From 7 to 9 July 2026, twenty-two experts from France, Germany and Poland worked together on educational policy recommendations in the face of generative AI. I took part in it. Here I recount the purpose of this gathering, its working method and the scope of what we built there.
A place founded in 1993 for dialogue between three countries
Genshagen Castle stands in Brandenburg, some twenty kilometres south of Berlin, in the middle of an English-style park. In 1993, Brigitte Sauzay, who had been the interpreter of three French presidents before becoming Chancellor Gerhard Schröder’s adviser for Franco-German relations, and the historian Rudolf von Thadden founded there the Berlin-Brandenburg Institute for Franco-German Cooperation in Europe. Their intuition, four years after the fall of the wall, was to carry the Franco-German dialogue into the Länder of the former East Germany, and to associate Poland with it from the outset, in the spirit of the Weimar Triangle that the foreign ministers of the three countries had just created in 1991. In 2005 the institute became the Genshagen Foundation, a public-law foundation supported by the German federal government and the Land of Brandenburg. It has been working since then on two grounds, artistic and cultural dialogue in Europe and European political dialogue, organising in this castle gatherings where artists, researchers, lawyers, heads of institutions and public policy makers cross paths.
Artificial intelligence has established itself in this programme as a social issue in its own right, at the crossroads of the foundation’s two missions. In June 2025, the foundation gathered there twenty experts from the three countries for a first forum on cultural policies, around culture, democracy, civil rights and the protection of authors’ rights in the age of AI. This forum led to the “Genshagen Proposals”, nineteen recommendations addressed to European policymakers, among which already figured the call to develop an AI literacy founded on the understanding of the interdependence between technology and society (I attach this document to this article). The July 2026 colloquium, “Between Creation and Simulation”, is its continuation, the foundation this time shifting the focus towards education, cultural practices and the training of young artists. The place, incidentally, is part of the method. People sleep there, take their meals together, walk in the park between two sessions, and the conversations that form on the sidelines count as much as the plenary sessions.
Two scenarios for 2036 by way of opening
In the invitation document, the organisers set out the starting point plainly: the spread of generative AI is driving a deep cultural transformation, comparable to the invention of printing, with a risk of deskilling, that is, of loss of individual skills, of elementary cultural practices and of collective knowledge, while no real public debate exists on what school curricula and the training of young artists should become. They stated their ambition to draw up trinational educational policy guidelines, and to contribute to a European alliance for an ethical, human-centred and creative education.
Susanne Keuchel, the foundation’s director in charge of arts and cultural education, opened the gathering with a prospective fiction in two scenarios for 2036. In the first, a student learns alone at home, schools having disappeared for lack of funding, and a commercial AI steers her studies as well as her leisure. In the second, in-person learning centres rely on tools designed according to European ethical standards, under educational objectives that remain public and human. Her argument bore less on technology than on decision, for the first scenario requires no choice from anyone, it is enough to wait, as society waited in the face of social networks before regulating once the harm was proven; the second demands choices, to be made now. This asymmetry between the inaction that decides and the decision that costs served as the backdrop for everything that followed.
Twenty-two people who do not know the same things
The composition of the group was the first working tool. Around the table were composers, lawyers from the collective management of rights (ZAiKS in Poland, VG Bild-Kunst and GEMA in Germany, CEPIC for photo agencies), a professor of computer science and artificial intelligence, a researcher from the CNRS working at IRCAM, teachers from art and design schools in Berlin, Geneva and Warsaw, a professor of translation studies, journalists and authors, a representative of the French Ministry of National Education, heads of networks of expertise on AI in culture, and the foundation’s teams, including two interns whom we insisted on having participate fully, because they had things to teach us. The full list of participants appears in the annexe to this article.
This diversity nourished the work at every stage. When a wording touched on authors’ rights, the lawyers immediately spotted that a word like “transformative” has a precise meaning in intellectual property and would be misread. When, among the draft recommendations, figured the idea that universities should develop their own models, the practitioners recalled what training a model really costs and what an art school can or cannot do, and we moved the recommendation towards what is within its reach, experimenting with workflows using open models. When some grew exalted about the creativity of machines, the representatives of authors brought back the figures of income losses, above fifty percent among translators, photographers and illustrators. Each of us held a piece of the real that the others did not see.
From the two boards of post-its to the eight workshops
The working method deserves to be recounted, because it embodied what we were seeking to recommend. After two framing panels, one on artistic training, the other on general education, we spent a whole morning collectively transforming the harvest of the discussions, noted on two large boards, into workshop topics. Eight themes came out of it, which we dealt with in two sessions of four groups, each group being composed of participants from the three countries: human relations, values, ethics and human rights; educational frameworks and rules for institutions; the human-machine relationship and AI literacy; skills, knowledge and unlearning; creative mindsets as a resource for society; sovereignty; the impact of AI on culture, society and the economy; language and narratives.
I took part in the group on human relations, values and human rights, then in the one on language and narratives. In both cases, the work was done with markers on flip-chart sheets written by several hands, and all the members of the group shared the plenary report among themselves. None of us was working in their mother tongue, English serving as the common language, which obliged each of us to a form of simplicity and gave the translators and linguists present a discreet role as arbiters of meaning.
A text negotiated sentence by sentence
We devoted the last morning, on the Thursday, to the synthesis. The drafting team handed us a first version of the future text, with a preamble containing our common positions and the skills to cultivate, then a list of recommendations extracted from the eight workshops, and asked us to concentrate our additions on concrete recommendations, for the materials abounded but many ideas remained at the state of notes, not yet formulated in such a way that a policymaker could take hold of them. We first went back physically to the flip-chart panels, spread through the rooms of the castle, to check that nothing essential had been lost along the way, each panel being reread by at least one person who had taken part in the workshop concerned. Then, during a round of the room of nearly three hours, we discussed, amended and weighed every sentence.
The discussions often bore on a single word, and that word engaged each time a vision. We replaced “teachers” with “educators” and “pupils” with “students” to cover school as well as higher education. A lawyer argued for speaking of “authors’ rights” rather than “copyright”, because copyright also protects companies whereas authors’ rights are attached to the person and belong to human rights. When I proposed to anchor respect for human dignity “in positive law rather than in declarations of intent”, a lawyer pointed out to me that “positive law” is a technical term that would not say what I meant in English, and we wrote “in legislation”.
Some debates ran through the whole room. One participant proposed treating European independence in AI as a matter of sovereignty, defence and security, on the grounds that cultural influence has become a weapon in a hybrid war; others refused that our group should plead for directing more money towards defence, when it is artistic and cultural sovereignty that needs defending. We ended up agreeing, all of us, on one formulation, sovereignty in the service of cultural diversity, democratic values and security in Europe. Likewise, those who know the European funding programmes set aside a proposal to create a new European research institution, because these structures already exist and directing more money to them would come at the expense of the arts and humanities; we therefore moved the recommendation, to ask that art schools, absent today from the European computing infrastructures, be given access to them and be associated with their development.
We added other points as the round went on: transparency of any use of AI in textbooks and teaching materials, with publishers’ responsibility; verification of school contents by human experts; disclosure of who funds apparently independent expert opinions, a lawyer having recounted the real case of an expert presented in the European institutions as independent and funded by large platforms; compulsory training of teachers in AI literacy, which people compared to a driving licence and which we set in tension with the right of each person not to use these tools in their classroom, the balance point being that training enables an informed choice without imposing use; and the idea, taken up from Axel Jean of the French Ministry of National Education, that for the first time teachers and students are learning a technology at the same time, which calls for mutual learning rather than vertical transmission.
I brought for my part three proposals for the preamble, coming from our workshop on human rights:
- the motto “the human person is an end, not a resource”, brought into our group by Dorota Danielewicz and whose genealogy goes back to Kant, with the requirement of anchoring respect for human dignity in legislation rather than in declarations of intent;
- the idea of a chain of resistance to manipulation more robust than intellectual education alone, running from emotional awareness to self-confidence and mental health;
- the affirmation that technology is not neutral, that AI must be treated as a socio-technical reality and that AI literacy is founded on the interdependence of technology and society.
And three proposals on the side of the recommendations:
- the pedagogy of play as a critical method;
- the systematic collection of empirical data on the real uses of AI by students and teachers, so that information about social and technical changes travels faster up to the institutions;
- support for horizontal peer-learning communities of teachers and educators.
From authors’ rights to the human relationship
The detail of the recommendations belongs to the official document to come. Their scope, however, can be described. They are organised around a few large grounds that imposed themselves over the course of the workshops:
- authors’ rights, with the transparency of models’ training data, the right to remuneration and moral rights, treated as human rights and not as a sectoral question;
- European sovereignty over tools and infrastructures, from the software used in schools to institutions’ access to computing resources;
- the governance of the educational sector, with standards, advisory boards linking school to the changing world and transparency of public-private partnerships;
- teacher training;
- a critical AI literacy throughout the educational journey, including the real harms of these systems and awareness of authors’ rights;
- pedagogy and assessment, with the primacy of process over product, final human decision in grading, AI-free spaces and laboratories for experimentation;
- and, running through all the rest, the human relationship and trust, education playing out first of all between human beings, in institutions that must remain environments of social and democratic learning.
I find in this architecture, almost term for term, the fields I have seen taking shape over the past three years in my own work accompanying cultural policies in the face of AI. This convergence, at which each of us arrived by their own paths, seems to me a serious sign that these grounds are the right ones.
A way of producing public policy
What I bring back from Genshagen goes beyond the content of the recommendations. We maintained real disagreements to the very end, between a computer science professor who defines herself as an AI dissident and practitioners who experiment with it every day, and we nevertheless produced a common text, not through a soft consensus but through a work of precision, each side accepting that its position be formulated in terms the other could sign. The method itself, trinational, horizontal, negotiated sentence by sentence, with interns taking part on the same footing as the directors of institutions, was already a putting into practice of what we recommend for education. The drafting team will now consolidate the results, send them to the participants, then present them and put them up for public discussion, in Berlin and elsewhere. I will for my part continue to carry this work into the French contexts where I intervene, for recommendations only exist if people circulate them, translate them and defend them, each from their own place.
