Is an unknown artist worth more than Picasso? The AI ​​says yes

What’s worth more: a Picasso or a painting by a street artist no one has heard of? According to the AI ​​model built by an Art Economics professor, the answer is the second.

The work of an unknown artist worth more than a Picasso. This surprising statement is the result of an experiment carried out by Magnus Reschprofessor of Economics of Art at Yale University, with the help of a data scientist and an expert in Artificial Intelligence from Silicon Valley.

Resch and his colleagues’ goal was to understand whether artificial intelligence could bring more transparencyand perhaps greater justice, by art marketexplains the American professor in article no.

The topic has real urgency, writes Resch.

The art world is in recession for 15 yearsgalleries are closing, young collectors are retreating, and artists trying to make it in the big market centers live on the edge of poverty.

The market is opaque and elitist. In contemporary art, more than 50% of the value at auction comes from just twenty artists. The attention generated by media exhibitions and record prices is reserved for a handful of artists and galleries — under the pretext that your art is simply “better”.

But is it really?

To find out, Resch and colleagues built an AI model capable of decode how artistic value is determined in the art market.

The team wanted to test whether or not it is possible evaluate visual quality in a way that was independent of context — such as gender, origin, training, representation by galleries, influence of collectors, price history or museum exhibitions.

At the heart of the project was an LMM (a Large Multimodal Model, like ChatGPT) designed to analyze not only the visual characteristics and the content of works of art, but also metadata such as technique, format and date of creation.

The starting point: a clean and standardized dataset with millions of images, including price information, which included masterpieces from major museums, from Mona Lisa to recent works of names like Rashid Johnson, as well as the most expensive works ever sold at auction.

Based on this dataset, the team trained what they called the “Fine Art Large Vision Model” (LVM) that could predict auction prices from what I could “see”.

Market price has become the pragmatic indicator of experience—one of the few widely available, quantifiable value indicators in the world of art, although heavily distorted by trends, access, speculation and power.

The first results were promising: in more than 50% of cases, LVM predictions, based only on visual data, got surprisingly close of real prices. But it quickly became evident that more reliable forecasts demanded additional metadata such as the artist nameorigin or representation by a gallery.

After months of training with millions of images, the conclusion was undeniable: the model could not realistically estimate the price of a work of art based on the image alone.

In a striking example, A.I. valued a Picasso at less than 1,000 dollarswhile at the same time assigning a greater value to a million dollars to a work by a unknown street artist which Resch had photographed in New York and uploaded to the system.

This revealed two things: Firstly, the AI ​​considered that the street artist’s work had greater visual quality than Picassochallenging market logic in its essence. Secondly, the model failed to achieve results viable in market terms. Technically impressive, yes — but scientifically and commercially useless.

Only when they were added artist names and gallery links is that the model predictions began to correspond to thes actual auction results.

After extensive testing and optimization, Resch and colleagues were faced with a uncomfortable truth: the problem wasn’t just AI. It was the training data itself — a reflection of a market distorted by social biases and economic.

Unlike object detection or medical imagingvisual quality in art cannot be objectively quantified. And as our data set was largely composed of works already “validated” by the market, we ended up reinforce circle patterns.

The results were revealing — and frustratingwrites Resch: the market does not reward the work of art itself, it rewards the name. It’s the galleries that define what matters.

What is the lesson for artists? That success depends less on the trait than on the contact network. Resch had already reached this conclusion years ago, in one of which he had co-authored, published in 2018 in the journal Science.

Therefore, what continues to surprise is how rarely art schools teach the business realities of an artist’s lifeand the frequency with which artists cling to belief that your art alone will make your careersays Resch.

As for Artificial Intelligence, artists probably shouldn’t fear it. No machine can replace a visit to the ateliera real conversation, or the thrill of connecting with a work of art in person.

Art is still a human business, based on trust, intimacy, and emotion, and AI isn’t replacing artists — it’s replace system guardians.

And the collectors, in turn, must trust their instincteven if it means buying a work they come across in a small gallery or even on the street. For what it’s worth, an AI is likely to agree with the choice.