Return-Path: <@SUMEX-AIM.ARPA, @MIT-MC.ARPA:DAVIS@MIT-OZ.ARPA>

Received: from rockefeller.ARPA (rocky4.ARPA) by rocky3 (4.12/4.7)
id AAO3002; Thu, 8 Aug 85 01:21:13 edt

Received: from SUMEX-AIM.ARPA (sumex-aim.arpa.ARPA) by rockefeller.ARPA (4.12
id AAOQ3973; Thu, 8 Aug 85 01:21:06 edt

Message-Id: <8508080521 .AA03973@rockefeller .ARPA>

Received: from MIT-MC.ARPA by SUMEX-AIM.ARPA with TCP; Wed 7 Aug 85 22:17:52-PDT

Received: from MIT-OZ by MIT-MC.ARPA via Chaosnet; 8 AUG 85 01:03:41 EDT

Date: Thu 8 Aug 85 01:01-EDT

From:Randal1lDavis<DAVIS,ZMIT-OZ@MIT-MC.ARPA>

subject: dendral performance

To: buchanan@SUMEX-AIM.ARPA

Cc: lederberg@SUMEX-AIM.ARPA, davis¥%¥MIT-OZ@MIT-MC.ARPA

Bruce/Josh:

I'm on a panel at IJCAI that has (among others) Stuart Dreyfus. He and
Hubert have taken out against expert systems this time and made some of the
usual unreasonable claims. I need a few facts to shoot them down. In one
remarkable passage, they cite Duda's expert systems article which says
roughly "dendral performs better than an expert chemist in some areas" and
then go on to say, "well that's no big deal, an expert chemist isn't going
to be an expert mass spectrum reader because most chemists don't read

mass specs". I suspect they are simply being mind-bogglingly literal
minded here and want to catch them at it.

So: is it in fact true that there are specific areas in which dendral

is as good as the expert mass spectrum interpreter? If so, can you
cite specific categories of molecules or organism families?

K Dosnp