Consensus, Polarization, and Knowledge: How AI Can Help Shed Light on Our Societies

I had the privilege to attend Wivace 2024 (https://events.info.unamur.be/wivace2024), an annual international workshop on artificial life and evolutionary computation that this year was held at the Université de Namur. It was an extraordinary opportunity to look into the most fascinating and intriguing questions in and around AI.

One of the focuses was consensus. In particular, how can you have different agents share different -and partial- pieces of information and converge on a common shared knowledge?

Ahmed Almansoori presented an astonishing project of a group a research at the University of Namur leaded by Elio Tuci . A group of robots is allowed to move on a plane squared surface. Each robot can move only on a limited section of that surface, which is like a chess board, with black and white little squares, except there are more black squares than white ones. Question: can the robots, sharing their own opinion, based on different areas of the surface, collectively determine whether the majority of the whole surface is in fact black or white? Yes, they can. In other words, you can have different agents with different opinions, based on their own different experiences, work together to reach a consensus on a general truth about the overall situation. Not only that, but the project also demonstrated that a neural network works better than a predetermined protocol in managing the process.

Sounds interesting? Here's another experiment, more directly concerning society. Eleni Michaelidou , along with a group of researchers from Edinburgh, presented their work on opinion dynamics, based on a model of Shane Mueller and Yin-Yin Sarah Tan, about the role of knowledge in consensus formation, opinion divergence, and group polarization. It turns out that a large number of topics publicly discussed, along with an easy possibility of debating opinions (social networks) and a large number of political formations, push the society towards polarization. Moreover, in the long run, it pushes it towards consensus too, which means that almost everybody will agree on some extreme view.

On the other side, the expansion of communication fosters cooperation too, according to another work, presented by Simone Righi of the University of Modena and Reggio Emilia. But in rather unexpected way. He focused his attention on informal communication between individuals about third parties -i.e. gossip. Yes, gossip! It seems that it serves as a mechanism for updating reputations and conveying information about others' perspectives, intentions, and tolerance, and thus it is vital for meaningful cooperation.

It's impossible to cite all interesting things that showed up in Namur. Among others, the experiments on an emergent naming system by Dat Nguyen and Nicolas Cambier from University of Amsterdam, the insights about graph neural networks by Pietro Lio' from University of Cambridge, the intuitions on the emergence of conscience by Fernando Rodriguez from University of Sussex, and the ones about a future "chemically embodied cognition" by Sergio Rubin from IULM University Milan. Talking of genes and embodiment, the gene regulatory network by Marco Villani and the influence of finite diffusion rates on protocell synchronization by Roberto Serra, both from University of Modena, and the emergent properties of groups of neurons of Aldo Genovesio , University of Eastern Piedmont.

All in all, a wonderful experience. Once more, a confirmation about how questions in and around AI can pave the way for a deeper knowledge about ourselves and our societies.