|Visions of the Future|
Some computer prophets say that the age of the thinking machine is near. They see a world in which machines could be gods of information. John Gilbert investigates the claims.
ARTIFICIAL Intelligence, AI, has become one of the most fascinating areas of interest to the computer fraternity.
An underlying interest in the production of 'intelligent' machines has always been evident in the computer world but it has not been until the last six months or so that the subject has caught the headlines in computer magazines and books. The reason for that is the difficulty in writing about a subject which has evolved no terms of reference at a simple level.
One of the books which has tried to do just that, and failed to some extent, is The Fifth Generation, by Edward Feigenbaum and Pamela McCorduck. The attitude of the American authors shows that they have no doubt that fifth generation artificial intelligence within a machine environment is possible. We are currently in the middle of the third generation of microcomputers, which involves integrated circuits. The first and second generations have evolved from gas heated valves and transistor technology. The fourth uses very large scale integrations, VLSI, and fifth generation will show dramatic leaps not only in hardware but in software.
The authors state that artificially intelligent machines will be able to manipulate information and come to conclusions, or reason, on the basis of that data. Unfortunately, their idea of AI seems to be confined to a machine which can amass huge amounts of data, using enormous memory banks, and offer that information to another user in any format required. Many expert systems can do that already, and they would in no way be termed intelligent.
Unfortunately, the concept of consciousness, though touched on briefly, is not dealt with in anything approaching enough depth. Some readers will, as a result, feel that the authors have not produced an adequate formulation of the different definitions of AI and have only put across their own views which, they seem to think, are unchallengeable.
Once the problems of terms of reference have been established and cleared the authors then launch into a look at the Japanese innovations in software and hardware techniques. They see information as the next great commodity on the world market and explain that the Japanese with their KIPS, Knowledge Information Processing Systems, are on the way to becoming the next great superpower which could have domination over the USA and USSR, at least in economic terms.
The fifth generation of computers, unlike the last four, will be one in which software, and not hardware, is most important. Feigenbaum says that 'significant levels of innovation in software techniques will have to be achieved before the fifth generation can be implemented. He then goes on to say that the Japanese are close to such breakthroughs and that they will have a dangerous monopoly on such new techniques unless other countries, for example the United States, do something to safeguard their interests. Such a viewpoint is slightly naive and shows the authors to be suffering from a highly developed sense of information paranoia.
Feigenbaum does, however, redeem himself by admitting that the Japanese need a lead in the new information revolution. He comments that 'Japan's survival as a nation is at stake' unless new quantum leaps in technology are made by that country. What he does not overtly say, however, is that his pessimistic viewpoint about the American lack of interest in the subject is fuelled by the fear that what could happen to the Japanese if they fail could also happen to the United States.
The Fifth Generation, despite its technofear style, is an interesting and digestible book which will appeal to computer historians and prophets alike. Sir Clive Sinclair thinks that it is 'essential reading for anyone concerned with computers' and what greater endorsement could you get than that?
On a more practical note Exploring Artificial Intelligence on Your Microcomputer by Tim Hartnell investigates the traditional idea of artificial intelligence. Unfortunately the book might have been better titled 'How to write strategy games or programs which will talk back to you'.
Hartnell's overview of the field of artificial intelligence is informed and concise. It does not side-step the issue but equally it does not go to much trouble to evaluate the terms of reference that were mentioned earlier. To he fair the book is not just another tome of listings. The examples are broken down so that the reader gets a few lines at a time together with a paragraph of explanation. Many of the listings are then reproduced as a whole, although it is not clear if that is to help the reader or whether it is just to fill space.
All the programs use conventional programming techniques and if The Fifth Generation terms of reference were used the book could not be described as a text about artificial intelligence.
To be fair, the author does deal with the subject of Syllogy, an area which figures greatly in the AI debate along with information processing. The area covers forms of deductive reasoning in the style 'if a and b are true then c is also true'. Computers can deal with such relational arguments and can also make the connections between relationships. Hartnell includes a program to show how it is done.
Although Exploring Artificial Intelligence is more a book for those casually interested in the thinking machine it provides a view of the subject which should appeal to many people who want to improve their programming skills. It takes the reader to an advanced level but, because of the limitations of Basic, does not even touch the realm of what would now be termed artificial intelligence.
The same can be said of Artificial Intelligence on the Spectrum Computer, by Keith and Steven Brain. The book contains little information that could not be acquired from good texts on adventure gaming or data processing.
Subjects such as entering English sentences and getting sensible replies from the computer are covered together with short examples which are not particularly imaginative. One good point about the book is that program listings are backed up with flowcharts which detail the techniques which have been used to create the revolutionary new program. That will better enable the reader to adapt techniques to specific programming needs rather than have to wade through the programs.
The authors have, like Hartnell, taken a simplistic view of AI. They see it as a method of communicating with computers and in turn receiving a coherent reply. They also touch on matters such as recognising shapes but make no attempt to distinguish the real points of issue in that area of AI.
The chapter on shape recognition deals only with input from the keyboard and not with senses such as touch, sight, and sound recognition. All those areas are under investigation by computer scientists but none of them are mentioned in detail by the Brains.
If other Sunshine books can include information about setting up hardware for simplified sensor devices then surely the Brains' book could make at least some effort to look at the subject in depth without shying away with a few examples which are old hat to most programmers.
Build Your Own Expert System by Chris Naylor on the other hand, is a welcome relief from the simplistic views of AI given by some authors when dealing with the subject.
The book is about building relational databases which can be questioned in order to obtain specific information. Naylor introduces the random element which occurs in the thought processes of most human beings and which probes for new areas of knowledge. For instance, the author gives an example of a database which will predict what the weather will be like the day after the prediction was made. If it is rainy today and has been raining all week, the chances are that it will rain tomorrow. With that supposition and a knowledge of cloud formations the computer might predict that it will continue to rain tomorrow. If the prediction is wrong the method used by the machine will he adjusted. That might be by providing better knowledge of weather movements or lengthening the odds of certain weather patterns occurring. It is a hit and miss business but it is a better display of AI than any of the books reviewed earlier could muster.
Naylor's book is a must for computer users clamouring for more information about AI or wanting to do something useful with their Spectrums. It is one of the few books worth reading on the subject and, with Naylor's unpretentious skill as a writer, it is certainly one of the most readable.
|The Fifth Generation, Pan Books, £2.95.|
|Exploring Artificial Intelligence on Your Microcomputer, Interface Publications, £4.95.|
|Artificial Intelligence on the Spectrum, Sunshine Books, £6.95.|
|Build Your Own Expert System, Sigma Technical Press.|
Theo Wood examines the case against AI
PUBLISHED earlier this year for the first time in Britain Computer Power and Human Reason, From Judgement to Calculation was originally published in 1976, but contains much that is relevant today.The author, Joseph Weizenbaum, must be considered a heavyweight by anybody's standards; currently Professor of Computer Science at MIT - Massachusetts Institute of Technology - he has had a career in computers since 1950.
Weizenbaum's book has received a fair amount of media attention, due mainly to the inclusion of a new preface to the 1984 edition. That contained a blistering attack on the computer games fever which has spread across America and Europe in recent years. The main basis for his attack centres on the process of psychic numbing of the individual who plays those games, similar to that which takes place for man to wage modern warfare. It was he also who described the computer junkie way back in 1976.
Although important, those are, however, only peripheral to the central concern of the book, which is a systematic description of how computers work and a similarly systematic attack on the work done by Artificial Intelligence researchers. With books such as The Fifth Generation published and Sir Clive Sinclair talking about the wonders of the new technology, it is still worth reading Weizenbaum to place those thoughts within the context of human values.
Weizenbaum doubts that Artificial Intelligence can be anything close to human intelligence, and portrays with scorn such statements made by eminent scholars such as Professor John Macarthy, then head of Stanford University's AI laboratory, who said, "The only reason we have not yet succeeded in formalising every aspect of the real world is that we have been lacking a sufficiently powerful logical calculus. I am currently working on that problem."
For Weizenbaum the idea that all human activity and thought can he reduced to formal equations capable of being computerised is beyond belief. Having worked on a natural language program, ELIZA, which allowed conversation between the user and the computer as to the user's mental state, Weizenbaum was horrified to find that serious attention was being given to it. Specialists in the psychiatric world were considering the use of such programs in place of human therapists.
Weizenbaum's book is a bellow of anguish from the heart of the computer establishment and, as such, requires some serious attention. However, it is an academic's book in that the main chapters concerned with AI are high-level intellectual infighting. For a more general purpose examination of the social implications of computerisation, Michael Shallis's book The Silicon Idol is more suitable.
Depending on which viewpoint it is judged from, the book can either be seen as the work of a Don Quixote tilting at windmills, or of a man trying to communicate the deepest reservations about the introduction of microtechnology into the world of work and human relations. Shallis owes a lot to Weizenbaum for covering the same ground but from a slightly different standpoint. The descriptions of how a computer works are not so comprehensive but, on the other hand, are much more accessible by the reader who may have no previous knowledge of the subject.
For Shallis the suggestion that computers can come anywhere near having what are essentially human characteristics is debasing the human condition. He is particularly strong on the history of computers and intertwined with this the history of man's attitude to service and technology and the power of technology to transform society.
In a world climate of increasing speed of technological change The Silicon Idol is ideal for the general reader to take stock of where that change will have maximum impact, as well as its social consequences. Shallis is sceptical of the shining brave new world of high technology, where most of the population do little or no work as we know it in its present form. He reaches the crux of the dilemma when he states that new technology is usually used for economic reasons, replacing humans in both the manufacturing and the service sector, leaving fewer and fewer people to work in factories and offices. Neither does that move to automation provide an alternative to work which is considered dull and repetitive.
Michael Shallis is an unashamed Luddite and is appropriately pessimistic about the future of work and the social disruption that might cause. He offers no solutions to the problem, merely setting it before the reader in what might be considered a sensationalist and extreme form. In so doing he provides enough fuel for discussion between here and Armageddon.
There seems to be a general consensus that we are in an age of transition, and the two books discussed are important in that they raise issues central to that. Are there activities which computers ought not to be part of? Are computers going to be 'more intelligent' than humans? How are masses of people going to react to enforced leisure/redundancy?
The answers to the first two involve philosophical enquiry, depending on the definition of human intelligence - are we as humans simple input/output devices whose though processes can be reduced to a logical calculus? The third question depends on the way we as a society organise ourselves in the next twenty years. The debate has already begun - it is too important to he ignored.
|Computer Power and Human Reason : From Judgement to Calculation.|
Joseph Weizenbaum, Pelican Books, £2.95.
|The Silicon Idol: The Micro Revolution and Its Social Implications.|
Michael Shallis, Oxford University Press, £8.95.
Sir Clive Sinclair sees a rosy robotic future. Alexander Macphee, systems analyst, disagrees
SIR CLIVE Sinclair has speculated on a vision of the world of the future, a vision in which, the editorial comment remarked, there was little to cause a raising of the eyebrows. I do not share the editor's eyebrows.
The central ideas in Sir Clive's remarks were firstly that current developments in Computer Science are leading to superintelligent silicon-based life forms and secondly that human problems can be solved by the application of advanced technology. Neither of those assertions is unchallengeable.
The speculation that computers will develop as living beings is a revival of an idea first advanced in the 1950s in Artificial Intelligence studies, often referred to as 'Strong AI'. It is not a new idea, but no more justifiable now than it was then. What is remarkable about those speculations is the number of uncertain, unself-critical and seriously flawed assumptions made, and the seductive use of vocabulary in foreign contexts.
The most primitive of those assumptions is that the way in which a computer processes information is in any way similar to or cognate with the way in which a human being - or indeed any other animal - processes information, or even that information processing is the sole or most important attribute of sentience.
There is no reason whatsoever to suppose that the way in which human brains work at the neurophysiological level is in any way similar to the workings of a computer at the electronic bit level. Yet it has always been a human failing to attempt to describe the workings and functions of the human brain in the vocabulary of current technology; formerly as a set of compartments in a Victorian office communicating by memos, later, at the turn of the century, as a telephone exchange switching messages, and currently as a bit-twiddling and byte-shuffling device.
None of these is either valid or instructive, although the computer analogy is clearly more seductive, so much so that as a result of idiot anthropomorphisation people have come to think of human brains as operating like a computer 'brain', a notion which has been reinforced by the use of computing vocabulary to describe human brains. Thus it is an easy step for Sir Clive to turn a computer into a real functioning brain. All you have to do is
"... render it intelligent by loading the proper software ..."
Under that view, thinking is no more than a set of rule-based operations, and the world is no more than a series of puzzles to be solved by primitive and computable rules.
Understanding is not derived from formal rules alone. John Searle demonstrated that in his Chinese Room analogy in The Behavioural and Brain Sciences. A man with no knowledge of Chinese is locked in a room. He has a script in Chinese - which he doesn't understand - together with a set of formal rules in English - which he does - which enable him, when given lists of Chinese symbols, to construct other lists of Chinese symbols, which he can pass back out of the room. Unknown to him, his Chinese script is a story, the list of symbols he is given is a list of questions, and the list of symbols he constructs are really answers to the questions about the story, which are indeed understood by native Chinese speakers. In what sense can that mechanical, rule-based symbol manipulation be called understanding?
Those areas that are claimed as successes in the field - the so-called 'expert systems' - owe their success to the fact that the conditions under which they operate are strictly defined, as are the problems which they are designed to solve. 'Expert' is, of course, another of those seductive words. Another example often quoted is the success of some chess programs. Here the restriction of operation to a rule-based and context-free world is signally clear.
Sir Clive also confuses knowledge with data when he proposes that "acquired knowledge of a man" could be given to a computer by "the transfer of data from human to machine mind". Yet those two are not synonymous, the implication being that knowledge is no more that a large database of context-free atomic facts, by which definition human knowledge therefore becomes reducible to logical formalism.
Sir Clive also speculates that these intelligent computers will evolve as silicon-based life forms, able to reproduce themselves. I wish I could be generous but that seems to me to be staggeringly stupid. Again the seductive use of vocabulary with human connotations continues with use of words like "see", "feel", "sensory devices". Optical or physical devices may indeed be used to supply information to a computing device, but the significance of those as sensory devices is a matter of interpretation. The ballcock in a water tank performs a sensory function, but I do not attribute to it the phenomenon of sensation.
It may be argued, however, that if the machine appears to show intelligence we may never know the difference, the Turing test frequently being invoked. The Turing test, however, is not a test to divine the existence of intelligence, rather it is a measure of the level at which we can be fooled into believing that the responses we detect may be coming from an intelligent source.
Suppose, however, we allow that machines can think in a way cognate with or superior to human beings. That then raises moral dilemmas of massive proportions. If we have truly created intelligent life then presumably this life form is entitled to what we regard as fundamental human rights. With the concept of life comes the concept of death and killing. The question is by no means trivial, for Sir Clive offers those "superintelligent" life forms as "menial slaves" to human beings. In what sense could beings of such superior intelligence be treated as menial slaves?
The second offering of Sir Clive's speech is of high technology as solutions to human problems. Here he is on equally shaky ground. He is not unique in that, for it is common to propose the salvation of the world by throwing technological solutions at human problems. This is doomed to failure because the problems themselves are not technological, but human and social.
What are the major problems faced by humanity in the twentieth century? And what is the role of advanced technology in solving them? They are not difficult to identify: hunger, poverty, war, unemployment, health, freedom, education, housing. The size of the list is as distressing as its content.
None of those problems is a merely technological problem; indeed, for many of them we know what the solutions are. What we lack is the collective human willpower and trust to take decisions to implement the solutions to end these problems. The immense effort and logistic enterprise used to transport the men, machinery and materials of warfare and mass public death around the world might equally be used to feed and clothe the sick and the hungry. We have massive food production capabilities, but most of the world is hungry. It is the human application to the solution that is missing, not the solution itself.
Technological solutions show equally great poverty of inventiveness in areas of unemployment. Sir Clive at least acknowledges the unemployment problems - though not the consequential problems - created by advanced technology. He believes those are only temporary; he does not say why. He believes those will widen horizons, he does not say how. He says "... goods are still needed but ... technological change will remove virtually all employment".
Who will buy the goods? And with what? In what way will the wealth (if any) created by this new Technological Revolution be distributed among those who, unemployed, have no part in it? Who will own the means of wealth production?
Sir Clive believes rather that our lives will parallel the lives of the Freemen of Athens. But the world in the twentieth century is not a City-State, and not all the inhabitants of Athens were Freemen. Poverty and misery were not absent in Periclean Athens, nor did educating children to an appreciation of the finer things in life do anything to prevent the wars which sapped, eroded and finally reduced Periclean Athens.
Where, however, human solutions to human problems seem difficult, it is not uncommon for the problem to be simply recast as a technological problem, to which of course technological solutions abound. It is just such a technological solution which Sir Clive offers to the problem of crime. He ignores the relationship between crime and the social problems of unemployment, deprivation, poverty, poor housing; instead, the 'solution' is to implant in the 'criminal' a device linked to a computer.
He dismisses fears of an Orwellian society by saying that we could offer a choice to miscreants. This dismissal fills me with fear and alarm. It presupposes that any State given such control would offer an alternative, it applies a narrow definition of what people would recognise as a miscreant, it presupposes that the definition of a miscreant would not be 'adjusted' to include anyone who opposed the State in any way.
That is not mere speculation on my part: the technology of surveillance has already been widely and illegally applied in this country to people who were not criminal in any sense of the word, as exemplified in a recent case. I am not sure if what Sir Clive demonstrates is innocence, naivete or just plain silliness, but our liberty should rest on stronger foundations than unproved declarations of benevolence from the State.
Even the problems of loneliness in old age are cast by Sir Clive as technological problems for which, of course, the solution is a quick-witted robot, not humanity, caring, love, compassion, or understanding. Those are qualities which, if they have them, people have from an understanding of human problems, human ills, human sufferings - in short, by being a human being sharing in those fears and difficulties with which so much of life is fraught. It is not clear how those attributes or properties could be shared by a robot.
A more persuasive application, perhaps, is in the field of education. Sir Clive argues that we can - and presumably therefore should - have our children taught by machines. Yet here again his arguments are faulty. Education does not consist of instilling a list of facts. It is, above all, a social process, not a mechanistic process. Apart from set topics dealt with in the classroom there is also the 'hidden curriculum' of learning - what it is to be human - by being in the company of, and interacting with, other human beings.
Robots may, however, be convenient for some in that they may never have the human attribute of accountability for their errors, a feature that is often found useful in computers when it comes time to lay blame at someone's door for some error.
What is really distressing about Sir Clive's views, however, is not so much that they seek to describe computers or robots as a special kind of Man, but rather they describe Man as no more than a particular kind of robot. If that is a vision of the real future, I hope it arrives with the punctuality for which Sinclair is renowned.
Sir Clive commented on this article in turn during an interview published in issue 35.