Artificial  Intelligence: A  General Survey  by  Professor Sir  James
Lighthill FRS

	This   survey  was  commissioned  by   the  Science  Research
Council,  the  organization  that  provides  government  support   to
university scientific  research in Great  Britain, to help  it decide
what  to do  about grant requests  in AI.   It starts  by dividing AI
research  into  "three  categories  A,  B  and  C  according  to  the
long-term motivations for the three different types of work."

	In the  report, "A stands for Advanced  Automation: the clear
objective of this category of work  being to replace human beings  by
machines for specific  purposes, which may be industrial  or military
on the one  hand, and mathematical or scientific on the other." Next,
C is "computer-based  studies related to  the Central Nervous  System
(CNS) in man and animals."  Finally, B stands for "'Bridge Activity',
but also  for the basic component of that activity: Building Robots".
This last is explained by "Thus, a Robot in the  sense used here, and
by most  workers in the field,  is an automatic device  that mimics a
certain range  of  human  functions without  seeking  in  any  useful
sphere of  human  activity to  replace human  beings".  The goals  of
activities  A  and   C  are  considered  worthy  in  themselves,  but
"Research in category B, if acceptable arguments for doing it  can be
agreed, works  by its  interdependence with  studies in  categories A
and C  to give unity and coherence to the whole field of AI studies."
The survey continues  by evaluating the  results of past AI  research
in accordance with  this classification.  It finds  that "work in the
categories A and C of section 2 has some respectable achievements  to
its credit (and  achievement in such  categories of work  with rather
clear aims is  clearly discernible), but to a disappointingly smaller
extent than had been hoped  and expected, while progress in  category
B has been  even slower and more discouraging,  tending (as explained
in  section 2)  to sap  confidence in  whether the field  of research
called AI  has  any true  coherence.   In  the meantime,  claims  and
predictions regarding  the potential results of AI  research had been
publicised which  went  even farther  than  the expectations  of  the
majority  of workers  in  the field,  whose  embarassments have  been
added to by the lamentable failure of such inflated predictions."

	This review contains the following: my view of the nature of
AI and the progress that has been made to date, criticism of
Lighthill's categorization and conclusions, criticism of several
subsidiary points with a view to establishing that Lighthill has
not understood what is going on, and finally, an attempt to account
for the attacks on AI by Lighthill and others.  

	The scientific object of AI work is the understanding of
how intelligence works, i.e. how a mechanism can
manipulate information to achieve goals.  It is specifically interested
in mechanisms for solving "intellectually difficult" problems.  We contend
that it has turned out that intellectual mechanisms can be identified
and studied by a combination of theory and experiment.  The theory consists
of devising mechanisms that are hopefully adequate to solve a class
of problems and studying their properties, and the experiment consists
of writing computer programs embodying the mechanisms and testing
their behavior when applied to the problems.  We further contend that
many of these mechanisms can best be studied independently of practical
applications and independently of how these mechanisms may be carried out
in the brains of humans and animals.  We work independently of practical
applications for any of the following reasons:

	1. We are studying only certain aspects of intelligence and
want a problem that requires mastery only of that aspect.

	2. We want the maximum amount of effective experiment with a
minimum of expenditure of computer time or programming.

	3. A practical problem that is suggested may require unavailable
data.

	We work independently of psychology and neurophysiology because

	1. The neuron is a universal computing element and so assemblies
of neurons can carry out any computational process.  Therefore, what the
neurophysiologist have discovered about neurons tells nothing about
the higher level processes of the brain.  What they may discover about
the larger structures is potentially more useful for AI, but many of the
mechanisms of intelligence do not directly correspond to presently
identifiable parts of the brain.

	2. Psychology has not been of much use to AI until recently, because
psychologists had talked themselves out of studying mechanisms of intelligence.
Neither behaviorism nor psychoanalysis whose ideas dominated psychology for
many years was compatible with the study of concrete mechanisms for
intellectual processes.  AI has helped psychology liberate itself from these
doctrines for which some psychologists are properly grateful, and two way
interaction has started in some places.

	Well, what are these intellectual mechanisms, and what success has been
obtained in studying them.  Here are some personal opinions:

	1.  The most studied mechanism is the search of spaces of alternatives
for a solution to a problem.  Many ways of conducting this search so as to
reduce the "combinatorial explosion" have been devised.  Some of them are
specific to the problem or category of problems and some are quite general.
These ways are often called heuristics.

	2. The success of a search depends not only on the heuristics used
but also on what information is represented in the memory of the computer and how it
is represented.  Unfortunately, the study of representation is much less
advanced than the study of heuristics, and this often leads to a program
being improvable only up to a certain point without a fundamental restructuring.

	3. The information about how to solve a class of problems may have
been collected, but it may still be necessary to develop an efficient procedure
based on this information.  This field is called automatic programming and
is just beginnning.

	4. Information from sense organs or input devices has to be manipulated
to determine how the sensed part of the outside world is divided into objects
and how these objects are related in space and time.  Some progress has been
made in using visual and speech information.

	5. Experience needs to result in new knowledge.  Programs have been
written to learn from experience the values of parameters and whether particular
configurations are good or bad, but further progress in learning depends on
new results on the representation problem.

	In my opinion, some success has been achieved, but we have a long way
to go.

	1. Present programs for mainly heuristic tasks like game playing and theorem
proving are much better than their predecessors mainly because new mechanisms
have been discovered.  However, when improvement requires fundamentally new
representations, progress has been slower.  For example, all the work in chess
so far has been done without explicit representation of concepts like fork,
double threat, king's side attack, cramped position, combination, and positional
advantage.  No present program can be told about smother mates much less learn the
concept for itself.

	2. The languages for the expression of procedures have been much
improved

(more successes here)

	When we have succeeded in understanding the intelligence well enough
then we should be able to make programs that equal and exceed human intellectual
performance and which could improve themselves further.  How far are we from
that?  It is difficult to say.  On the one hand, conceptual breakthroughs
are required.  No-one could outline a development program that would achieve
human level of intelligence in a fixed number of years.  On the other hand,
maybe only one conceptual breakthrough like that which produced the theory of
relativity is required.

	What can we guarantee to achieve in a fixed time.  In my opinion,
rather little.  We cannot guarantee to produce master level chess in five
or ten or twenty years, because this may require the ability to represent
chess ideas like those mentioned above.

	There were over-optimistic predictions.  In my opinion, these predictions
were due to incorrect ideas about what intellectual mechanisms were required
to solve certain problems.