Under group selection, AI complementarity can prevail—because groups that preserve human variation explore more, and perform better over time. Strong group boundaries, institutions, and norms can act as a safeguard against long-run cultural decline.

Using evolutionary models, we compare AI substitutes and AI complements. We show that in a population with mixed strategies, AI substitution wins with higher efficiency, at the cost of a reduction in group variance.

New preprint with @thomasfmueller.bsky.social Julian Garcia @iyadrahwan.bsky.social: Group Selection as a Safeguard Against AI Substitution. arxiv.org/pdf/2602.03541

This paper is now published. www.nature.com/articles/s44...

Inspiring experience working with a group of great researchers. Now it’s the time for us to understand the current social changes and their influence on the information society.

Discussions on abortion are still very much needed, especially right now. Glad to see this paper finally came out. Do we still have the opportunity to do research like this?

After two years of trial and error, here are my tips on how to validate your model empirically. When we can't have perfect measures in social science, this is what we usually do and should be careful about: selfmademodeler.wordpress.com/2024/09/24/h...

Two weeks left to submit a poster to the multidisciplinary Machine+Behavior conference (Berlin, April 29-30, submit here: easychair.org/cfp/MaBe24) This is a great opportunity for early career researchers, I'm looking forward to the conference! Learn more about it here: machinebehavior.science