Interview with Daniel Innerarity. Artificial intelligence dreams of redefining the arts: between creativity and simple repetition.

Trump announced a $92 billion investment in artificial intelligence (AI) in data centers in Pennsylvania . Mark Zuckerberg of Meta plans to invest "hundreds of billions of dollars" in artificial intelligence (AI) infrastructure. What is his stated goal? To build a "superintelligence." For some years now, it has been claimed that Meta has lagged behind in generative AI , dominated by OpenAI since the launch of ChatGPT , and by Google . They all have one thing in common: they aspire to create a so-called "general" or "superintelligence" AI, with cognitive abilities superior to those of humans, capable of making scientific discoveries and inventing technologies on its own. We are living through a revolution. And this definitely impacts the cultural field and its spaces.

The first step to delving deeper into this infinite and invisible territory is knowing how to write a prompt . What is it? It's the instruction, question, or request given to an AI system to obtain a specific response. When it's accurate, complete, and imaginative, it produces better results on the screen. In fact, private universities around the world already offer short two-month courses on "How to Write a Good Prompt." The techno-optimist who teaches it has found a new job.

Daniel Innerarity is a Spanish philosopher and holder of the Artificial Intelligence & Democracy Chair at the European Institute of Florence , where he directs the Institute for Democratic Governance . He has just published A Critical Theory of Artificial Intelligence (Galaxia Gutenberg) and explains, among other things, whether AI can become intelligent: “Artificial intelligence, as it is known, is only intelligent if we apply the word intelligence improperly. It has aspects very similar to activities that we humans carry out, some even better than ours, such as the ability to calculate, to handle data, etc. These are some of the fundamental properties that we humans have,” he notes with his powerful voice from Spain.

–Many celebrate the fact that it's possible to not recognize whether a painting was made by a human or an AI. What implications does this position have for this and other cultural fields?

–All the arts in general, especially music and painting, are experiencing spectacular development thanks to AI processes, and above all, it's becoming more feasible for many people to create music, compose, and paint in ways that are unimaginable. However, AI has limits. It's quite possible that the difficulty isn't so much creating paintings like Van Gogh's , or expanding Beethoven 's symphonies and Bach 's sonatas, as inventing a new Bach, Beethoven, or Van Gogh. The originality of these geniuses wasn't due to AI. Another thing is that once these geniuses existed, we know their works and their fantastic artistic strategies. AI can increase the number of their works; is that a problem? No, whether we can invent a new genius with those characteristics is another matter.

–What remains of Walter Benjamin's seminal essay, "The Work of Art in the Age of Technical Reproduction,"? Does anyone wonder about the aura?

–Thanks to AI and the ease with which it composes, paints, customizes, and modifies works of art to our liking, it has become clear that the idea of creativity is less exclusive, and is within reach of more people than before. Today, anyone with a phone has become a photographer. Does this mean that any of us will be on the level of the great photographers of history? Probably not, but we can create things of a certain value without them needing them to be exclusive or especially accessible only to a select few.

–What has the idea of creativity transformed into? Is AI art art or repetition?

–AI-generated art is art that captures the patterns found in a Van Gogh painting or a Bach sonata very well and combines them in a somewhat original way. Now, those patterns already existed, and ultimately, its creativity is a kind of combinatorial approach, not so much a particularly intense originality.

–More and more people are asking ChatGPT absolutely everything, googling much less, and spending less time on social media. Everything is centralized around Chat, and there are already confirmed addicts. What do you think about this situation?

–Generative intelligence and language models are replacing and sophisticating the simple search technology that represents what is becoming obsolete. Chat has greater agency, greater intervention than a traditional search engine. It must be acknowledged that traditional search engines also had a pre-design element that configured possible searches; that is, they weren't random searches, but rather searches guided by designers. But these new chatbots establish a more intense type of interaction with humans, and I think this has led Google to also incorporate AI processes into its searches.

–There are those who argue that the “inauthentic language” emerging from machines, robots, and AI is destroying languages like Spanish. What is your opinion on this?

–I think the analogy of what happens to those of us who don't have English as our mother tongue, but use it in our communication, in our teaching, and in scientific conversation, might be helpful. We use a standard English with little richness, with few nuances. Something that, for example, is also noticeable when one uses a translator. They are enormously effective, and their approach is almost unchallengeable when it comes to plain language with little meaning. They lack the richness of mother tongues when one wants to explore nuances, irony, double meanings, and implicit allusions, and this has already led to the English used in scientific communication—which one acquires by reading scientific journals of any kind—being a particularly poor English.

Here, a balance will have to be struck. I think it's very important that one level doesn't cancel out the other. There must be a universal language of communication, but we should also protect linguistic diversity because, at its core, it's about protecting the diversity and worldviews of the human capacity for nuance, for a language rich in meaning that brings us closer to a very specific field of human beings: ambiguity, irony, inaccuracy, double meanings, and the implicit—something that an AI translator finds difficult to grasp.

–Where does this leave nature? Will AI be able to interact with it to protect it? Or will it do so to extract more profit from it?

–AI has a problematic relationship with nature. We're describing it as an innocuous technology from the perspective of its environmental impact, when in reality it's less cloud-like than it's said to be and more matter, rare metals, humans working in harsh conditions, garbage production, and extremely high energy consumption. It's one of the major issues that needs to be investigated: how would an AI be possible that consumes less data and, therefore, less natural resources? At the same time, it's true that AI can help us economize on processes. The most banal of these is online communication, which, for example, would save us many polluting trips, as well as measuring the environmental impacts of the things we do.

–You argue that ChatGPT is extremely powerful when it comes to processing a large amount of pre-existing data, but not when it comes to making recommendations about new phenomena for which there is no information. Will there come a time when AI will no longer be lacking in information?

–Generative AI may significantly increase the available data and information it offers us, but it has a limit because there is no data about the future . It is not capable, and it will not be capable because it cannot anticipate in a way that is not in continuity with past data, nor behaviors, nor scenarios that are by their very nature unpredictable. The limitation of generative AI is not the past or the present, but the future. It also costs us humans because we are very repetitive and routine, but as Hannah Arendt said, the ability to give rise to new things, to the unpredictable, although it is verified only a few times in the history of societies and people, is a property we cannot do without.

–There seems to be an idea floating around that with good AI, anything is possible, even having a friend, as Mark Zuckerberg suggested. Can an AI assistant become a human's best friend?

–AI is an excellent substitute for human friendship . It can reach unsuspected levels of closeness, support, even pseudo-conversation. But what humans really seek in other people, and appreciate, is their unpredictability, their freedom, their ability to contradict our opinions or personal inclinations. A relationship with a robot has many advantages over a person: it's always available, we can modulate the type of responses we want, making them more accommodating or more contradictory, and so on. The fascination of relationships with other humans has a lot to do with our inability to dominate them. And that inability connects us with the strange, with the different, with free wills, and it's something that can never be replaced by a machine, which can be very gratifying, but doesn't have the infinite nuances and complexity that our relationship with another person has. Another thing is that many humans may want a company that is accommodating, simple, immediate, and of course, if we are only looking for that, it is better for us to go to a machine than to a person who might contradict us, but I believe that deep down in our hearts and affections, we do not want to deal with people who agree with us, but with those who exert some kind of complementarity, fascination, or mystery over us.

–How does AI fit with democracy and current policies? Will AI give me the political answers I want?

–AI poses two types of problems for democracy. The democratic system is an organization of coexistence that presupposes a type of conversation and a type of decision. Conversation and decision are the two pillars that support the democratic edifice. Both have been profoundly disrupted by the emergence of AI. On the one hand, conversation as a result of social media has become horizontalized, more democratic, but also chaotic. In contrast to vertical communication, where authorities are more or less secure, but at the same time imply a structure of verticality that is undemocratic, conversation in the new communicative landscape of networks is more democratic and chaotic at the same time .

The second problematic aspect is decision-making. We cannot cope with the complexity of the contemporary world without automated systems, without algorithmizing many of our decisions. The big question is to what extent certain problems, or certain aspects of the problems we face, require human or chemical treatment. And the distinction, it seems to me, has to do with the fact that humans are better than machines at making decisions when there is little data, an environment of uncertainty and ambiguity, and machines, on the contrary, make better decisions than we do when there is abundant data, situations are unambiguous, and there is no uncertainty, but rather binary solutions. Delineating these problems is the fundamental question that will occupy us in the coming years.

–Is the field of AI infinite?

–We're in a hype moment, one of enormous exponential growth in AI, and that makes the field of things it can do for us or instead of us seem endless. The history of AI combines moments of spring and moments of winter. When there are moments of spring, expectations soar, but so do fears, and the current one is such a moment. When we reach a plateau, when generative intelligence and the great language models hit certain limits, we will surely consider taking their limitations more into account, and therefore our expectations and fears will also be more limited.

–We want AI to be used in healthcare, education, and housing, but is there really the will to write prompts that address that? There are millions of trivial questions being asked in the chat, ignoring the energy consumption these requests generate.

–AI is introducing us to an unreflective field, and in a sense, we are digital sleepwalkers; we use tools without having sufficiently reflected on their impact, limitations, advantages, and regulation. I would say that one of the main challenges is to introduce, at the various stages of the AI lifecycle, humans who reflect, who ask questions, and not simply sleepwalkers responding to impulses they cannot consider themselves the creators of.

–What will happen to our decision-making capacity? Is it already affected by the use of AI?

–Our decision-making capacity is something that has already been greatly affected by automation. Without going into the case of AI, humans are increasingly using things that relieve us of having to make certain decisions, and it's not a bad thing that we stop deciding on banal or everyday things; it's certainly a benefit for us to focus on other types of decisions. The point is that we should design machines in such a way that we reserve certain dimensions for humans because decision-making is something specific to humans . Perhaps more disturbing is that we think we've decided when in reality someone else has decided for us.

This isn't exclusive to the AI field; it also occurs in the analog realm, where we often make decisions, for example, about consumption, thinking they're sovereign, and there's actually a system that has induced us to decide one way and not another. This is exacerbated in the case of AI, because the decisions it makes are more sophisticated, but we should be very vigilant, both individually and collectively, in regulation, etc., so that we can say that the decisions we've made thanks to AI are ultimately our decisions or that, indirectly, we have them as the author. In the same way, we can say that the decisions our legitimate rulers make are not decisions we make individually, but that our authorized rulers, legitimized on our behalf, act. They are decisions we can share and that bind us.

He is a professor of Political Philosophy, an “Ikerbasque” researcher at the University of the Basque Country , director of the Institute for Democratic Governance, and a professor at the European Institute of Florence , where he holds the chair of Artificial Intelligence & Democracy . He has been a visiting professor at universities including the Sorbonne, the London School of Economics, Georgetown University, and the Max Planck Institute in Heidelberg. He is the author of Politics in Times of Outrage (2015), Democracy in Europe (2017), Politics for the Perplexed (2018), Understanding Democracy (2018), A Theory of Complex Democracy (2020), Pandemocracy: A Philosophy of the Coronavirus Crisis (2020), The Society of Ignorance (2022), and Democratic Freedom (2023), among others.