Image: Creatives Unite
By Gleb Tsipursky
Imagine a community arts centre using artificial intelligence to draft its next funding application. The tool proposes a compelling programme, an attractive description of the target audience, and a persuasive account of the social value the project will create. The document looks ready for submission. Nobody has yet asked the young people who use the centre whether this is the work they want to make.
This is the difference between producing material about a community and giving that community power over its cultural life. Artificial intelligence can make the former dramatically easier. Whether it strengthens the latter depends on who controls the decisions.
The question matters wherever public bodies, charities, cultural institutions and creative workers face pressure to do more with less. A system that produces posters, funding bids, captions and project ideas at speed can save precious staff time. It can also give established organisations another way to decide what counts as culture without opening those decisions to the people whose lives are supposedly being enriched.
Cultural democracy has always demanded more than access to finished products. It requires opportunities to create, to challenge the gatekeepers, to learn the skills of expression, and to influence the institutions that allocate resources. AI changes the cost of producing cultural material. It does not automatically distribute those opportunities.
Who sets the questions?
Culture Matters has already made the case for a digital commons. Dennis Broe’s January 2026 essay, “Our Culture: Our Code”, argues that access to AI systems and control of the knowledge they contain are questions of ownership and class, rather than technical convenience alone. https://www.culturematters.org.uk/our-culture-our-code-ai-class-and-the-fight-for-a-digital-commons/
That argument should be extended to the decisions made inside cultural organisations. A library might offer a free AI writing workshop, but who selects the themes, the languages, the models of good writing and the criteria for success? A heritage project might use AI to sort recordings from residents. Who decides which memories become prominent and which remain inaccessible? A council-funded festival might automate the first round of proposals. Who gets to challenge a recommendation that feels plausible to the software but wrong to the community?
These are ordinary questions of power. Access to a tool is valuable, especially when money and expertise are scarce. Yet handing out accounts while reserving all important decisions for managers leaves the old hierarchy largely intact.
The most useful first question for any publicly supported cultural AI project is therefore simple: whose judgment becomes more influential as a result of this change?
The Apprenticeship that disappears
The cost-cutting case for AI often begins with minor tasks. A junior worker no longer needs to draft the first press release. A volunteer coordinator can produce a rota without learning to resolve competing needs. A newly hired producer receives an instant project plan rather than helping to assemble it from conversations with artists, technicians and audiences.
Some repetitive work should disappear. Nobody gains dignity from filling in the same form for the tenth time. But a surprising amount of practical learning takes place inside supposedly routine tasks.
Someone drafting a description for a community exhibition must listen to the people whose experiences it represents. Someone organising a workshop discovers that the best timetable on paper may exclude shift workers or carers. Someone preparing a funding budget begins to understand whose labour is paid, whose contribution is invisible, and where promises exceed resources.
A machine can help with the document. It cannot be assumed to supply the relationships and judgment the employee would otherwise have built while preparing it.
This is where the question of class becomes acute. People who enter cultural work with family connections, financial security and experienced mentors may still have ways to learn the unwritten rules. Those without such advantages often depend on their first paid assignments to encounter them. If those assignments vanish, the knowledge gap can widen even while software access becomes more equal.
Research offers reasons to value AI assistance without pretending that the learning question has been settled. An experiment by Shakked Noy and Whitney Zhang in Science found that access to ChatGPT helped participants complete certain professional writing tasks faster and improved evaluated quality. The evidence concerns productivity on specified writing tasks; it does not tell us whether new cultural workers become better at listening, negotiating or making accountable artistic choices.
In another setting, Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied AI support for customer-service workers. They found productivity gains that were particularly large for less-experienced workers, suggesting that some expert practice can be made available through software. It is a promising finding. The task for cultural employers is to discover whether workers can eventually explain, adapt and challenge such advice, rather than becoming permanently dependent on it.
The Cultural commons needs paid learning
An apprenticeship is a social arrangement, not simply a course. It creates time to practise, an experienced person who can demonstrate why a decision matters, and space to make mistakes without being discarded.
Cultural organisations should preserve those features as they adopt AI. They could give early-career staff responsibility for assessing a generated exhibition text against oral histories and collection records. They could ask participants in a community arts programme to judge a draft publicity campaign for accuracy and respect. They could pay young producers to test whether a proposed event schedule actually works for the people involved.
The key word is pay. The cultural sector already asks too many people to acquire experience through unpaid activity. A transition that removes paid entry-level tasks and replaces them with unpaid AI training risks making this pattern worse.
In a strong programme, tools would take away low-value repetition while the worker keeps ownership of a decision that matters. A new staff member might draft three approaches to an inclusive outreach plan with software assistance, then meet local partners to learn which assumptions fail. Their final responsibility would be to explain what changed after those conversations.
That process creates a more useful cultural worker and gives the community an actual voice. It also makes the organisation’s decisions easier to question.
Let people disagree with the machine
Cultural institutions should teach people to recognise when a fluent answer carries a false assumption. An AI-generated curatorial label may sound authoritative while smoothing over contested histories. A grant narrative may describe a neighbourhood in terms residents would never choose. A draft audience strategy may treat everyone outside an established customer database as an unreachable demographic rather than an excluded participant.
The answer is to build meaningful challenge into the workflow. Before a publication, programme or funding decision is final, the responsible group should be able to ask where an assertion came from, whose perspective is missing and what evidence would change the recommendation.
This is especially important when a tool is used to rank ideas, applicants or projects. Criteria that appear neutral can reward people already skilled at describing their work in the preferred language. A polished proposal and a valuable cultural idea are not identical things.
A community panel should be allowed to challenge the weighting of a selection tool, not merely comment on its output after the budget is allocated. A project worker should be able to say that a generated summary misrepresents an artist’s intention. A participant should have a clear route to correct an inaccurate description of their community.
These rights have practical implications. Institutions can record who made an important decision, what information supported it and where people may seek correction. They can allow applicants to explain nonstandard work rather than being filtered out by a narrow template. They can make human deliberation part of evaluation instead of treating it as an inconvenience after automation.
Count participation, not just output
Organisations will understandably measure time and money saved. Those figures matter to bodies fighting for survival. They are poor substitutes for judging whether a cultural programme has become more democratic.
A fuller assessment would ask who now gets to contribute to decisions, whether first-time artists and workers can develop paid skills, whether more people can contest a mistaken description or selection, and whether resources reach groups previously excluded.
Consider a venue that uses AI to increase the number of funding bids it submits. If successful applications keep rewarding the same established organisations and their professionally polished language, output has risen while cultural opportunity may remain unchanged. Compare that with a venue that produces fewer documents but uses the hours saved to mentor new community producers, translate materials with local review, or open programme decisions to residents. The second venue may be generating more cultural value even though its document count is lower.
No algorithm can settle these priorities. Institutions, workers, artists and communities have to negotiate them in public.
More profit or a richer cultural commons?
The debate over AI and culture is often framed as a conflict between protecting human creativity and embracing a new technology. There is another question worth asking: who receives the benefits when work becomes easier?
If the gains mainly reduce employment and consolidate editorial or curatorial power, AI can make cultural production cheaper while leaving cultural life less open. If the gains fund paid apprenticeships, enable community decision-making, widen access to archives and languages, and give workers room for judgment, the same tools can support a richer cultural commons.
That choice cannot be delegated. It belongs to the institutions allocating resources and to the people demanding a say in how those resources are used.
Cultural democracy begins when more people can influence what gets made and how. The measure of a useful AI system should be whether it helps that happen. A world full of effortless cultural content will be a poor bargain if fewer people have the skills, the income or the authority to shape the culture around them.
