Skip to main content

Review of the world's first AI product launched in 1988 - Tome Searcher

I am tired of reading and hearing statements by press and politicians alike that companies like OpenAI (the creators of ChatGPT) started the AI revolution. Just today, with the announcement of the Stargate data centre project, the BBC's Verify (?!) website stated "OpenAI kicked off the AI race in 2022". I'm not claiming that a little company in London called Tome Associates Ltd, of which I was a founding director, started AI in the 1980s. There were already plenty of people exploring it around the world at the time. Indeed a project called Logic Theorist, developed in 1955 by Newell and Simon at the Rand Corporation in California, is probably the world's first AI application. But to the best of my knowledge, Tome Associates were the first company in the world to launch a commercial product based on AI... which subsequently met a resounding blank look from everyone we showed it to. People were fascinated, but confused about what to do with it (don't forget Tim Berners Lee only invented the worldwide web the following year, so most people had no idea about going online to search for information). 

We'd solved a problem few people knew they had. How very different might the UK and Europe's fortunes been if just one of today's politicians clamouring for investment in AI had recognised what we were struggling to do nearly 40 years ago.

Below is a review of the launch of our first product, Tome Searcher (a forerunner to Google as well as ChatBots et al) by an AI expert at the time who published his piece in a specialist publication presciently titled AI Business. Judge for yourself whether OpenAI really did "kick off" the AI race.


NATURAL LANGUAGE PROCESSING

AI Business March 3 1988

Software revolutionises access to databases

By Jim Wright

Just two days ago (this Tuesday) a new AI-based product was announced which, it is claimed "understands ordinary English", and sets out to facilitate access to central commercial databases by enquirers completely untrained in their use. At the launch, AI Business was able to examine these claims in some detail.


TOME SEARCHER: the world's first computer device to understand English." This is the claim being made for a new software product developed from a pioneering Al research program at the University of London, under the sponsorship of the British Library.

The aim of the project was to find a way of making it easy for anyone to use call-up database services without specialist training-which effectively meant it had to be driven by ordinary English. Not deterred by the lengthy Iist of Al projects with this aim which have foundered, the team set out to crack the problem afresh.

And they have succeeded. "Understanding English" is perhaps making too much of it of course, it is perfectly possible to fire valid English statements at the system which it has no hope of understanding.

But if you answer its questions reasonably accurately, using unconstrained English, it can extract from your statements the information it needs. Which is a great deal more than can be said for the myriad of other query systems currently on offer, claiming "natural language- style interfacing" - Al specialists often regard the claim as laughable when they actually see the product. But not so in this case. It works.

And because the concept of an instruction-less database query system (it doesn't even have an instruction manual- it doesn't need one) has such enormous potential, Tome Associates Ltd was formed in 1986, incorporating many of the original research team, to exploit the project commercially.

In fact, the 'Searcher' employs Al at two levels-and it is easy to envisage extensions to three, four or even five levels.

The first level is the piece which "understands the English which actually picks out the bits of your statement which it needs. And the second level is the 'librarian expert' package which understands how to get the information you really want out of a large commercial database. The two are, of course, fully integrated, but nevertheless there are two discrete functions in evidence.

A third level of Al could be easily achieved by putting a voice recognition/synthesis package at the front end instead of a keyboard - the technology is readily available now (see, for example, Al Business No 34).

A fourth level of Al would be brought into play if the database being accessed was, say, a medical diagnostic expert system. And it is easy to conceive of fifth and sixth layers of Al-going all the way from the user to the problem, and back with the solution, in the language and format the user prefers. An exciting prospect.

The initial version of the Tome Searcher deals only with searching Inspec or BT databases for information in the areas of electrical engineering, electronics, and computers (and Al) - at one stage, the system was pulling information out of the European Space Agency's computer at Frascati in Italy.

One hates to think about the commercial charges involved-but in fact, the charges come down to local telephone calls, plus the per-item charge for information received, and a few relatively insignificant overheads.

But extension into other database-search areas is obviously in view. Financial systems, legal systems, press cutting services, abstracting agencies, news agencies the list is endless.

And most of these enquiry systems offer information which is both accessed and presented in the English language and there are probably thousands of them. People simply do not recognise that their telephone lines can deliver such quality information-and products such as the Tome Searcher aim to make the process even easier.

On looking at the English-collecting end of the system, one is strongly reminded of Weizenbaum's ELIZA system, which incorporated algorithms for recognising the 'important' words in English statements. If one typed in "I don't really like computers" it would immediately focus on "like" and "computers" and come back with "What is it that makes you say you don't like computers?" In this way, one could conduct an apparently meaningful dialogue with the machine-of which it understood nothing at all. But that was more than 20 years ago.

Apparently the Tome Searcher doesn't work like that. It already knows that it has references for computers, optical fibres, VLSI and so forth-so it can check for the 'reference words' in input statements, focus on them, and discard the remainder.

Which means that when one types "Please give me some information on Hybrid-T repeaters as applied to local area networks" the system will simply focus on "Hybrid-T" and "local area networks" - and ignore the rest. It is a bit cleverer than that-there is a full-function semantic analyser in there-but what I have described is the end result. If I typed "What I really need to know is what you have got on Hybrid-T for networks" the system would still attempt to search for the same information.

But if it only searches for keywords it already has listed, one wonders if the front-end processor actually has 'intelligence'.

The semantic analyser (working with unconstrained natural language statements) is obviously an Al system. And when one looks into the system a little more deeply, it is obvious that it has intelligence. Tome call this part of the system a "free expression processor because they understandably feel that the term "natural language input" has been somewhat abused.

The front end also has a Thesaurus. A simple
Statement, but one of substantial importance. Suppose you have posed a hypothetical question about optical computers used in the car industry, and asked for the 30 papers, and it has only come up with two. To satisfy your demand, the system will propose how you can broaden the search. It will suggest that it might also march for "automotive industry" for example - giving you buses, trucks, tanks, hovercraft and any other land vehicles it can identify.

In fact, if your grasp of English is limited, and you don't understand what "automotive industry" means, and start off enquiring about "vehicles", the system will quide you relentlessly towards asking the right question. This is a very powerful function, with significant potential in international markets especially Japan and Germany. In fact, it can be of value in any situation where the database is in English (which is usually the case) but the user is not a native English speaker.

This aspect of the system was something I tested. I typed in "sysstems" on one enquiry-not deliberately, it was a typing error, but I let it stand to see what the response would be. The system sharply responded that elther I had made a spelling mistake, or it was dealing with a new word, and would I kindly tell it what it was supposed to mean (from the following categories)?

From an Al product, that is a response which is no less than excellent. The basic vocabulary is around 7,500 words at present but will grow, and can easily be "trained" to particular words the user requires.

Further, the system does try to make sense of whatever nonsense it is asked to deal with. I asked it to search on "Reports of plain English systems for accessing databases" (knowing there are several hundred of such claims).

And to the amusement of onlookers the system which "understands English" didn't understand the word "English" and wanted me to tell it what particular species of electronics that might be. (Well, they'll get some of these warts out soon when real customers get their hands on it). But we quickly identified the correct set of search keywords.

So much for the English-acquisition end of the system. The other strong Al element is the database-driver. This, in fact, is a hand-tailored expert system not built upon any particular commercial shell or environmental product (Tome claims nothing suitable was on the market in 1983 when the project began) but on an expert system written by their own specialists.

The entire product, interestingly, is written in Pascal. Not Lisp, not Prolog, not C or even Cobol "When project began," said MD Carl Mattocks, "Lisp and Prolog compilers were a good deal less refined than they are today, and Pascal seemed the best choice at the time." They will probably have to go for a Lisp version soon, but at least Pascal will give them an easy path to Ada for defence-related work.

Searching a large database is no fun for the beginner. (The Inspec system, for example, puts more than three million references on line in the electronics/computers area alone-hundreds of other databases have similar numbers for chemical, legal, financial or other sectors).

Tome Searcher aims to conduct all the database dialogue on behalf of the enquirer. In essence, it plays the role of a 'specialist librarian' who is asked to find specific references by whatever means available. In this respect, I found the product reminiscent of modern modems, which, once a date link has been established, negotiate between themselves matters such as packet size and error-checking routines, without the user knowing anything at all about the negotiations.

Tome Searcher does the same sort of thing. Once it has got into the database it is asked to access, it undertakes all the necessary protocol without the operator being involved. It simply does it for him. To be precise, the user obviously has first to call up the package. Then, on screen, he/she has to identify the database which is to be accessed. One must then provide a password for the database host (the Inspec system, for example, charges for usage, but there is no licensing charge' for being granted an access password).

Next comes a screen menu, inviting the user to specify search criteria. One may search by subject, author, title, etc - but perhaps the most important item on the menu is "number of references". If one specifies "find 30" and the system actually finds 10,000, it will simply report on the 30 most recent items - which avoids dumping vast amounts of unnecessary data onto the local disc (and eliminates the associated communications time and cost). which

Then comes the "free expression" part. One can type in "tell me about" or "show me information on" or "I want to know"- all these expressions are treated in the same way. The system even sets about correcting spelling errors.

And then the expert part of the package sets about conducting the dialogue with the database system. The user can see, but need not understand, the dialogue which takes place which includes all the logging-in functions, all the query functions, and all the data-extraction functions which normally have to take place in a database-query session. It all takes place completely automatically.

And finally, the requested references, honed into shape by the user (having been led in the right direction by the Searcher system) can be acquired and written on to his local disc for subsequent perusal. Which all makes for about the easiest database access system I have yet seen.

The software runs on any PC-compatible machine with a hard disc and 640K of RAM, although a 512K release is planned soon (which will be quicker). The system needs about 5Mb of the hard disc.

And finally, we should report on the price. This also is good news. Despite the fact that the product does more, and more cleverly, than almost every other database access system on the market, the software is competitively priced at under £500 per installation. (Of course, the purchaser also has to have a PC and a modem-but there are plenty of those on the ground). It should sell well.

One of the longest-outstanding aims of Al research has been to achieve natural language input (if Tome will forgive me for using those words) to information retrieval systems. On the evidence I saw on Tuesday, that barrier has finally been breached. And the potential applications of the technology assuming Tome manages to keep its head above water are virtually limitless.

Al Business March 3 1988

Comments

Popular posts from this blog

Phillips screws - yes I'm angry about them too

Don't get me wrong. They're a brilliant invention to assist automation and prevent screwdrivers from slipping off screw heads - damaging furniture, paintwork and fingers in the process. Interestingly they weren't invented by Mr Phillips at all, but by a John P Thompson who sold Mr P the idea after failing to commercialise it. Mr P, on the otherhand, quickly succeeded where Mr T had failed. Incredible isn't it. You don't just need a good idea, you need a great salesman and, more importantly, perfect timing to make a success out of something new. Actually, it would seem, he did two clever things (apart from buying the rights). He gave the invention to GM to trial. No-brainer #1. After it was adopted by the great GM, instead of trying to become their sole supplier of Phillips screws, he sold licenses to every other screw manufacturer in the world. A little of a lot is worth a great deal more than a lot of a little + vulnerability (watch out Apple!). My gromble is abo...

To kill or not to kill.

Had an interesting discussion with a Muslim friend today about the ethics of killing. Could it ever be morally justifiable? Abrahamic scriptures, especially the old testament, are awash with murders and killings, some sanctioned by the prophets and assorted mouthpieces for god. Some killing is even mandatory. For example all Jews are instructed in the old Testament to kill everyone belonging to the 7 Canaanite tribes for example - Deut 20:17 , or to slaughter Amaleks, especially their children - Deut 25:19 . So accepting for a moment that these draconian instructions were written in times when tribal leaders had fewer options available to them with respect to managing miscreants and maintaining some sort of law and order, I suspect that most people today would agree that killing people is a bad thing and should not be condoned except under extraordinary circumstances. My friend and I then proceeded to try to list those circumstances. We started with self-defence or perhaps protecti...

Successful Entrepreneurs Don't Aim to Make Money

Of course all entrepreneurs want to make lots of money. Who doesn't? But the difference between entrepreneurs who do make money and those who don't, is that successful ones don't focus on making money. They focus on building their businesses. And that relies on having an attitude of pouring any money their businesses do make, back into them, rather than rubbing their hands and taking it out as soon as they can. True entrepreneurs are gamblers and thrifty by nature. Given the choice of a holiday of a lifetime versus the chance to create a great business, they'll always choose the business - and take it for granted that if the business does eventually make surplus money, they can have that holiday - although entrepreneurs can become so hooked, holidays become a guilty wrench away from the businesses they need to protect. I didn't have a single days holiday, or off sick, for 10 years after I started my first business. I probably could have afforded it (in fact my wif...