Hacker News Reader

Comment by sago | Hacker News Reader

Try it. With actual evolving populations with inherited strategies, and without hard-coding group selection.

Without hard-coding the groups as separate evolving units, you'll have a hard time.

In general, it is easy to show the group behaviors you want if you can treat the group as atomic for the purpose of the model. But you've just moved the goalposts, your 'individuals' are not the unit of evolution any more, and you've simply set the required behavior in the fitness function.

So try it and see.

View on HN

But... under his conditions it is clear that everyone in the group with the least altruists dies. There also seems to be no particular benefit for an individual of getting more points than anyone else. No point trying anything, it is clear prima facie.

Whether this has any relevance to the real world is the debatable matter. I would argue that it does, there are many examples where strong communities have grown as a result of general goodwill and altruistic behaviours, and these communities prosper compared to those where exploitative behaviour is common. There is of course another element which a simplified model neglects - communities, once formed, exert a strong pressure on its members to conform to the norm, be it good or bad. This pressure has the effect that the individuals who depart from community norms are "evolved out" so to speak. In case of altruistic communities, this protects the group from infiltration by exploitative individuals. In case of exploitative communities, unfortunately, this perpetuates the communal problems. Eventually, if the two communities come into a conflict, the exploitative community will be eliminated.

What do you mean when you say "without hard-coding group selection"?

Are you saying the simulation is not valid unless collaboration between individuals arises as an emergent behavior?

If you hard-code group selection (setting the group as the unit of evolution), then you are already assuming what you want to prove.

> without hard-coding group selection

Unless you're disputing the existence of group selection in real systems, what does hard coding it change vs. a given system where it is an emergent property?

You have a model with 3 demes, with the deme having the lowest proportion of A-individuals being killed. This is no different from a model with 3 individuals with the individual with the lowest number of A-alleles being killed.

You can, of course, give a narrative argument on why you feel it is justified to exclude the interplay between these levels. But if you, at some point, try to show it is justified, by removing the assumption from your model, you'll struggle, I predict.

Evolving systems are very complex, it is an unwise assumption that you can chunk their dynamics into levels of abstraction and expect those dynamics to function independently or predictably when combined.

Reply on news.ycombinator.com