From: scerir (scerir@libero.it)
Date: Tue May 01 2001 - 06:12:28 MDT
> Extropians is a forum for exchange of ideas about the future,
> and it is also an ongoing party of sorts. Many pleasant
> acquaintances in RL start here.
> spike
Is Extropians, also, a collective intelligence?
Collective intelligences are large, sparsely connected,
recurrent neural networks, whose "neurons" are "reinforcement
learning" algorithms.
The distinguishing feature of collective intelligences is that
their dynamics involves *no* centralized control, but only
the collective effects of the individual "neurons",
each modifying their behavior via their individual
"reinforcement learning" algorithms.
This restriction holds even though the "goal" of the collective
intelligences concerns the system's global behavior.
So we must assume, essentially, no knowledge concerning the dynamics
of the full system.
And this rules out any approach that models the full system.
In a perfect collective Intelligence the *local* utility (function)
must be initialized, and updated, without centralized control,
so that as the "neurons" improve their utilities, *global* utility
also improves.
But, for a perfect collective intelligence, an additional condition
is required: that the "neurons" do not frustrate each other
as they attempt to increase their utilities !
(That's why the human economy is not a collective intelligence).
(D. Wolpert and K. Tumer - "Collective Intelligence" - In J. M. Bradshaw,
ed., "Handbook of Agent technology" - AAAI Press & MIT Press, 1999.)
This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 08:07:24 MST