00050	DRAFT	DRAFT	DRAFT	DRAFT	DRAFT	DRAFT	DRAFT	DRAFT
00075	
00100	             SYLLABUS FOR THE QUALIFYING EXAMINATION IN
00200	                       ARTIFICIAL INTELLIGENCE

00300	Computer Science Department, Stanford University, November 1969
00400	
00500		The subject of artificial intelligence  may  be  regarded  as
00600	divided  into  three  sub-topics: heuristics, perception and control,
00700	and representation theory.  The examination will  have  questions  in
00800	all  three  areas  with some emphasis on heuristics which is the best
00900	developed of the thrree.
01000	
01100		1. Heuristics.    This is the  problem  of  search  of  large
01200	spaces  of  alternatives for some particular kind of information. The
01300	best developed examples are: Game playing programs  search  the  move
01400	tree  of  a game in order to determine the best move. Theorem proving
01500	programs search spaces of  sentences  in  order  to  find  proofs  of
01600	theorems.  Dendral searches spaces of chemical structures in order to
01700	find one that satisfies conditions of mass spectra.  The  student  is
01800	expected  to have a specific familiarity with the state of the art in
01900	theorem proving as described in AA and with game playing as described
02000	in   AB.  Besides  that  he  should  be  familiar  with  the  general
02100	discussions of heuristic search in AC.  The examiners are aware  that
02200	all  these  discussions  are limited in scope and that there does not
02300	yet exist a systematic theory of  heuristics  nor  is  there  even  a
02400	single  source  that summarizes what is known. Questions that require
02500	the student to go beyond published  material  in  discussing  current
02600	issues may be included.
02700	
02800		You  may be asked to write a short program in LISP or another
02900	language, if you prefer, to express some heuristic procedure.
03000	
03100	
03200		2.  Perception and control.  Work in perception has taken two
03300	main directions: classification and  description.  Classification  is
03400	the  older problem and involves making the computer put patterns into
03500	categories corresponding to letters of  the  alphabet,  spoken  words
03600	from  a  fixed  vocabulary,  etc.   The  major  approaches are either
03700	mathematical oriented towards decision statistical decision theory or
03800	physiological  coming  from  ideas about the structure of the nervous
03900	system.  In either case they involve programs or hardware that  learn
04000	to  classify  the  patterns. This field is not much emphasized in the
04100	Stanford Computer Science Department and the cited literature  AD  is
04200	all the student will be held responsible for.
04300	
04400		The  problem  of  describing stimuli so as to be able to take
04500	some  action  has  received  greater  emphasis  in  our   department.
04600	Examples  of  this  are  visual  scene description for the purpose of
04700	redisplay of selected information, for  the  purpose  of  picking  up
04800	blocks  or  constructing  a tower, or the description of a road scene
04900	for the purpose of driving a vehicle.   The  literature  is  somewhat
05000	behind  in  this  field,  but  we  shall  ask  you  to  read AE.  The
05100	description of connected speech has also been  studied  here  and  is
05200	described in AF.
05300	
05400		3.  Representation  theory.   General intelligence requires a
05500	means of representing in the memory of the computer a picture of  the
05600	world adequate for receiving problems and for constructing individual
05700	representations for them appropriate for  their  solution.   This  is
05800	described in AG.
05900	
06000		The  examination  will be open book, will be held on March 7,
06100	1970, between the hours of 10AM and 5PM; bring a lunch or buy it from
06200	the machines.