00100	                         REFEREE'S REPORT ON
00200	
00300	     INTERACTIONS BETWEEN PHILOSOPHY AND ARTIFICIAL INTELLIGENCE
00400	
00500	                 by A. Sloman, University of Sussex
00600	
00700	
00800	⊂	⊂The  paper  should  be  published,  because  it  makes  some
00900	interesting  contentions  and  represents  a  first  discussion  by a
01000	philosopher of how to represent the world usably  by  an  intelligent
01100	machine.     However,  the  author  should  try,  if  he has time for
01200	further revision, to anticipate as many as possible of the  following
01300	criticisms from the A.I. side:
01400	
01500	
01600		1.  McCarthy and Hayes do allow other representations than by
01700	sentences  (see  pp.  467-468  of their paper), but give the dominant
01800	role  to  representation  by  formal  sentences  because   of   their
01900	attachment  to  the idea of making formal proofs that strategies will
02000	work.
02100	
02200		2. Your description of various analogical representations  do
02300	not  give  a  computer  programmer  confidence  that  he could make a
02400	machine construct them from observations and decide what to do on the
02500	basis of the information they contain.  Precisely, how is the picture
02600	information to be represented in  the  computer.   In  fact,  picture
02700	information  has  been  used,  but  the  M.I.T., Stanford, and S.R.I.
02800	groups are all using the pictures to generate sentences, in order  to
02900	decide what to do.  What should they do instead?
03000	
03100		3.   At the bottom of page 16, you will be accused of begging
03200	the question.  What precisely, does "select one leading to the  goal"
03300	mean  if not to select one that has been concluded to reach the goal?
03400	Do not the astronauts trust their lives to the correctness of  rather
03500	long  chains  of reasoning? Do they just envisage or imagine possible
03600	changes of configurations, select one leading to the goal,  and  then
03700	attempt to put it into effect:  ...
03800	
03900		4.   In   general,   it  would  be  worthwhile  to  make  the
04000	distinctions  between  and  the  respective  uses  of   Fregean   and
04100	analogical   reasoning   more  precise.   Incidentally,  your  fellow
04200	philosopher Hubert  Dreyfuss  claims  that  analogical  reasoning  is
04300	impossible  for  a  computer and therefore artificial intelligence is
04400	impossible.