POSTAMBLE: TOWARDS THE ULTIMATE PERSONAL COMPUTER

"Good heavens, are people really working on artificial intelligence?"
Isaac Asimov, Preface to "The Robots are Coming"

When Lawrence R. Zeitlin asked his class of graduate students at the City University, New York, what use they would make of their own small portable computer, the consensus of serious replies concentrated on information storage and retrieval tasks. The role assigned to the strictly mathematical processing power of such a machine was quite limited, involving relatively simple operations, and it ranked a poor second to artificial memory activities. Science, in Francis Bacon's words, is organized knowledge. Once the first requirement of a good laboratory computer is satisfied - that of providing accessible processing power (p. 10) - it would seem natural for the scientist also to concentrate on the ability of his own personal machine to handle information, the "knowledge" required by his science. The ability to organize a sizeable and personal store of knowledge, to retrieve items from it at will using keywords and contexts, to develop new relationships between selected pieces of information and to do all this relatively painlessly, surely comes close to a scientist's dream of Utopia. Certainly in the minds of those New York students, presumably facing examinations at some stage, information storage and retrieval represented a major problem.

Allowing, then, that the most useful aspect of a scientist's personal computer is its capacity for information handling and that the scientist spends a good deal of his training and working life wrestling with problems of information handling, one very pertinent question presents itself: how much information needs to be stored to make such a capability useful? Let us take the Chemical Society's Annual Reports on the Progress of Chemistry as an example. This is a series of reviews which deals with highlights of the past year's work as reported in "the literature" and covers some 35 chapters (or major topic divisions) in 1300 pages. With an average of 45 lines to the page, including references, and with each line containing 80 character positions, that book (or those books, as is usually the case) would contain an approximate total of 4.5 million characters, making some allowance for coding of chemical structures and diagrams. For the serious research chemist, a reasonable current information set might include coverage of the previous decade through these reports. Since the research chemist is by nature a well read person, and by training an inquisitive one, he would probably want to have at his disposal an equivalent amount of information from a related discipline, such as physics, biochemistry, or medicine, another equivalent portion from more general science - perhaps a few years' back copies of popular magazines such as Nature, New Scientist, Science, and so on, and finally a fourth equivalent portion of information from general reference texts, tables of physical and chemical constants and the like. In all, this fairly modest library would contain about 52000 pages or approximately 164 million characters which, coded as seven bits per character, is l.15 gigabits of information.

The present (1977) peak of the state of the art, high density semiconductor memory packaging, represented by IIL circuits (Section 4.2.5), requires a volume of about 6 mils cubed (6 x 109 cubic inches) per bit of information stored. Neglecting packaging and connection requirements and support structures, this is an information density of 1.67 x 108 bits per cubic inch. Thus the whole of our selected library could be contained in 6.9 cubic inches of present state of the art storage. The arithmetic and logical processing requirements of a fairly capable computer would not add significantly to the overall size, since this would take up space equivalent to less than 0.1% of the total space contemplated for the information storage. Allowing a further 100% increase in volume for packaging and another similar amount for l/O and power connections, the 20 cubic inches then arrived at is still only two-thirds the size of the current paperback Microcomputer Handbook (1976-1977 edition) published by DEC - an eminently portable volume (20 cubic inches of silicon and glass, density 2.4 g/ml, weighing 1.8 lbs).

With information storage requirements compressed to this sort of size, it is hardly likely that the user would wish to carry around a Teletype or similar keyboard data entry device to communicate with it. In addition to the size restrictions imposed by such a device (even the smallest and lightest keyboard must be of such a size that its use does not require needlepoint fingernails„ the downfall of many a good miniature calculator), there are additional disadvantages advantages in the mechanical skill required to operate a keyboard with any efficiency, and the limit on ultimate speed. By far the most satisfactory method of communication would be the spoken word. A great deal of research effort has been invested in hardware and software systems which recognize and obey a limited number of spoken commands. Most recognition devices now construct their own "dictionaries" with tolerance to dialect and intonation. At the present time hardware for one such system is contained on two printed circuit boards 10 inches by 8 and it seems reasonable that a custom-built chip-based version could reduce that size by an order of magnitude, again with currently available technology. For output, voice synthesizers have been in operation for some time (one or two years) in a number of systems, some commercially available, and some small enough to be contained in calculators of the size postulated for our personal computer library.For a voice I/O scheme for our personal machine, perhaps operating with rather restricted commands at input (say, a vocabulary of 256 words - the average level of a two-year-old child), but with a considerable output capability, based on synthesis of words from the stored information, no conceptual changes to present technology are required, just a little more effort in miniaturization. Acceptance of this speech based system is actually quite helpful, since our library of physics and chemistry could be held in phoneme (speech element) form instead of character form. This cuts storage requirements by half, and slims down our pocket computer by the same factor. So from a rack mounted, Teletype driven monster of a minicomputer, we have derived a truly pocket-sized machine with full computational power, which can carry as much information as about 65 major textbooks, accept spoken commands, and reply to them - all this with current technology and well proved concepts, and only a little development effort.

If the implications are far-reaching, lying well beyond the simple aspects of a dictating machine with "intelligence" as well as "memory", consider the projection of current technological trends over the next decade or so as they relate to this system. Circuit modules about the size of a common operational amplifier (1 x 1.25 inches) and with a memory-packing density 22.5 times that of llL (92 000 bits per chip of magnetic bubble memory, Section 2.2.2(c)) have recently become commercially available because of advances in memory technology. These devices are being designed into systems now, and although they require more connection and support space on a chip for chip basis than the semiconductors at the present, the packing density of the next generation of devices is close to three times that of the first generation, and must continue to rise. Exploiting this trend for our personal computer (again from known technology) brings us down to a size estimate of about 0.17 cubic inches, or somewhat less than 3 ml. Zeitlin points out with considerable insight that this is about the volume of one large and probably redundant molar. The listening, talking personal computer can be held exactly where it is most effective - in the mouth. Piezo electric transducers would quite happily pick up vocalized commands and return answers direct to our ears through bone conduction of sound. Allowing a little more licence with technological advance, such a device would almost certainly be designed for low power consumption from CMOS-like logic, and this power could be generated by chemical reaction in situ using the acidity of saliva as electrolyte and dissimilar metals in the casing. The mouth is nearly a perfect environment. Temperature stabilized, protected, easily accessible. We have a high tolerance to foreign bodies in the mouth so long as they are stable and non-toxic; nearly everyone has a filling, a false tooth, or some other artificial construction there. Also the tooth implanted computer would be more difficult to lose than the conventional pocket kind.

Zeitlin highlights the real economic advantages of such a device, for example in implanting arithmetic into children by this direct method instead of spending thousands of dollars and many months on very inefficient ways of teaching how to add, subtract, multiply, and divide. Language skills (very similar to the concept of a wig-mounted "translator device" used by Dan Dare on his hazardous voyage to Venusian Atlantene territory in the Eagle comic strip 20 years ago), job skills, and social skills could all form part of the implant program. Re-evaluation of the role of education (not to mention teachers) becomes essential when the cost of providing the equivalent of a degree course is perhaps 50 dollars and an hour in a dentist's surgery.

But the working scientist, you will say, must have more direct access to information than that provided by a mere talking book, however convenient. What about the interface to such a machine? After two minutes biting on the end of a pencil, the answer becomes obvious. Pick up your "interface wand" - a slimmer version of a light-pen - locate it on the prepared serial input contact pair of your lower left seventh, and bite gently to take data: a really direct scientist experiment coupling, and one which would encourage him to get his interface electronics right. High speed optically isolated serial links with data rates sufficient for most on-line laboratory instrumentation are well within the scope of present technology, as is the miniaturization required for the production of a mouth-held interface connection. Pipe-smokers would no doubt be at a considerable advantage in terms of technique when it came to issuing selective commands with the interface wand still in position. A short-range radio linked version would also allow for meditative pacing around the laboratory whilst assimilating data, and lengthy calculations and data reductions could always take place overnight with the usual good effects of having "slept on it". GPO-approved "telewands" would allow direct tooth-to-tooth communication over the telephone network, and who knows what advantages might accrue from a double ended (null-modem or twisted link) wand which would allow you and that attractive new colleague from biochemistry to get your heads together and chew over your research findings (perhaps after dinner, to keep the cells fully charged...).

The technological projections by which we arrive at a powerful computer the size of an (aptly named) wisdom tooth are only slightly "ambitious" and are firmly grounded in reasonable, foreseeable developments. The technological capability is undoubtedly available; the economic drive is already positive and not far from commercial reality; the implications for life as a whole, even to hardened users of jargonistic superlatives, are of truly epic proportions. Conceptually there are few real obstacles to the construction of such a device. From the practical standpoint we still have some distance to travel. But who would be so bold as to say that by the turn of the century the big names in the industry will not be taking rather more than a "clinical" interest in this sort of possibility? Zeitlin makes his projection to an information processing computer of this capacity eventually little larger than a grain of sand, implanted directly in the brain and as accessible to it as any other neurone. A trace difficult to interface to, perhaps, requiring a little scratching of the head....

This document created and maintained by Acolyte Science
For comments, suggestions, or contributions,
E-mail to alan@acolyte.u-net.com
© Acolyte Science, 1996
Last Update, 20-May-96