There are, however, glimmers of hope on the horizon, a way forward. Recent writers, such as Bolter in his Writing Space, have argued that computing is a way of writing. This idea, not surprisingly, has its origins in hypertext, the use of computers to create interactive, non-linear text systems, but it can be generalised to all computing. ai has failed to recognise or acknowledge its real origins and hence grasp its future. I want to claim that it applies to ai more than any other aspect of computing- ai is in fact an extension of writing - a grammatology of reasoning.
My own search in this area began many years ago when I was working within the traditional ai paradigm. Concerns about both the technical possibility and the desirability of such a project soon began to arise. I happened upon an article by a major ai researcher, Mark Stefik, in which he argued that ai should change its goals and try to create a Knowledge Medium, a means of communicating knowledge between people. The revised aim being to create intelligent people rather than intelligent machines. He used a number of metaphors to state his case, but one that he missed was the very one he was using to communicate with his audience - the written word. This observation prompted an investigation into writing, which threw up a number of interesting parallels between ai and writing.
Yesterday
Artificial intelligence is a curious beast. Many a popular text on the subject will begin with an introductory chapter on the sources of man's desire to create a being in his image. We read of the splendid robots of the Illiad, the Golem created by the Rabbi of Prague, and Mary Shelley's Frankenstein. It is strange that a subject seeking to be a science and based on logic should appeal to such mythical beginnings.
The history of writing, too, is full of stories that ascribe its invention to the divine, a gift from the gods. This belief is particularly strong in cultures in which writing was restricted to a special class or caste of priests. The ancient Near East, where normally only the priest class could write, abounds with mythical stories about the creation of the first writing. In Greece, however, where writing was widespread - such myths are lacking. They knew where it came from and had no need for such stories. Like the ancient Greeks, ai should acknowledge its true roots in the technologies of the word: writing and print. It may be less exciting in the first instance, but the potential is far greater. One only needs look at the significance and value of the written word to our culture. It is not for nothing that it has been called The Great Invention.
Even the claim to independent intelligence has remarkable precursors. Among non-literate cultures, writing created a great deal of astonishment. For them, as Gelb has put it a book is a living being which can speak. There are many stories of autonomous intelligence being attributed to the text. In one case a native messenger refused to carry a written message, as he was afraid it would speak to him. A native Australian stole some tobacco from a package, which he was carrying with an accompanying letter, and was completely astonished that the recipient caught him, despite having hidden the letter in a tree trunk during the theft so that it could not see him.
No one today would say that a text was alive, except in the poetic sense. I strongly suspect that in few years we, too, will look upon the reaction to ai programs with similar amusement. But the desire to anthropomorphise is strong (and profitable) as Disney discovered, and may take some time to decline.
Of course, the true beginnings of writing have little do with any of the mythical beliefs, or the creation of a new intelligence. Recent work by Schmandt-Besserat indicates that we should look to counting stones that were used for trading purposes. That is, initially the push for the development of writing was from accountancy. As Bottero (in Goody) has remarked, Mesopotamian civilisation was quickly caught up in a widespread economy, which made necessary the meticulous control of infinite movements, infinitely complicated, of the goods produced and circulated. It was to accomplish this task that writing was developed: indeed for several centuries, this was virtually its only use.
Likewise, ai, through its roots in computing history, is the son of the accounting and banking software so pervasive today. As writing sought through myth to elevate itself beyond the mundane, although this was actually achieved through literature, so ai must abandon its own myths and accept that it is a form of writing. Recently it has become noticeable that as the creation of interesting conversational partners has proved so difficult, ai has ironically returned to accounting - but without looking to its potential future, as a writing, that this implies. Consider the comment of the director of one ai company who considers that the big question in knowledge based systems (the primary form of ai system) is - can the customer develop the applications he needs to realise more efficient operation of his business? Well, if that is still the big question, no wonder the interest in ai is dwindling. Can you imagine anyone saying that the big issue in literary theory is the content of bank statements?
Writing and ai have also shared denunciations. Despite the high status it holds today, writing has not always been considered as benevolent. Writing, Plato has Socrates saying, is inhuman, pretending to establish external to the human that which can only really exist within it. It will destroy memory, making one rely upon external means and thus weaken the mind. Similar complaints followed the development of print. Hieronimo Sqarfiafico, as early as 1477, argued that the abundance of books makes men less studious, presumably, as one would not have to learn everything but could turn to references. ai too has had its share of critics, often on the grounds that if we can get the machine to give us the answers, our own intellects will diminish. If this technology is going to take over much of our intellectual work, what remains? The critics seem to have a case. The trade magazine Electronics has put the case for ai thus: the computerised expert or advisor is always alert, is never under the weather or temperamental, can work 24 hours a day, seven days a week, and cannot accept a better offer from a business competitor. Little wonder that the public get concerned.
But Plato's worries were unfounded. In fact, writing engenders understanding by virtue of the possibility of exploration and re-examination that it affords. As Goody has remarked, when an utterance is put in writing it can be inspected in much greater detail, its parts as well as its whole, backwards as well as forwards, out of context as well as in its setting; in other words it can be subjected to quite a different kind of scrutiny and critique than is possible with a purely verbal communication. It is easier to recognise contradictions in the word set down, rather than in the constant flow of conversation. The invention of print continued this process, allowing the reader to consult and compare many different texts.
Similarly, one of the unexplored side effects of ai comes from that area known as expert systems. In these, one tries to elicit the reasoning employed by an expert in some subject. This is then formalised and implemented within the computer. It has been observed that this very process aids the expert's understanding of their own reasoning, leading to a refinement of that knowledge. Some years ago I worked on a conventional ai project, designing a system to simulate the reasoning of an economist. Certainly, I would say that the most significant result that came from the project was not a working system, but myself having an understanding of economics, and the economist too found the process of trying to 'externalise' his reasoning helped him in his subsequent work. More concretely, Musen and his colleagues, while designing a system for the handling of cancer treatment, found many logical inconsistencies in the original printed material, materials that have since been improved.
Today
Maybe these parallels are more than coincidence, and ai is a form of writing. The techniques of ai have the potential to be employed as a Typography of Thought, allowing an author to 'write down' their thoughts. Just as conventional writing expresses what can be said, so this new form of writing can express what can be thought. It is dynamic and runs on a computer but, unlike the pretences of ai, I make no claim that this new system is thought, anymore than text is voice.
Thus, a computer procedure for metaphorical reasoning can be considered not as a simulation for human metaphorical reasoning, but as a sTimulation for it. The written letter b does not simulate the sound but, through convention, stimulates us into saying it. We have conventionalised symbols for certain families of sound, and it not difficult to image creating an alphabet of reasoning. This cognitive alphabet would allow the expression of many forms of reasoning that ai has uncovered and implemented: causal, metaphorical and diagnostic to name a few. These can be assembled by an author to express an argument or narrative.
This new form of writing is unlike conventional writing in that it will be dynamic, allowing the reader to explore the implications of the line of thought. Like Derrida';s Machine Programmatrice, we can read the range of possible meanings within the text, seizing the text's various resources and bend them to our own purposes. This is possible because the computer allows the reader to edit and run the text in a way that the static manuscript or printed text does not easily permit. The question-answer tradition within ai is subverted and employed to allow an author to create a text. The author asks questions of his own models to create the desired appropriate response. We thus have a simple implementation of the notion of the problematological approach to text proposed by Meyer. The reader is now in a position to edit the questions and can generate (as the text can be run like a computer program) their own answers, addressing their own concerns.
The value of the comparison with writing is that it gives us some guidelines for designing such a cognitive alphabet and using it to create Running Texts (so called because they are texts that can be run). Specifically, in the first instance, significant advantage was gained when the written symbols came to be associated with the existing practice - speech. Havelock notes the original problem with a writing that is ideographic: The shapes were used to symbolise (...) mental acts directly. They went straight to psychological processes in the brain. In a sense they were too ambitious. They were not content to deal solely with phonetic (...) finally came the systems that sought only this limited aim, of copying linguistic noises. The attempt to bypass speech and directly represent the world was attempted in seventeenth century Europe. Some of the best minds worked on the Universal Language Project (described in detail by Slaughter) which, despite much effort, failed. It is noticeable that traditional ai continues this approach, designing knowledge representation languages that are 'deeper'; than external language expression. The lesson is surely that just as writing used speech as a crutch, so any Running Text system should represent existing practice - conventional text - to carry it through the initial process of development. The output from the executable processes should be tied to, and interpret, everyday text. Once conventionalized, just as conventional writing has left the bosom of 'mother speech', Running Texts will emerge with their own expressive form - a poetics of reasoning.
Tomorrow
Should such Running Texts become widespread, this conversion of ai to a communicative medium would raise a new set of questions. Conventionally, arguments about the social impact of ai are predicated on the goal of creating intelligent machines. Should such systems have rights? Could they be held responsible for any mistakes? The new questions follow from the notion of ai as one of many media we might use, a knowledge medium.
It is certainly the case that our media seem to infect our minds. A film student of Salomon relates how their daydreaming is influenced by movies (...) I have observed third person narration, flashbacks, zooms, slow-motion emphasis of action, audience viewing, re-takes, 'voice of conscience', multipersonality dialogue, background music (...) I fear that there is very little original style to my daydreaming. It is all influenced by celluloid.
One might say that the computer is somehow different. Being a calculating device, could a computer really get into our minds? But even the simple abacus exhibits such effects. It is a technology of the intellect. It has been observed that certain users of the abacus exhibit finger movements when doing mental arithmetic. When prevented from moving their fingers, performance in this mental task fell significantly. They seem to be in the process of internalizing the functionality of the abacus into their cognitive apparatus.
There is also little disagreement that writing and print have changed both our psychology and our culture, although we may debate details. Ong has gone so far as to argue that literate people are beings whose thought processes grow not only from their own natural powers, but also from their restructuring through the technologies of writing. Indeed, for him, all major advances in consciousness depend on technological transformations of the word. Bolter has also made the point that writing gives the writer an awareness of himself that is beyond that given by speech; ''the technology of writing is customarily regarded as the creation of the human mind, possibly its greatest creation. In fact, it is the other way around: the mind is the creation of writing.
It is an interesting thought that by viewing ai as a medium to be used by people, ai may achieve that final goal which has proved so illusive. If a technology can truly infect those who use it frequently, artificial intelligence could arise. We would be that artificial intelligence, our cognitive apparatus restructured through the use of this technology.
Other impacts of writing were more social than individual. The concretisation of language through writing and printing produce consequences that were inconceivable with a purely verbal form of language. For example, once printers knew that each page on a large print run would be the same, they began to produce indexes. Indeed having a large library as printing facilitated made its almost imperative. But suddenly ideas became objects that could be located spatially in the text (within the volume), not a loose bundle of sounds lost forever. As locatable entities, authors sought copyright over their words, and citation became standard. Equally, in Running Texts ways-of-thinking become like printed words; we can inspect them, index them, give them to others. The potential of taking such an approach is suddenly problematic. What might an index of ways-of-thinking look like? How would I cite an extract from a Running Text? How would an author claim copyright over such material, allowing for its infinitely editable nature?
We should also be concerned with the pressure to standardize. Spellings and grammatical forms converged due to print. Will ai do the same to 'thought'? It is not difficult to imagine a dictionary of thought patterns, citing their first use by the great and the good. The lesson from both writing and music notation is that, initially at least, the original practice - the oral tradition - becomes reduced in status. The established authorities use standardised notation to encourage standardised performance. It would be a tragedy to see the primacy of individual thought diminished.
Post-Script
Compare the public perceptions of the two technologies under discussion, ai and writing, as commonly seen through film. In Kubrick's 2001 the intelligent computer Hal is the source of all that goes wrong. In contrast, in the film version of Fahrenheit 451 books are seen as containers of all that is great in human culture - something for dictators to destroy and for liberators to preserve.
The public perception of what ai tries to undertake suffers from this problem. But the solution is not, as one all too often sees, to try change of interface colour or improve the verbal output of computer systems to make them more personable. To do so is like arguing that the best way to encourage greater interest in nature is to improve the typography of signposts on public footpaths. No, a leap of perception of the natural world had to occur, equally the ai community and its potential users must view ai in a new light. It is this prospect that using ai as a communicative and expressive medium provides.
Too often sci-fi presents us with images of a dead society survived only by its computing machine, the assumption (either explicit or implicit) being that the computers contributed to the decline. But if we find a ruined city with written records we do not (unlike Plato) think that the speakers of this city were in some way destroyed by the documents. On the contrary, we think that their speech was more appreciated, amplified by their writing. So Running Texts, the knowledge medium, might amplify our thoughts into new directions, facilitating the human dialogue between authors and readers that is a prerequisite for rational human progress. The alternative, the classic ai position of mythic intelligent machines can only contribute to a concentration of authority in those who create such systems - systems that give answers - over those who are supposed to use them.
We must return authority, authorship, to the human who creates such Texts, and who can then be called to account. The question now becomes not is this computer behaving intelligently? any more than we would ask is this book behaving intelligently? What one should rightly be asking is ''is the author behaving intelligently?
''
Only yesterday I watched a television preview of the new film based on Stephen Hawking's A Brief History of Time. Unable to speak due to motor-neurone disease, Professor Hawking must write his words into a computer and the machine - in a voice that seems to come from science fiction - speaks to us. A typical ai project, but the author of the words is a human, a remarkable human, and they carry his humanity with them. What machine could have so much to say?
References:
- David Bolter Writing Space: the Computer, Hypertext, and the History of Writing, Lawrence Erlbaum Associates, 1991
- J. Gelb The Study of Writing University of Chicago Press, 1963
- Jack Goody The Logic of Writing and the Organisation of Society, Cambridge University Press, 1986
- E. Havelock The Literate Revolution in Greece and its Cultural Consequences, Princeton University Press, 1982
- Michel Meyer Meaning and Reading Pragmatica & Beyond IV: 3, John Benjamins Pub, 1983
- Musen et al Knowledge Engineering for a Clinical Trial Advice System: Uncovering Errors in Protocol Specification Report Number ksl-85-51, Medical Computer Science Group, Knowledge Systems Laboratory, Stanford University, 1985
- Walter Ong 'Interfaces of the Word Cornell&' University Press, 1977
- Gavriel Salomon 'The Use of Visual Media in the Service of Enriching Mental Thought Processes', Media, Knowledge and Power, Boyd-Barrett and Braham (eds). Croom Helm, 1987, pp. 251-265
- Denise Schmandt-Besserat ‘From Tokens to tablets: A Re-evaluation of the So-called ‘Numerical tablets’’, Visible Language, xv, 4, Autumn 1981, pp. 321-344
- Mark Stefik ‘The Next Knowledge Medium’, The ai Magazine, Vol. 7, No. 1. Spring 1986