Cooperation and the Evolution of “Intelligence”

I ran across a reference to a study by Luke McNally, Sam P. Brown and Andrew L. Jackson, exploring the evolution of simple neural nets in a “social” population. The idea is that each “individual” in the population has its own neural network, and the individuals play simple strategy games against each other [1]. The neural networks decide the strategy and remember the outcomes. The games require decision to cooperate or not, and the payoff of the game are fed to the network.

This population was “evolved” by selecting the “fittest” networks, which reproduced with random mutations that modify the neural network. The children then repeat the “social” interaction, for 50,000 iterations.

The idea was to examine how the neural networks evolved in the presence of this social selection. Does the need to “get good” at social interactions affect these neural networks over generations?

In nature there is a general correlation between large brains and social behaviors, i.e., animals that live in packs, flocks, pods, families, and so on, tend to have large brains. Is there a causal relationship, and if so, which way? Do large brains enable complex social behaviors? Or does social living select for larger (or more complex) brains?

In this study, the social interactions control the reproductive “fitness” of the networks, with all other factors eliminated. It is an abstract test of the “social behavior drives brain evolution” hypothesis.

The results showed a couple of things. The neural networks learned various strategies to “win” the social games, and these were passed on to offspring. The strategies were not always the simplest or most optimal theoretical strategies, In particular, the networks generally remembered quite a few iterations of history, leading to idiomatic decision making strategies, functionally similar to each other (and to game theory), but highly individual.

These strategies required increasingly complex neural networks, i.e., the evolving population evolved networks with more nodes and connections to represent these strategies.   The authors label this complexity of structure and function as “intelligence”, hence their title “Cooperation and the evolution of intelligence”.

These results suggest that living in social groups could, just by itself, influence the evolution of brain structures supporting the cognitive demands of social behaviors. Living in cooperative groups might select larger brains to cope with the need to remember and strategize.

Is this increasing neural complexity justifiably called “intelligence”? I guess you could call these evolved networks with their Byzantine decision rules “more intelligent” than simpler networks. I would eschew the baggage of the term “intelligence”, and just call them more complex (though that term has its own baggage).

The authors are careful to point out that this small experiment cannot represent the complexities of natural evolution. The evolution of “intelligence” or brains was and still is likely influenced by many factors, including tool use, sexual selection, and social learning. For that matter, cooperative social behavior can be a very complicated “game” indeed, played on a board comprised of complex social networks and with multiple kinds of payoffs (e.g, advantageous access to food, safety, or sex).

Still, even with the limitations and the problematic invocation of the term “intelligence”, this is a neat study. It shows that social interactions between individuals might be select for the development of more complex neural structures.

---

1. Luke McNally, Sam P. Brown, and Andrew L. Jackson, Cooperation and the evolution of intelligence. Proceedings of the Royal Society of London B: Biological Sciences, 279 (1740):3027-3034, 2012. http://rspb.royalsocietypublishing.org/content/279/1740/3027.abstract