Life & Sport

What keeps me going when I am not training models or reviewing papers.

Research is a big part of my life, but it is not all of it. These are the things that keep me balanced (or at least try to).

Long-distance swimming

Lanes of an indoor swimming pool

Long-distance swimming is my favourite way to switch off. After the first few hundred metres the rhythm of strokes and breathing takes over, and the only things left are the water and my own thoughts. No e-mails, no notifications and no deadlines can follow me there. Those sessions give me a calm that is hard to find anywhere else: it is the closest I get to mindfulness, and I usually come out of the water with a clearer head than I went in with. More than one research idea has taken shape somewhere between two breaths, although I have learned that the hard part is remembering it once I am back on dry land.

Reading about AI, philosophy and the future

I enjoy reading about the philosophy of Artificial Intelligence and about where the technology may take us: what intelligence and understanding really mean, how humans and machines should share decisions, and which futures (good and not so good) AI could open for society. Two questions interest me in particular:

  • The transformative role of AI. Could AI reshape science, the economy and the way we live as deeply as earlier technological revolutions did, and how do we tell realistic prospects apart from hype? Jesús (Txus) López Lobo and I explore these questions in our book Can Transformative AI Shape a New Age for Our Civilization? Navigating between Speculation and Reality (Routledge).
  • Cognitive deskilling. As we hand over more and more of our thinking, writing and decision-making to AI assistants, we risk losing the very skills that allow us to judge whether their outputs are right. How to benefit from AI without letting our own abilities wither is, to me, one of the key questions of the coming years.

These readings rarely end up cited in my papers, but they shape how I think about trustworthy, safe and responsible AI, and they remind me why the questions we work on matter beyond benchmarks and citation counts.

Hiking

San Juan de Gaztelugatxe on the Basque coast
Green hills above a coastal town in the Basque Country

The Basque Country is a wonderful place to walk, with mountains, forests and the coast all within reach. Hiking is my way to disconnect, enjoy some slow time outdoors and, when the legs allow it, reach a summit with a good view. It also turns out to be one of the few activities in which nobody in the family asks me about neural networks.

The hardest optimization problem: research and a large family

Illustration titled Research, family and the art of balance: the researcher at his laptop with Nekane, Garazi, Hodei and Maialen, surrounded by conflicting objectives, changing schedules, uncertainty, explainability and the toughest reviewers asking but why

My most important project, and by far the most demanding one, is my family. Nekane and I have three children, Garazi, Hodei and Maialen, and reconciling them with a research career is a daily multi-objective optimization problem with many hard constraints and very few feasible solutions:

  • Conflicting objectives. Paper deadlines, project reviews, school runs, homework and weekend activities all compete for the same 24 hours. I have not found the Pareto front yet, but I keep searching.
  • Non-stationary environment. Just when I think I have adapted to the family schedule, the concept drifts: new timetables, new hobbies, new illnesses. Continual learning is not only a research topic at home.
  • Uncertainty quantification. Any estimate of how long a task will take at home should come with very wide confidence intervals.
  • Explainability. Teenagers are the ultimate black-box models: their outputs are hard to predict, their reasoning is hidden, and they badly need explanations. I have spent years working on explainable AI, and I still have no method that works on them.
  • Toughest reviewers. Explaining what I do for a living in a way that satisfies my children is harder than any rebuttal I have written, and “a computer that learns” is usually followed by a long series of “but why?”.

Jokes aside, Nekane and the children are the best thing that has happened to me, and they keep me grounded every day. Whatever I achieve at work makes much more sense when I can share it at home.