Cultural investigation with artificial intelligence

How can we transcend the limitations of traditional cultural surveys? By combining spontaneous dialogues and AI analysis, a new approach could allow us to discover what we would never have thought to look for.

The limits of traditional surveys: a simplistic worldview

Cultural surveys are important for multiple reasons: understanding citizens’ practices in a given territory, collecting their expectations, better supporting professionals, imagining new projects that respect cultural rights, improving reception and communication, and even enriching artistic creation. Surveys, which constitute the very heart of sociology, are a valuable tool for working on the meaning of cultural policies. Survey has nothing to do with polling. Moreover, there are numerous survey methods across different sociological currents, and we have the right to invent our own.

In the field of cultural surveys conducted around cultural venues to inform cultural policies, the preferred method remains the questionnaire, generally administered online. These surveys are either conducted by professionals in a territory or entrusted to a specialized agency. The fundamental problem with this type of survey, essentially quantitative and statistical, lies in the fact that the framework of thought is entirely determined by its designers.

To be more direct: these questionnaire survey methods are, in my opinion, 95% useless for generating genuine innovations. Why such inefficiency? Because the predefined framework acts as an invisible prison. Physically, when you face a questionnaire, your unconscious immediately understands that it cannot step outside the box. Certainly, free response spaces are provided to allow respondents to formulate proposals outside the framework, but these free responses will never have the same analytical impact as formatted responses, precisely because they cannot be typologized within the analysis grid.

Statistics, which forms the basis of these surveys, represents a form of thinking that is certainly interesting, but fundamentally simplistic. Even cross-tabulated statistics remain a reductive view of the world, extremely partial in what they can teach us, particularly about potential innovations.

Moreover, who actually responds to these online surveys? What are their motivations? These fundamental questions about the meaning and impacts of protocols generally remain unexplored in the survey itself. Added to this is the punctual nature and high cost of these systems, even though cultural expectations and practices in a territory can evolve rapidly.

Yet, the very essence of a survey consists precisely in discovering what we had absolutely not anticipated. That’s its whole point. I am convinced that the quality of surveys, the openness of perspectives, and the richness of results largely depend on the method employed.

For a diversity of survey methods

This is why, in the support sessions I facilitate for cultural sector professionals, I often encourage them to design their own survey methods, to evolve them, or even to solicit others to imagine new methods. The diversity of methods naturally generates a diversity of results—it’s common sense. I also think that the survey process should be intrinsically linked to programming work, funding choices, and the development of cultural policies. It should be a continuous process rather than rare, punctual surveys that are costly and freeze our representation of the context.

I have developed a survey protocol proposal using artificial intelligence, which could, in my opinion, produce particularly useful, perhaps surprising, and potentially transformative results. I have already conducted some marginal experiments with this approach, without having deployed it in its entirety, and the initial results suggest considerable potential.

An innovative protocol: spontaneous dialogues and recordings

Here is the protocol I propose: in a cultural venue—let’s take a theater as an example—I suggest that reception staff engage, for a week for instance, in in-depth dialogues with visitors. This system somewhat formalizes the informal: we explicitly propose to discuss in this particular place, creating opportunities for exchange that go beyond usual interactions.

It is crucial to ensure that this week welcomes sufficiently varied groups and individuals, different types of audiences, different generations, different backgrounds. Ideally, it would even include conversations with people around the venue who don’t enter it, but let’s first focus on actual visitors.

Let’s imagine five people in this reception staff, each equipped with a portable dictaphone (and it’s important to choose quality devices, as some models are inadequate). The instruction is simple: dialogue with visitors, more than usual, take the initiative to engage in conversations and exchanges. Discussions can be individual or in groups. No other directive is given. No pre-established grid of questions.

Visitors are of course informed about the recording and its purpose: “I have a small dictaphone attached to my clothing that’s recording us, would you agree to chat a bit?” This transparency constitutes the only necessary framing.

Naturally, since these exchanges take place in a cultural space and are conducted by reception staff, the venue itself is implicit in the conversations. However, to maximize the openness of exchanges, it is crucial to frame them as little as possible while encouraging them. Discussions can be about the theater, but not necessarily—the important thing is to let the conversation follow its natural course.

Each evening, the recordings are classified, archived, and identified according to their context: discussion with a group of children, exchange with a retired spectator, with a family, etc. Each recording must be carefully contextualized and documented. After this week of recording, voice recognition is applied to all conversations, thus producing a collection of textual documents, each contextualized, relating the exchanges. These can vary considerably in duration, from thirty seconds to an hour. Voice recognition, a reliable technology that has existed for about thirty years and has improved considerably recently, is not strictly speaking artificial intelligence.

The analytical power of artificial intelligence

The next step consists of submitting the entirety of this “dataset” (as we say in the jargon) to an artificial intelligence. Claude or ChatGPT currently seem the most suitable for this type of analysis, notably thanks to their capacity to process vast volumes of documents and establish complex connections. A paid subscription is necessary to process such quantities of data, as free accounts are too limited.

The crucial aspect lies in formulating the request to the artificial intelligence. We must absolutely avoid asking specific questions so as not to filter responses through our own framework of thought and thus reproduce the restrictive logic of traditional questionnaires. We simply ask the AI to draw conclusions from these documents and formulate recommendations, without predefined objectives. This is where we touch on the singular power of these tools: their extraordinary capacity to connect a very large number of heterogeneous data points, to identify patterns invisible to the human eye, to formulate deductions from disparate elements. Certainly, AIs have their own filters and biases, as we have ours, but their mode of information processing differs fundamentally from ours.

They can analyze not only the explicit content of conversations, but also thought movements, underlying reasoning structures, implicit thematic recurrences. Thus, even a conversation apparently distant from cultural subjects can, through its thought structure or indirect references, nourish relevant ideas for the institution.

Artificial intelligence, by establishing multiple connections and precisely because no directive question is posed to it, can generate reflection totally outside our initial framework of thought. I am convinced that this approach would produce completely unexpected and relevant leads, precisely because it relies on real human exchanges, much less formalistic than a list of questions.

These proposals will often appear to us as obvious in hindsight, even though we would never have thought of them spontaneously. This is the whole strength of this approach: revealing what was there, before our eyes, but that we could not see because of our own cognitive limitations and pre-established frameworks of thought.

Perspectives and future experiments

This method represents much more than a simple alternative to traditional surveys. It embodies a different philosophy of research in social sciences applied to the cultural sector: an approach that trusts in the richness of spontaneous human exchanges and in artificial intelligence’s capacity to extract meaning from them, without imposing a preconceived reading grid.

The relatively modest cost of this approach (a few quality dictaphones, staff time, an AI subscription) makes it accessible to most cultural institutions. Its flexibility allows it to be adapted to different contexts and repeated regularly to follow the evolution of practices and expectations.

This would be an experiment worth conducting, which in my opinion could bring a lot, including many salutary questionings. The initial trials I have been able to conduct at the margins of this approach suggest a transformative potential for our survey practices and, beyond that, for our understanding of territorial cultural dynamics.