the image data to rank the most visually similar objects first. If those comparables are old and the market has moved, we rely on our custom market indexes to adjust that value over time.

How have you seen people’s attitudes or openness to AI change since you started the business? Do you still encounter a lot of resistance?

When people talk about AI for creating art, that’s controversial, though artists like Refik Anadol are doing really cool stuff using AI as part of the creative process. For what we do, there was initial fear and skepticism; people love posting on LinkedIn saying, “I used AI to value a painting and it didn’t realize it was by my grandmother.” The worst things are black box AI systems where a number shoots out and you can’t audit the reasoning. With Artsignal, you can edit or delete comparables, adjust condition information, and add assumptions. Some clients use it as an independent assessment of value, while others get intimately involved by adding images and adjusting variables.

There is a real anxiety around whether artificial intelligence will sanitize the creative industries, and when we dissolve the traditional boundaries between human curation and machine learning, it forces us to navigate a brand-new relationship with how knowledge is produced. I’m interested in better understanding how AI can change the art business more broadly. What in the art business is AI really good for, and good at?

I think AI opens up the art market to many more people. Using a tool like Artsignal, any collector can understand what a work is worth, the factors behind it, and its market history. While it isn’t as detailed as a top art advisor—and top collectors will always want advisors for human nuance and access—most collectors can’t afford one anyway. Currently, the size of the art market is constrained because human expertise simply doesn’t scale. By using AI to scale that expertise, we can open the market up to far more people and objects. For institutions like galleries and auction houses, they simply don’t have the time to dedicate to mid-value objects, keeping in mind that most objects in the market sell below $50,000. Our system enables them to offer detailed data on every single object, whether it’s worth $1,000 or a million.

What is AI not good at… yet?

For one, AI will always struggle with taste. It’s incredibly hard to train a model on good taste because it’s just not clearly defined—it’s why AI-generated emails still look so obvious. Another big issue is consistency. If you ask a model the same questions twice, the logic can look completely different each time because controlling how it weighs data just through prompting is tough.

What do you think it will never be good at?

AI will never replace human relationships, human interaction, and the actual passion for collecting. The art world relies on that in-person discovery, that “scratch and sniff” element of going to art fairs to appreciate the work. I look at it optimistically: By handling time-consuming data processing or building pitch decks in five minutes instead of two days, AI frees us up to have three times as many client meetings and allows us to focus on the human aspect of the business.

In a blue-chip scenario where timing and rarity are everything, a tool’s reliability becomes the ultimate benchmark. If I have a client looking for a red Rothko, how could Artsignal, or AI in general, help me better advise them?

If you have a specific piece in mind, you just drop the image and whatever information you have into the system. It instantly finds the most similar things that have sold before and breaks down the nuances of size, date, condition, provenance, and even auction house guarantees. This builds a benchmark list to guide your valuation, giving you a clear view of what the red Rothko is worth and what factors could push the price up or down. You can also zoom out to look at the broader Rothko market and track how red ones perform against other colors to get a really granular understanding of their value. Now, what we don’t do yet is actually source the red Rothko for you—though it would be great if we could.

What use cases does AI have for tracking provenance and determining authenticity? It seems like people are divided on how reliable its authentication really is.

Right now, you see companies analyzing paintings and saying they’re 80 percent confident it’s a Rembrandt, but it becomes a bit of a losing battle because people instantly question the credibility whether the AI is right or wrong. My view is that AI isn’t a definitive stamp of authenticity, but rather a tool to help you cross-reference a massive amount of data points to reach a conclusion. 

I agree with you on that, but I am curious as to the long-term impact of democratizing data. The art market has historically functioned as a democratic medium in theory, but as an opaque, exclusive domain in practice. The entire ecosystem has thrived on information asymmetry—where knowing a singular detail or identifying a market trend before anyone else is how fortunes are made. Any time the business becomes more transparent, some people who have been profiting off of its opacity lose out. Who stands to gain and who stands to lose from an art business that diminishes information asymmetry through AI?

It’s a really interesting question. The art market absolutely runs on that kind of insider access—like knowing Gagosian is about to take on an artist before it’s announced. Because it’s less regulated than financial markets, a few people make outsize gains while the rest of the market loses. When new collectors come in and get exploited by information asymmetry, overpaying for an artist only to watch the value drop 90 percent, they never buy again. Greater transparency will minimize that. Yes, some insiders will lose their advantage, but we’re interested in a much more expansive market. There’s a massive amount of unmet demand from people who just want to buy art with real history and meaning rather than overpriced decor. In five years, the art world will look very different.