[Bmi] Discussion on group intelligence

[Paul Werbos]

This subjective of collective intelligence is certainly an important one. Sometimes it works, sometimes it doesn't.

Last year, MANY proposals were rejected here when people said:
"We will solve the problem by using a multiagent system. We all know that the whole
is greater than the sum of the parts. We can even prove that our system as a whole will converge to
a Nash equilibrium, so it gives the best result."

The problem is that Nash equilibria are not in general Pareto optima. When each agent maximizes its own thing, treating other agents like mindless physical
objects, without taking advantage of what might be gained by making deals... the result is often so far suboptimal that it breaks things.
The design of games and markets SO AS TO MAKE the Nash equilibrium BECOME closer to a Pareto optimum
is one of the key ongoing challenges, especially in areas like energy. One of the possible approaches is to design distributed systems which are
physically distributed 9made up of many agents) but mathematically designed to be one integrated sparse system.

Of course, analyzing human brains and societies adds many other dimensions of complexity,
beyond the scope of a quick email. Certainly human groupthink screws up very badly at times, but equally certain there are types of corporate culture which generate very useful dialogue. (Like my next panels, if I can find the right people...)

Best of luck,

    Paul