Daniel Innerarity: "It is crucial to feed AI with more diversity"

Daniel Innerarity, one of the most prominent Spanish thinkers in the contemporary world, raises in his latest book the need to “feed AI with more diversity” so that it does not become an instrument of exclusion.

Innerarity (Bilbao, 1959), Professor of Political and Social Philosophy, Ikerbasque researcher at the University of the Basque Country, presents these ideas in A Critical Theory of Artificial Intelligence (Galaxia Gutenberg, 2025).

And along with expanding diversity in the sources that feed these new systems, the professor adds another idea: the need to “supervise human interaction to ensure its compatibility with democratic values.”

In this interview, he discusses this new book, awarded the III Eugenio Trías Essay Prize. On the cover, a seagull on a bent railing facing the sea suggests, contrary to expectations, another way of viewing the technological revolution.

Why this cover and this book?

I didn’t want the typical image in blue tones, associated with the ethereal and the cloud, because I conceive of artificial intelligence in a more material way, in contrast to the human body. The photo, slightly retouched with AI, plays with that idea: a machine would interpret that the bar has bent under the weight of the seagull, while humans understand the joke immediately. It’s a metaphor about the differences between our intuition and AI processing.

I wrote this book so that it’s clear what’s at stake. There is a lot of hysteria about technology, with exaggerated expectations of technosolutionism. Silicon Valley wants us to believe that technology, not democratic deliberation, will solve the big problems. There is also exaggerated panic not aligned with real risks or our capacity to face them. AI does not impose a path; it opens several possibilities, as the current historical, political, and geostrategic context shows. It’s not predetermined that this will end badly or that technology will reduce our freedom; it can have an emancipatory potential.

Is ‘technosolutionism’ the ideology of the digital age?

It is the ideology of certain major players in the digital world, reinforced by spectacular technological development. Every day, promising or disturbing news emerges: yesterday, a woman regained speech thanks to AI; today, translators protest so that their work does not feed these systems. This carousel of advances and conflicts generates perplexity.

What is the role of political decision-making in an algorithmic democracy, considering that AI influences both debate and decision-making?

The decision-making process in AI involves design, data, and user interaction, and the key is when, how, and for what purpose humans should intervene. The lack of diversity among those designing these systems, especially the scarce female presence, can generate biases. An example is facial recognition that confused Black women with orangutans due to biased training with more images of white men. It is crucial to feed AI with more diversity, as Spain does with AI in Spanish and co-official languages [ALIA], against the current anglocentrism, and to supervise human interaction to ensure its compatibility with democratic values, as Algorithm Watch and AESIA (Spanish Agency for AI Supervision) do.

Why is it undesirable for AI to be like humans? Can we control it?

In aviation, the Wright brothers demonstrated that instead of imitating birds, one should study aerodynamics. The same applies to AI: instead of replicating human intelligence, we should leverage its computational capacity and handling of large data volumes. It works well in contexts with clear data and binary solutions, but in complex or ambiguous problems, humans remain irreplaceable.

We have not found an adequate control concept for AI: either we leave too much margin, risking values, or we apply excessive control that hampers its learning. The European AI Act moves in the right direction but remains generic. It’s like in 1868, when the UK required a person to walk in front of the car with a red flag to prevent accidents. The key is to combine speed and safety, as with modern cars or ABS, where technological improvements prevent the risks of abrupt braking in panic situations.

What legitimacy do we have to impose a European regulatory perspective?

The ideal would be to reach a global agreement on ethical, political, and democratic standards, but we have different cultures. Like climate change, we know it’s urgent, but progress is made through partial pacts. Nevertheless, European standards influence other actors. Additionally, AI and ecology overlap because the energy consumption of this technology is enormous and already part of the international climate agenda.

What does it imply that AI regulations are formulated by non-native digital people, with different perceptions of risks and benefits than new generations?

Digital transformation is not just about digitizing administrations but transforming society as a whole. Such a transformation must be democratic and inclusive, open debates, allow criticism, and consider generational, economic, geographic, and cultural gaps. It cannot be managed by ignoring diversity: digital natives, the elderly, migrants, rural inhabitants… It’s a litmus test: will we develop inclusive technology or let it divide and expel? We must prevent digitalization from becoming a tool of exclusion, as was perceived with the ecological transition during the yellow vest protests in France.

In The Age of Revenge, Andrea Rizzi proposes a global body to govern AI, like the IAEA for nuclear energy. How to do this in a context of distrust?

We live in that era, and global institutions are questioned for ideological or geostrategic reasons. The logic of “every man for himself” and zero-sum games prevails, making solid international governance difficult. Today, the win-win model is neither present nor expected. Regarding this, the pandemic had ambivalent effects. Positively, it promoted a cooperative response in Europe, boosting solidarity mechanisms like the Next Generation funds, unthinkable in 2008. Negatively, it deepened inequalities, such as online education, which excluded students from the school environment, and generated feelings of vulnerability and social irritation, fueling distrust towards authorities, migrants, and other countries, which fuels certain extremisms today.

What are the repercussions of that distrust?

Now, citizens have a greater critical sense, which can be seen as a democratic advance. However, I believe the idea of transparency is overrated, and our visual culture mistakenly assumes that seeing equals understanding. Instead of transparency—understood as simply “looking inside the black box”—I propose focusing on explainability: it’s not enough to show; it must be made understandable, because current technological systems are so complex that they can only be properly evaluated through reliable public mediation institutions.

How do we address algorithms that favor polarization and disinformation?

We are moving from an ecosystem with few sources of information to one with many, where the challenge is to discriminate the reliable ones. This abundance, beneficial and democratic, is distorted by social networks with algorithms that favor extremism, polarization, sensationalism, and punish moderation. However, algorithms can be regulated to promote dynamics compatible with a democratic culture, as the EU does.

How do we address the conflict between AI and copyright?

Europe leads in copyright and privacy regulation, but the balance is complex. Much of human creativity involves recomposing others’ creativity; we are not as creative as we think. It’s not about justifying plagiarism; there are some clear limits, but in other cases, it’s not easy to delineate what is original and what is repetitive, largely because we live in a culture with many products—such as scripts for series or bestsellers—that are trivial or based on unoriginal formulas.

AI is revolutionizing fields like medicine, but training lags behind. What would you say, in general, to teachers and professors, like yourself?

In areas like medicine (or law), the gap between societal expectations and a memoristic, repetitive, and uncreative training is very evident. Although in health and justice we tend not to experiment, a more holistic training would be necessary.

In general, I would tell teachers that we are no longer just transmitters of information. Since Google exists, that role has become trivial; students no longer need us to find data. What can we offer them that they cannot find on their own? We can help them formulate the right questions—because searching is not easy or spontaneous—and teach them to truly find what they seek. Google competes with us only in gathering information, not in the most creative aspect of teaching.

Source: Aser G. Rada / SINC Agency / CC By

Photograph: Juantxo Egaña / CC By-sa