Some Principles Of Artificial Learning That Have Emerged From Examples

We argue t h a t A . I . shou ld no t l ose s i g h t o f the need f o r gene ra l p r i n c i p l e s when w o r k i n g w i t h problems in s p e c i f i c domains. We a l s o argue the case f o r s t u d y i n g A r t i f i c i a l L e a r n i n g . We presen t a computer program w i t h a ve ry l i m i t e d sense m o d a l i t y t h a t acqu i res some f a c i l i t y w i t h the E n g l i s h l a n g ­ uage and l ea rns some n u m e r i c a l concep ts . The p r i n c i p l e s by which such l e a r n i n g takes p lace are expressed in terms of a concept of process and they prove to be a p p l i c a b l e to l e a r n i n g the dec ima l numeral system and f o rm ing e lementary u t t e r a n c e s as w e l l a s l e a r n i n g t o i n t e r p r e t E n g l i s h sen tences . 1 . I n t r o d u c t i o n Being conv inced o f the c e n t r a l impor tance o f l e a r n i n g i n i n t e l l i g e n t behav iour and b e l i e v i n g t h a t r e a l l y complex "pe r fo rmance" programs w i l l be e x c e e d i n g l y d i f f i c u l t t o w r i t e and even harder t o ex tend by any o t h e r means, we have been add ress ing the prob lem o f A r t i f i c i a l L e a r n i n g ( A . L . ) . The term " a r t i f i c i a l " i s i n tended t o c a r r y c o n n o t a t i o n s s i m i l a r t o those i n " A r t i f i c i a l I n t e l l i g e n c e " . The i n v e s t i g a t i o n is to be in fo rmed by what we t h i n k we know about n a t u r a l i n t e l l i g e n c e and i t s development but the computer program wh ich is one o f the p roduc t s o f the s tudy w i l l d e p a r t f rom the p s y c h o l o g i c a l paradigm whenever t h i s i s e x p e d i e n t . The two a l t e r n a t i v e approaches to A . L . t h a t we have cons ide red are to inc rement a p e r f o r m i n g system t h a t a l r e a d y embodies some knowledge s p e c i f i c a l l y r e l e v a n t t o the m a t e r i a l b e i n g l ea rned o r , on the o the r hand , to se t up a program endowed w i t h the minimum o f s e n s o r i m o t o r c a p a b i l i t y and the s k i l l s needed f o r expans ion . Since the incremented program must possess such s k i l l s in any case , the na i ve e n t i t y i s i n h e r e n t l y s imp le r t o b u i l d . What i s more, i t becomes in t ime a p e r f o r m i n g system t h a t can be i nc remen ted , thus subsuming the a l t e r n a t i v e approach. The lesson t h a t A . I . has lea rned s ince 1960 (1) i s t h a t f r u i t f u l research i s most l i k e l y b y c o n s i d e r i n g s p e c i f i c domains i n d e t a i l . Th i s p o i n t o f v iew i s e l a b o r a t e d by Minsky and Paper t ( 2 ) . That does not mean, however, t h a t we shou ld abandon a l t o g e t h e r the search f o r genera l p r i n c i p l e s , o n l y t h a t success i s more l i k e l y by g e n e r a l i s i n g f rom some t h o r o u g h l y worked p a r t i c u l a r examples. The need f o r gener ­ a l i t y i s p a r t i c u l a r l y obv ious i n l e a r n i n g , s ince a system t h a t l ea rns o n l y a few w e l l c h o s e n t h i n g s may be s u p e r f i c i a l l y l e s s impress ive t h a t a w e l l w r i t t e n per formance program. The methods employed by the l e a r n i n g program must , as f a r as p o s s i b l e , be made n o n s p e c i f i c to the examples chosen. Our approach d i f f e r s f rom some ( e . g . ( 3 ) ) e s p e c i a l l y i n t h a t our program is no t g i ven express goals to which a p lan can be r e l a t e d l o g i c a l l y . Such would be i n a p p r o ­ p r i a t e to a na ive e n t i t y ( a p a r t f rom genera l goals l i k e " a s s o c i a t e symbol w i t h meaning" o r " c a r r y out i m p e r a t i v e s " wh ich are i m p l i c i t ) . Rather i t r e a c t s t o i t s env i r onmen t , i n c l u d i n g feedback f rom i t s own a c t i o n s , l a y i n g down p a t t e r n s o f behav iou r f o r f u t u r e use and r e t r i e v i n g them a c c o r d i n g to a s u i t a b l y a b s t r a c t e d r e p r e s e n t a t i o n o f what was p e r c e i v e d . The l a t t e r is n o r m a l l y a p a r t o f the p a t t e r n o f b e h a v i o u r . The aspects o f i n t e l l i g e n c e wh ich are examined in the p resen t work i n v o l v e a d i a l o g u e conducted in E n g l i s h and a l s o some numer i ca l concep ts . The program acqu i res bo th semant ic and s y n t a c t i c knowledge o f some words , wh ich i t subsequent l y uses and "comprehends" . I t a l s o acqu i res the meanings o f the numerals i n the dec ima l A rab i c sys tem, u s i n g the same bas ic r e p e r t o i r e o f a b i l i t i e s even though numerals are not c o n v e n t i o n a l l y cons ide red to be p a r t o f a n a t u r a l l anguage. The domain in which these a c t i v i t i e s take p lace has been s e l e c t e d f o r the utmost t e c h n i c a l s i n p l i c i t y . The program reads f rom and w r i t e s to a t e l e t y p e w r i t e r where a person i s s i t t i n g . The d i a l ogue i s concerned w i t h p r i n t i n g c h a r a c t e r s , i n c l u d i n g l e t t e r s t h a t make up words and sen tences . So i t i s a l s o p o s s i b l e t o d i scuss what i s b e i n g s a i d ; no u n n a t u r a l s e p a r a t i o n a r i s e s between speech and o t h e r a c t i v i t i e s . Beg inn ing w i t h no v o c a b u l a r y and o n l y the g e n e r a l a b i l i t i e s desc r i bed i n s e c t i o n 2 be low , the program acqu i r es s u f f i c i e n t knowledge to engage in a d i a l o g u e o f which the f o l l o w i n g i s an anno ta ted sample. The conven t i on is t h a t l i n e s en te red by the human t u t o r are preceded by a c o l o n ; the o the rs are produced by the p rogram. : PRINT THREE ASTERISKS AND A DOT * * * . : WHAT DID YOU PRINT BEFORE YOU PRINTED A DOT THREE ASTERISK ( c f . ( 4 ) ) : WHAT DID YOU SAY THREE ASTERISK : SAY TRIPLE AFTER I PRINT THREE CHARACTERS The word " t r i p l e " i s new to the program. " A f t e r " i s l i k e "when" i n t h i s c o n t e x t but i n the domain i n use t he re is no s i m u l t a n e i t y or ambiguous tempora l p r o x i m i t y and so " a f t e r " is more a p p r o p r i a t e . A k i n d of "demon" has been set up to awa i t the s p e c i f i e d c o n d i t i o n . These demons are not t r i g g e r e d by goa ls l i k e those of Papert (5) but by events as they can be c h a r a c t e r i s e d in the p rogram's r e p r e s e n t a t i o n . I f , l a t e r , we en te r a group o f th ree c h a r a c t e r s , the d e s i r e d r e s u l t o c c u r s . TRIPLE The program l e a r n s the rud imen ts o f the dec ima l number system so t h a t i t can do a d d i t i o n , sub ­ t r a c t i o n and m u l t i p l i c a t i o n w i t h two -p lace n u m e r a l s , : WHAT ARE 6 12S 72 The ma t te r i s d iscussed b r i e f l y in s e c t i o n 3 . Before t h a t we d e s c r i b e the p r i n c i p l e s on wh i ch t h i s work is based and then p resen t a d i a l o g u e demons t ra t i ng a complete l e a r n i n g sequence. At the