[MUD-Dev] More on Small Worlds

Koster Koster
Fri Mar 14 14:05:19 New Zealand Daylight Time 2003

This is a brand new bit of research:


start quote--->
Pro bono publico
Mar 13th 2003 
>From The Economist print edition

The evolution of punishment explained

HUMANS, unlike most other animals, often co-operate with individuals
to whom they are unrelated. That puzzles evolutionary biologists. It
means they have to explain the spread of genes which cause an
individual to engage in altruistic activities that are costly to
perform, and yet benefit only others. The usual assumption is that
favours will be repaid. The question is how, given the number of
cheats and freeloaders around, that repayment can be relied on. And
one of the answers is to punish the cheats.

However, imposing punishment, too, is costly-often, it seems, more
costly than an individual's own interests could justify. So the
problem is merely pushed back a step. There still needs to be an
explanation for the evolution of this so-called altruistic
Robert Boyd, an anthropologist at the University of California, Los
Angeles, and his colleagues, think they have found it. In a paper
just published in the Proceedings of the National Academy of
Sciences, they discuss a series of computer simulations that show
how, once they appear, genes for altruistic punishment tend to

Dr Boyd modelled 128 groups of individuals who, together,
constituted a "virtual" society. He then let his model society
evolve. There were two distinct evolutionary pressures at
work. Individuals competed with one another in a classic Darwinian
manner. But entire groups competed as well.  In Dr Boyd's model (as
in the real world), these two pressures tended to work in opposite

Each group comprised individuals who adopted one of three
behavioural strategies. "Co-operators" helped the group to improve
its fitness at the expense of their individual rewards. "Defectors"
willingly accepted the help of others, but did not
reciprocate. "Punishers" behaved like co-operators, but also
punished defectors at a cost to themselves.

At first, the researchers ran their model without including
punishers in it.  In this case, when the groups were small (four
individuals per group), co-operators came to dominate. Freeloading
off other group members is not a great strategy when the others are
few in number. Co-operation, by contrast, helps everybody in the
competition with other groups. When the groups were larger, though
(up to several hundred individuals), defecting came to dominate. In
this case the benefits conferred on an individual by defecting
outweighed the cost to the group. Which is where punishment came in.

Dr Boyd showed that when punishers were introduced to the mix, even
if they started in only one group, their strategy spread rapidly
through the population. The benefits to a group of having punishers
to keep defectors in check outweigh the cost to individual
punishers. This is particularly true when defection is rare, as the
cost to punishers is then lower. And because defection does not pay
when punishers are common, it tends to be rare-a virtuous circle.

So much for the theory. But it also seems to illuminate
reality. Modern hunter-gatherer societies, which are assumed to be
similar to those of early man, have a maximum size of 150-180
people. Without punishment, the computer simulations suggest that
co-operation would have died out in groups of this size.

There is a long way to go, of course, before computational
anthropologists can accurately simulate human societies. Dr Boyd
hopes to move a bit in that direction by making the models of both
individual behaviour and group structure more complex. For example,
the existing model has no geography.  All the different groups are
thus equally likely to encounter each other.  However, the model can
already say something interesting about what might have happened
hundreds of thousands of years ago. And that is no small feat.
<---end quote

The abstract of the paper:



MUD-Dev mailing list
MUD-Dev at kanga.nu

More information about the MUD-Dev mailing list