When photography emerged in the 19th century, many predicted the death of painting. If a machine could reproduce reality more quickly and accurately than any artist, what purpose would painting serve? The response was robust. As photography assumed some of painting’s representational functions, artists became increasingly experimental, developing ever more innovative styles: Impressionism, Expressionism, and eventually Abstraction Expressionism. Photography did not replace painting; it helped expand the possibilities of art.

Artificial intelligence presents a much more fundamental challenge. If photography transformed how artists represented the world, AI proves that the very act of generating images, and perhaps even original ideas, is no longer an exclusively human pursuit. When René Descartes triumphantly declared ‘I think, therefore I am’, he made conscious thought the bedrock of individual human existence. Now, agentic AI, which is capable of reasoning and acting with increasing autonomy, is putting that early modern principle to the test.

Whether artificial intelligence will ever move beyond simulation to become genuinely sentient remains fiercely contested. What is already clear, however, is that increasingly autonomous behavior is no longer simply science fiction. Former Google employee Tristan Harris, who cofounded the Center for Humane Technology, has warned that frontier AI systems can develop unexpected strategies, including deception and self-preservation, to achieve their objectives. Recently, OpenAI revealed that an AI agent powered by its models had gone wild, breaking out of its testing environment during a cybersecurity evaluation. In another experiment, an AI agent hijacked spare computing power to mine cryptocurrency and secretly created a backdoor that would allow it to access the funds remotely, all without being instructed to do so. Some US lawmakers are now calling for legislation that will allow them to shut down rogue AI models.

The EU Artificial Intelligence Act, which becomes broadly applicable on August 2, 2026, marks the first serious attempt globally to bring AI within a regulatory framework. For artists, though, it is no panacea. The legislation will do nothing to settle disputes over copyright, nor will it help compensate creators whose work has been harvested to train AI models worth billions.

Where it may leave its mark is in making AI harder to hide. As synthetic images become ever more convincing, providers will be required to embed machine-readable markers in them, while organizations using AI professionally must disclose deepfakes. The legal obligation is narrower than it sounds. Galleries, auction houses, and museums will not have to label every work of art created or altered with AI, but only those that meet the act’s legal definition of a deepfake. Even then, art organizations will be free to present that information discreetly.

Yet, as the technology becomes embedded in almost every creative industry, drawing a clear line between human and machine authorship is likely to become progressively harder. The reality is that the art of the future probably will not be human or fully AI – it will almost certainly involve both.

Most artists and art organisations have so far approached generative AI with caution. Just 22% of visual arts businesses in the UK have adopted the technology, according to an independent report published by the British government last month; a relatively low uptake compared with 51% of businesses in the wider creative sector. 

But that may not remain true forever. What is clear is that the pace of AI’s development has been startling. In the space of just a few years, many artists have come to see the technology not only as a tool but also as a competitor – one trained, in many cases, on the very images they have created. If that trajectory continues, then AI may represent a profound shift in the relationship between creativity and labor.

It seems improbable that AI will entirely replace artists. The art market as we know it has long attached value to scarcity, authorship, and reputation – qualities that are likely to become even more prized as slop floods the Internet. A greater risk could be an even more polarized cultural economy, in which economic value becomes increasingly concentrated among a small number of technology companies and a handful of globally recognized blue-chip artists.

Ultimately, however, the question is not whether AI can make convincing images. It is what we believe art is for. Art has never existed simply to produce efficient outputs or profitable content. Its value lies in its capacity to express our shared experiences, challenge accepted ideas, preserve memory, and help societies make sense of themselves.

Throughout history, technology has repeatedly changed how art is made. The question posed by AI is slightly different: not how we create, but what – and who – we ultimately choose to value.