From: Peter de Blanc (peter.deblanc@verizon.net)
Date: Thu Apr 07 2005 - 01:38:36 MDT
On Thu, 2005-04-07 at 01:40 -0500, Tomaz Kristan wrote:
> Predicting how much pain, pleasure or some other arbitrary measure, a
> possible action will bring, and selecting the best action by its
> _predicted_ value - that's the decision mechanism.
Thank you for explaining that to us.
I think Eli's point was not that, given a predictive algorithm, it's
difficult to implement a decision mechanism as a separate module; this
is easy. The problem is implementing the predictive algorithm. Of course
there are plenty of great algorithms for making predictions - some even
work in finite time.
But if your system works by *thinking*, then it has to decide what to
think about, how to structure itself to think more efficiently, and how
to structure the universe to give it access to the information and
resources it needs to make the predictions you require. These are all
subgoals of predicting. If the thinking module can decide how to
organize its thoughts, then it can decide how to organize the universe.
So the thinking module is by necessity an agent in itself, and its goals
are different from the goals of your decision module.
This archive was generated by hypermail 2.1.5 : Tue Feb 21 2006 - 04:22:55 MST