Newsgroups: comp.ai.alife
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!EU.net!uknet!comlab.ox.ac.uk!sable.ox.ac.uk!trin0008
From: trin0008@sable.ox.ac.uk (Rick Heylen)
Subject: Re: memories as mass, gravity as desire
Message-ID: <1995Feb28.014038.26075@inca.comlab.ox.ac.uk>
Organization: Oxford University, England
References: <dobrzelewski.11.0014C2AF@cvm.msu.edu>
Date: Tue, 28 Feb 95 01:40:38 GMT
Lines: 45

In article <dobrzelewski.11.0014C2AF@cvm.msu.edu>,
Joel Dobrzelewski <dobrzelewski@cvm.msu.edu> wrote:
>
>At any one time, to determine what the bug should do next, we simply place 
>a small probe in the bugs mind.  Then, we let it go and fall for a short
>time.  The place that the probe ends up is the sensory place that the bug 
>would like to be in.  Now, we route the coordinates of the probe back into 
>some outputs.  Thus, the bug attempts to put itself in the sensory state that 
>it likes best.

I understand up to here. The masses are placed at coordinates in an 
N dimensional space. N is the number of sensors and the coordinate in that
dimension is the value that the sensor is returning. When we let the probe
fall, it gravitates to the sensory states it likes best. 
How do we translate this point on the sensory space into actions which the
bug must carry out in order to achieve this new sensory state?
I have some ideas but they involve calculus in N dimensions and higher.
How would you translate a certain sensory state into the bugs next actions?

>The only part I have left out is how to create the memories:  Each sense has 
>a component to evaluate its overall goodness or happiness.  For example, 
>the bugs eye might determine that too much light is bad because it will 
>damage the eye.  Again, these evaluations are genetic, hard-coded algorithms.

I have to question the design philosophy on the grounds that it doesn't appear
to generate any new information or to manipulate the information in any 
interesting way. The only novel thing you're doing to these hapiness values is 
applying a function to combine several different values into one mass.

I am not happy with the algorithm as it's not powerful. It is conceptually
simple but relativly wasteful on the memory/CPU time side of things.
All the work is done by the genetically derived rules for determining the
happiness associated with a certain level of stimulation of a sense.
There is no concept of cause and effect in the algorithm. The actions of
the bug are determined by the instantaneous position of this probe and I can't
think of a way of improving the method. Giving the probe an inertia would
only make the bug more manic. (work this one out!)

When I first read this article I thought it was quite enlightened and had
some novel and good ideas. I would dearly love to know how the author
proposes to map the sensory state of the probe to the actions that the bug will
take.

Rick Heylen

