00100	       ARTIFICIAL INTELLIGENCE AND INTERSTELLAR COMMUNICATION
00200	
00300	                          by John McCarthy
00400	
00500		As  an  "expert"  on  artificial  intelligence asked to speak
00600	about interstellar communication, I am first obliged  to  attempt  to
00700	establish  a  connection  between  the  two  subjects.  There are two
00800	connections:  first,  the  study  of  artificial   intelligence   may
00900	contribute  to  stretching the imagination aboutthe possible forms of
01000	intelligence in the universe; and,  second,  the  possibility  exists
01100	that   we   might   find   ourselves  communicating  with  artificial
01200	intelligence as well as with natural intelligenc.  I shall  begin  by
01300	giving a summary of the state of research in artificial intelligence,
01400	then comment on the two above-mentioned questions, and  finally  make
01500	some  remarks  on  the  problem  of  interstellar  communication  and
01600	interstellar travel that are independent of my status as a specialist
01700	in artificial intelligence.
01800	
01900		Artificial intelligence refers to the attempts to make
02000	machine, specifically digital computers, behave intelligently,
02100	i.e. solve problems that humans require intelligence
02200	to solve.  There is no single reference that gives a summary
02300	of the present state of research in this area, but I can
02400	recommend (Minsky 1961), (Minsky 1968) and (McCarthy
02500	and Hayes 1969) for a start.
02600	
02700		More specifically, artificial intelligence research involves
02800	the search for intellectual mechanisms and their study
02900	mainly by means of their incorporation in computer programs for
03000	specific intellectual tasks.  We shall give some examples:
03100	
03200		1. Search of trees of alternatives.  Various search
03300	algorithms have been devised and tested in programs for
03400	playing chess and other games and in programs for proving
03500	mathematical theorems.  Good abstract search algorithms implemented
03600	on high speed computers play some simple games better than
03700	humans and can solve some toy mathematical problems, but are
03800	inferior to humans at chess and can't solve difficult mathematical
03900	problems.  To go beyond this performance seems to require several
04000	new methods which have not yet been implemented in a general way.
04100	Some of these are: the decomposition of problems into parts that
04200	interact weakly so that the parts can be solved separately
04300	and then combined; methods of making the specific
04400	problem solving situation control the search procedure more
04500	closely than at present (In some cases this involves the use
04600	of meta-statements about the state of the problem solving 
04700	process);.