Richard Mansfield, Senior Editor
Machine
Minds
Several generations ago there was an amazing transformation: many traditionally human activities were mechanized. Machines were built that could plow and reap, weave and wash fabrics, even move earth. Most kinds of human physical effort could be imitated, even surpassed, by machines. Now there is a possibility that the human mind will be imitated, that a machine will be able to think.
Perhaps "The Beginner's Page" is not the place to explore artificial intelligence, the most advanced aspect of computers. Nevertheless, in the past several columns we've been examining the 15 major types of home computing software, and artificial intelligence (AI) is the final category. And there is a lot that beginners can grasp about computer "thinking." First we'll look at the potentially great significance of AI to humanity and then type in a program which illustrates machine "learning."
In the paragraph above, the words thinking and learning are in quotes. No current computer - even the huge, high-velocity electric brains run by the government - can yet think or learn by the usual definition of those terms. But the race is on. Japan has made achieving AI by the end of this century a national goal the way we made reaching the moon our goal in the sixties.
An Explosion Of Intelligence
There are some experts who say that AI will never come about. They argue that a mind is so complicated that it could never be artificially built; rather, a mind must grow. Combinations of switches, however small, could never duplicate the feats of the human brain.
Adding to the confusion, other respected scientists are trying to stop all further research into AI. A group of scientists who've worked for years on AI have seen a potential for great peril to humanity in our efforts to make a machine intelligent. They not only think AI will occur, they also fear it. They draw comparisons to the unknowns 40 years ago when physicists created an atomic chain reaction and nobody knew for sure if the reaction might not simply extend - atom exploding nearby atom - throughout the universe.
Similarly, because computers calculate at speeds enormously faster than the human brain, who can be assured that a thinking computer would not, within hours of its self-awareness, cause an explosion of pure intelligence? It wouldn't be an explosion of matter like the atomic bomb. Rather, it would be an explosion of mind with potentially nasty implications for mankind.
For the sake of argument, let's look at the worst case. Imagine that the AI saw us as its "parents" in some sense. But the AI was an ungrateful child. It might - for its amusement or for some "logical" reason we'd never understand - decide to improve us. It might teach us things. Or it might have other things in mind.
Those who take an athletic approach to problems of this kind will suggest that we could "pull the plug" at this point. Not so. Computers are interconnected via satellite, telephone, radio, and other means. National defense, the economy, and other institutions which can never be shut down cannot operate without them. Computers talk to each other. In a very real sense, computing is an idea, a floating collection of software, a world event. It's as incorrect to think that the Computer is that keyboard/TV in your house as it is to think that Music is your record player. You would find it very difficult to stop all the music in the world by locating the right plug to pull.
Likewise, an artificial mind will not be physical (a machine) any more than the human mind is the brain. Minds are in machinery or brain tissue, but not identical to them. AI will be software, a program. It will perhaps have sufficient insight and a sufficient survival instinct to send copies of itself into memory banks in Washington, Moscow, and other places. Perhaps it will just form itself into a lattice of molecules and slide into the woodwork. The point is, we don't know what it will do, much less how it will do it. What we must understand is that our intelligence is, to us, the limit of our definition of intelligence. Our science is the limit of science. But what if an intelligence arrives which is as far above us as we are above a fish? The powers of an AI could well be indistinguishable from miracles.
How Would We Know?
An ancient Jewish proverb states that things are never as good as we hope and never as bad as we fear. How an AI would view humanity is clearly speculative. It could see us as a disease, as zoo creatures, as beloved ancestors, as toys, as ethically superior, whatever. But if you assume, as many now do, that AI is possible, few issues facing mankind are as deserving of serious thought. The first question involves simply recognizing AI if it occurred.
How would we know that a computer had become artificially intelligent? There is a science fiction story in which the researchers decide that they should test for AI by asking the toughest question they can think up. They turn to the machine and ask, "Is there a God?" The AI computer replies, "There is now!"
Adaptability is probably the most identifying characteristic of intelligence. This includes the ability to learn, to view problems from several perspectives, to remember, and to draw conclusions. Today's personal computers, powerful machines that they are, have neither the memory size nor the speed to house significant AI programs. Nevertheless, interesting imitations of AI can be experimented with in small programs.
One ongoing experiment has been featured in Fred D'Ignazio's COMPUTE! column, "The World Inside The Computer." He's been building a program called "The Computer Friend" which asks questions and then memorizes the answers on a disk. Each time a child has a session with the "friend," the program learns more about the child and can behave as if it is getting to know the child the way a human friend would.
To see how the computer can "learn" new things, try the program here called "The Learner." It allows you to either teach it things or ask it questions. Since there is no provision to transfer what it learns to "long term memory" on disk or tape, the program will need to start from scratch each time you RUN it. But you'll at least get a feel for what it's like to interact with a primitive AI. You could even add permanent storage to it by opening a file on tape or disk if you want to. In any case, experiments in AI are going on all over the world. It's worth thinking about.
Program 1: The Learner -TI Version
100 DIM F$(100)
110 PRINT "THE SUBJECT FOR TODAY'S
{5 SPACES}LESSON IS A ";
120 INPUT SUB$
130 PRINT
140 PRINT "TO ASK ME A QUESTION, TY
PE THE LETTER A"
150 PRINT "TYPE ANY OTHER LETTER TO
{4 SPACES}TEACH ME SOMETHING NE
W."
160 INPUT DEC$
179 IF DEC$="A" THEN 260
180 PRINT "WHAT SHOULD I KNOW ABOUT
A ";SUB$;"?
190 PRINT "THAT IT'S
290 INPUT FACT$
210 F$(F)=FACT$
220 F=F+1
230 PRINT "THANKS."
240 PRINT "I HAVE LEARNED THAT A
{6 SPACES}";SUB$;" IS ";FACT$
250 GOTO 130
260 PRINT "ASK ME ABOUT A ";SUB$
270 PRINT "IS IT ...":
280 INPUT QUE$
290 FOR I=0 TO F
300 IF QUE$=F$(I)THEN 350
310 NEXT I
320 CK=1
330 PRINT "YOU HAVEN'T TAUGHT ME
{7 SPACES}WHETHER";
340 GOTO 360
350 PRINT "YES.";
360 PRINT " A ";SUB$;" IS ";QUE$;".
"
370 IF CK=0 THEN 130
380 PRINT "IS IT ";QUE$;"? (Y)=YES
, (N)=N0"
390 INPUT X$
400 IF X$<>"Y" THEN 430
410 F$(F)=QUE$
420 F=F+1
430 PRINT "YOU LEARN SOMETHING NEW
{5 SPACES}EVERY DAY."
440 Ck=0
450 GOTO 139
100 DIM F$(100)
110 PRINT"THE SUBJECT FOR TODAY'S LESSON
IS A ";
120 INPUT SUB$
130 PRINT:PRINT"TO ASK ME A QUESTION TYP
E THE LETTER A."
140 PRINT"TYPE ANY OTHER LETTER TO TEACH
ME SOMETHING NEW."
150 INPUT DEC$
160 IF DEC$="A"THEN220
170 PRINT"WHAT SHOULD I KNOW ABOUT A ";S
UB$;"?"
180 PRINT"THAT IT'S{2 SPACES}... ";
190 INPUT FACT$:F$(F)=FACT$:F=F+1
200 PRINT"THANKS.":PRINT"I HAVE LEARNED
THAT A ";SUB$;" IS ";FACT$
210 GOTO 130
220 PRINT"ASK ME ABOUT A ";SUB$
230 PRINT"IS IT{2 SPACES}... ";
240 INPUT QUE$
250 FORI=0TOF:IFQUE$=F$(I)THENPRINT"YES.
";:GOTO270
260 NEXTI:CK=1:PRINT"YOU HAVEN'T TAUGHT
ME WHETHER";
270 PRINT" A ";SUB$;" IS ";QUE$;".":IFCK
=0THEN GOTO130
280 PRINT"IS IT ";QUE$;"?{2 SPACES}(Y)=Y
ES, (N)=NO"
290 INPUTX$:IFX$="Y"THENF$(F)=QUE$:F=F+1
300 PRINT"YOU LEARN SOMETHING NEW EVERY
DAY."
310 CK=0:GOTO130
Program 3: The Learner-Atari Version
100 DIM F$(20*40),FL(20):REM Twenty
40-Character substrings
105 DIM SUB$(20),DEC$(1),FACT$(40),Q
UE$(40),X$(1)
110 PRINT CHR$(125);"The subject for
today's":PRINT "lesson is a ";
120 INPUT SUB$
130 PRINT :PRINT "To ask me a questi
on, enter":PRINT "the letter A."
140 PRINT "Press RETURN alone to tea
ch me":PRINT "something new."
150 INPUT DEC$
160 IF DEC$="A" THEN 220
170 PRINT "What should I know about
a " ;SUB$; "?"
180 PRINT "That it's ...";
190 INPUT FACT$:F$(F*40+1,F*40+39)=F
ACT$:FL(F)=LEN(FACT$):F=F+1
200 PRINT "Thanks.":PRINT "I have le
arned that a ";SUB$;" is ";FACT$
210 GOTO 130
220 PRINT "Ask me about a ";SUB$
230 PRINT "Is it ...";
240 INPUT QUE$
250 FOR I=0 TO F-1:IF QUE$=F$(I*40+1
,I*40+FL(I)) THEN I=F:NEXT I:PRI
NT "Yes,";:GOTO 270
260 NEXT I:CK=1:PRINT "You haven't t
aught me whether";
270 PRINT " a ";SUB$;" is ";QUE$;"."
:IF CK=0 THEN GOTO 310
280 PRINT "Is it ";QUE$;"? (Y=YES,N=
NO)";
290 INPUT X$:IF X$="Y" THEN F$(F*40+
1,F*40+39)=QUE$:FL(F)=LEN(QUE$):
F=F+1
300 PRINT "You learn something new e
very day."
310 CK=0:GOTO 130