Anyone looking for visible applications of artificial intelligence in the cultural sector quickly ends up with the publicly visible. A chatbot on a website. A talking robot in an exhibition. A work of art that behaves as a conversation partner. This produces pretty headlines and a certain curiosity. But those who look a little closer see a different picture. For now, the public use of AI in the cultural field is not very much. The real advance is taking place much less visibly, in the internal systems on which institutions are now running and becoming increasingly dependent.<\/p>\n
This is relevant precisely because the debate on AI in the cultural sector in recent times has often focused on supervision, responsibility and governance. Previous articles of mine have addressed the question of who is in charge of governance when institutions deploy AI, how supervision relates to technological dependencies and how cultural organisations can guard public values in a time of rapid digital adoption. This follow-up requires a more practical question. What does the public actually see from it?<\/p>\n
For now, the answer is: not that much. Certainly not in the form of what the technology sector likes to call \u201cagentic AI\u201d, systems in which artificial agents perform tasks to a greater or lesser extent independently on behalf of or for users. In the cultural domain, such public use still seems extremely limited. It usually remains limited to a simple chatbot on the site, a search function in natural language or an accompanying digital guide that mainly makes an existing audio tour a bit smarter.<\/p>\n
That does not mean there are no examples. Just that they are scarce, experimental and often still rather tentative. In July 2025, the London Museum launched Clio 1.0, an experimental conversational search agent that helps visitors navigate through collections, stories and blogs using plain language. In itself, this is interesting precisely because the museum emphasises that the system relies on its own, trusted data rather than the open web. This limits interpretation and keeps it more controllable. It is an important detail because it shows that cultural institutions are indeed aware of the risks of free generative systems. At the same time, Clio is not yet an autonomous museum peer, but rather a neat, delineated entry point to existing knowledge. And to a lot of questions Clio has no answers, which is a pity<\/p>\n
In Tokyo, the move is slightly more ambitious. The Mori Art Museum experimented with the ARTLAS AI Companion for Roppongi Crossing 2025. That personalised guide adapts to the visitor's age, language, interests and available time and puts together a route and commentary based on that. There, the technology is already shifting a little more towards agentic behaviour. Not because the system is spectacularly autonomous, but because it actively makes choices in what it shows to a visitor and how it structures the experience. There, the classic audio tour changes from a linear tool to a modest, guiding guide.<\/p>\n
The Netherlands has similar incitements. At Discovery Museum in Kerkrade, visitors can meet Ami, a humanoid robot with remarkable communication skills, in AI: the expo. This is an appealing example on the audience side, precisely because it is visible, tangible and directly experienceable. But again, this is a conversational interface rather than a truly autonomous cultural agent performing tasks independently on behalf of the visitor. It is a demonstration of possibilities, not a new foundation for museum practice.<\/p>\n
At the other end of the spectrum is Amsterdam's Nxt Museum. There, 2025 teamed up with Prosus for a residency around intent-driven AI, systems that don't just respond to explicit commands, but try to understand a user's underlying intent and act on it. This may be closer in content to the idea of agentic AI, but here it mainly took the form of an artistic and investigative testing ground. It is less a direct public product than a laboratory in which artists, designers and programmers explore what such systems can mean culturally and aesthetically.<\/p>\n
This is precisely where the core lies for now. Where public institutions are still cautious, artists appear more willing to use AI agents as subjects, mediums or antagonists. This produces more interesting questions than the average website chatbot. The Brooklyn Rail in 2025, for example, pointed to Chiara Kristler and Marcin Ratajczyk's Agentic Fatigue Prevention Unit, a work in which the conversing agent is not only functional but also critical. The work exposes something of digital fatigue, the culture of constant affirmation and the curious intimacy that arises when machines pose as attentive companions.<\/p>\n
Botto, the decentralised autonomous artist who continues to develop through community direction, also shows where the experimental vanguard sits. With Mirror Stages at Art Basel Hong Kong in March 2026, Botto advanced further from generated image to live installation and presence in the mainstream art world. Such examples are important, not because they will become the norm in Dutch museums tomorrow, but because they make visible where the imagination of agentic systems is actually being tested now: not in the audience desk of the average museum, but in art practice itself.<\/p>\n
This creates a somewhat paradoxical picture. Public use of AI in the cultural sector still seems modest, sometimes even remarkably modest given all the social fuss. The most visible applications are often simple and supportive. A chatbot. A robot. A personal guide. The big leap to agents that plan, select, advise and act on behalf of visitors has not yet been made. Not because the technology is lacking, but because cultural institutions are naturally reluctant when interpretation, reliability and public responsibility are at stake.<\/p>\n
Meanwhile, behind the scenes, the use of AI is moving much faster. Not as a spectacular agent for the visitor, but as an invisible layer in search systems, collection registration, audience analysis, marketing automation, planning, word processing and digital infrastructure of suppliers. This is precisely where monitoring will probably have to focus most in the coming years. Because the biggest risk for cultural institutions at the moment is not that visitors will be overwhelmed by too many autonomous museum assistants. Rather, it is that AI is tacitly normalising in internal systems, without the board, supervision or public still seeing clearly where exactly the dependencies have arisen.<\/p>\n
So anyone who wants to know how far along the cultural sector is with AI should not only look at what is in the foyer or appears on the homepage. The real movement is taking place deeper inside the organisation. Publicly, it still remains quiet. Behind the scenes by no means. And that is a pity because cultural institutions can play such a strong role in interpreting social developments.<\/p>\n