30-May-83 20:48:50-PDT,2062;000000000001
Received: from SUMEX-AIM by SCORE with Pup; Mon 30 May 83 20:48:47-PDT
Date: Mon 30 May 83 20:46:10-PDT
From: Derek Sleeman <SLEEMAN@SUMEX-AIM>
Subject: EURISKO...
To: lenat@SCORE
cc: hayes-roth@SUMEX-AIM, sleeman@SUMEX-AIM

Doug,
 I enjoyed your talk on Friday, but I was left with a number of points
I'd like to follow up with you:

1. How do you do credit/blame assignment - this is clearly critical
for your system, and I'd like to know more about how you do that.

2. Essentially, no progress has been made on the change of representation problem:
and so I think you were hasty to suggest that adding additional slots would
be a contribution. Would that not merely have the effect of adding an
attribute ie turning an N attribute description into a N+1.

3. I'm concerned about the use of EURISKO to produce heuristics to control 
dialogues. My understanding of EURISKO is that many heuristics are 
produced, and that only some are retained. If a "transient" heuristic were
to be used (before it is rejected) in a dialogue control, then the user
might well be hindered rather than assisted by the "intelligent program".
A possible use seems to be where the user acts as TRAINER, and provides
feedback on the dialogue and hence on the heuristics.
(However, I can't help but feel that better techniques exist with which
to tackle even this problem.)

4. At a recent seminar on Cognitive Psychology I noted that experiments
to show a relationship between performance and ability and performance
and aptitude had shown little of interest, however the study of 
Performance v (Ability and Aptitude) had shown several interesting
correlations. (And indeed several workers are now looking at N factors)
These observations do appear to have direct importance to EURISKO, it
says that many of the major POTENTIAL findings will be missed because
you are only changing one factor at a time. [May be not something to 
investigate tomorrow, but this point should be bourn in mind.]

I hope we can talk sometime...
Derek
-------