Watson

I mentioned Watson on 5th July and now get around to a more substantial mention.

The long name of this little book is 'Smart Machines: IBM's Watson and the era of cognitive computing' and the main author is one Steve Hamm, an IBM staff writer. A very US sort of book, full of optimism about how technology is going to save the world. A very company sort of book, full of short but edifying - rags to riches - biographies of illustrious IBM staffers and of tales of IBM contributions to the world. I read incidentally that they employ lots of mathematicians, so maybe I should have joined them, a blue chip operation which has remained in business, stayed the course, instead of joining the civil service! It was also an easy read which does not dig very deep into the technology, about on a par with an article in the Economist.

The story starts with a computer which I remembered as being called 'Deep Throat' but which was actually called 'Deep Blue', a chess playing computer which defeated the flashy world champion Gary Kasparov in 1997. It then moves onto 'Watson', a Jeopardy! (see http://www.jeopardy.com/) playing computer which defeated various important people on television in 2011. The first point of interest is that playing Jeopardy! is a lot harder than playing chess, which made me stop and think a bit. The second point of interest is that Jeopardy! appears to be the most successful & longest running television quiz game the world  has ever known, despite my never having heard of it. And all credit to IBM for having a crack at something which it was not clear at the outset that they were going to be able to do - and for going on live television with what they were able to do. However did such a thing pass me by?

The next thought is that rather than have computers playing against humans, we have them helping humans. So, for example, IBM and others are working away to adapt Watson to help oncologists, to help them make better use of the huge amount of information about cancer sculling around the world, far more than any human is going to get a grip on. How do we get Watson to cope with all this data structured or unstructured in a million and one different ways? I notice in passing that a plus point for Watson is said to be his ability to interact with the computer systems of health insurers, important as an oncologist needs to get any proposed course of treatment OK'd by the customer's health insurer before (s)he can go ahead. Not quite how we do things over here, at least not until Cameron and his crew have finished with us. (Although we do need to remember that someone has to do the rationing bit. Maybe I don't want an insurance company doing it, but someone has to).

Next question, how do we cut down on the huge amount of human input which was needed to teach Watson how to play Jeopardy!? How do we teach computers to teach themselves and save us the bother?

Then the argument is that we need to move processing out to the data, to blend process and data rather in the way that a human brain does. Which reminded me of the CAFS (content addressable file store) which was being promoted by the company then known as ICL back in the eighties of the last century and which, according to Wikipedia, had its problems and was eventually killed off by increasing central processor speeds; it became quicker to move the data to the central processor and look at it there, rather than to look at it on the disc. The IBM argument seems to be that stagnating processor power combined with rapidly increasing data volumes mean that this is no longer true.

Then we have an aside about something called the Cognitive Enterprise Lab, a fancy computer assisted meeting room which sounds very much like something called the Pod invented by guess who, back in the eighties of the last century - ICL. Maybe lots of people were playing with the things at that time and maybe the time has now come when they might work. Maybe the COBRA room in Whitehall we hear so much about is one of them.

Before moving onto the need for computers to understand stuff from our five senses - sight, sound, touch, taste and smell - as well as they can understand words now. And to soak up all the stuff swirling (or perhaps swilling) around the mobile phone networks. Ten years from now, for example, the chic thing to wear might be a head mounted smell sensor, smelling and recording everything that the wearer smells. And then doing something useful with all this stuff. Perhaps the sort of thing an industrial chemist ought to wear when touring his factory, perhaps with some kind of a display to tell him about anything odd.

And to power all this we are going to need a new kind of chip, something that gets around the physics barriers that our chips are hitting just presently. Maybe a chip made of neurons like a brain rather than transistors like a radio. Go parallel processing down at the level of the chip rather than pretending at some higher level. Somewhere I read that provided you keep your wires at least ten molecules thick, the current in the wire still behave like the currents we learn about at school. Go under ten and you go quantum which is a whole new world.

Along the way we get glimpses of applications to help me decide which new car to buy, to help build giant radio telescopes, to help manage country wide power grids and to help manage large cities. Which last resulted in a waking wonder this morning about what happened when one had a computer model of a city which was interacting with the real city in real time, perhaps managing the traffic lights and the dispatch of police cars to hotspots. What is the relation of the model to the city? Who is in charge? Are the model and the city in symbiosis? What about the computer model of the city including a model of itself? By which time I had woken up, which was perhaps just as well.

But a good read, well worth the money. Maybe I shall return to it when I have read it again, a bit more carefully.