Hacker News Reader

Comment by zinodaur | Hacker News Reader

Emergent behavior in animal groups is very cool. I once saw another interesting algorithm in the wild: schools of anchovies sharing the most dangerous roles in the schools.

The front and back of the school is the most dangerous spot - since it is the most exposed to predators. However, anchovies share this responsibility of being in the most dangerous spot by following two simple rules.

Rule 1. When you get to the front of the school, slow down to a speed slower than the school.

Rule 2. When you get to the back of the school, match the speed of the school.

Following these two rules means that the anchovies will constantly rotate between all the varying degrees of safety of the school, a problem that would be difficult to orchestrate from an outside perspective, but one that is easy for the anchovies to figure out for themselves.

View on HN

Why would they do this though? Why wouldn't a gene for 'cheating' (e.g. not slowing down when you reach the front) to avoid danger be favoured by natural selection? Animals with complex social structures (like us) avoid this by evolving to identify and punish cheaters, but there must be some other mechanism in these fish?

If you don’t slow down when you get to the front, you’ll end up separated from the school and more vulnerable to attack. Likewise if you don’t match the school’s speed when you’re at the back.

Staying in the middle would be advantageous, but when the whole school is cycling slowly, that’s tricky to do. The cheating behavior would need to be a little more complex so that’s an obstacle to its evolution.

Additionally, the more cheaters there are, the slower the school will move overall, which harms everybody (including the cheaters) so that limits the spread of cheating somewhat.

Overall, I’d guess that cheating could evolve, but only in a careful balance with the normal behavior. If the cheating starts to get significantly more complex, policing methods would co-evolve with it.

Interesting. So maybe it is an evolutionarily stable strategy, if you assume it's too hard for the fish to keep track globally of where they are in the shoal (and are therefore limited to strategies where they only react to what's immediately around them). It would be fun to simulate this and see if there are any other strategies that offer an advantage.

Because then you evolve societies that spend so much time fighting each other for marginal increases in individual safety that the benefit of working together is lost.

To be clear, though, there's no evolutionary function that optimizes for the overall benefit of unrelated individuals. So-called group selection has never been shown to exist in nature.

Cheating is moderated by the success rate of cheaters and non-cheaters individually. If cheating strategies come easy but counter-cheating strategies difficult, then the species success in the larger ecosystem will indeed be limited in as much as organized, goal-oriented behavior is beneficial. There's no group-wide selective pressure to favor the non-cheaters over cheaters as there's no group-wide inheritance mechanism that can accomplish that.

It's why schools of fish aren't as efficient as one might naively think they could be at evading, e.g., a giant whale. The cheaters can and do drag down the whole group because the equilibrium rate of cheaters in the school isn't responsive to the success rate of schools as a whole.

The same algorithm is found among human racing cyclists to rotate position at the front of the pack.

Reply on news.ycombinator.com