Monday 9 May 2011

Communication languages and Intelligence

  I have just posted the follow comment on Babel's Dawn - a blog about the development of Language.

You write
When you think how much computational effort is required to support a machine playing chess or Jeopardy, you realize that it will be some time before even sophisticated ape interactions can be simulated, let alone plausible human conversations. Even so, these small efforts encourage a thought. Language is the keystone that brings cooperation and understanding together.
There is an assumption underlying chess playing programs,  Jeopardy, and your comments about them, that you are certain that the stored program computer architecture form a suitable starting point for understanding the role of language cooperation and understanding.
     Perhaps we should remember the words of the Irish yokel, who, when asked by a stranger the way to Balimoney thought for a moment and responded "If I was going to Balimoney I wouldn't start from here."
    The whole stored program computer philosophy resolves around the concept of a task which can be precisely predefined as a global model, and written as an algorithm. I am sure you will agree that both chess playing systems - and the programs in your ipod depend on the role of "intelligent designers". and both relate to activities which have no relevance to the origins on language on the African plains.
    Forty five years ago I was a naive newcomer to the computer world, having been very much involved in complex manual information processing activities. I was asked to familiarise myself with a vast sales accounting system (say 250,000 customers varying from private households to the U.S. Air Force, and say 5,000 different products aimed at about a dozen different markets). All the time there were new customers, contract changes, and old customers dropped out, which the products and sales promotions were changing to meet the real world market. Any solution needed to be simple enough to be able to process tens ot thousands of transaction a day on late 1960s computers. No knowing any better I used my knowledge of mentally working in non-computerised information systems to come up with a "simple" solution, modelling how I thought the sales staff modelled the problems in their heads.
    The starting point was "language." Sales staff needed to be in active control of the system and they could only control it if they fully understood what the computer was doing for them. What was needed was a contracts language which was simple but flexible enough to cover any reasonable contract - and - most importantly, was symmetrical. The sales staff would use the language to tell the computer what they wanted it to do, and the computer could tell then, IN THE SAME LANGUAGE, what it was doing for them.
    The reaction to my suggestion was - "That's research" - sales staff are not clever enough to tell the computer what they want it to do - they need very clever people such as programmers and systems analysts (the priesthood of computing!) to act as intermediaries.
    Shortly afterwards I became the ideas man on a future planning team of an innovative computer company. Within a few months John Pinkerton and David Caminer (the pioneers of UK comuting who built the Leo computers)  rushed me into research to look at the design of a revolutionary new type of  information processing "white box" system which generalised the contract processing language to handle a very wide range of open-ended problems. The elements of the system were sets and partitions of sets, used recursively, to allow any level of nesting. Processing was by a very simple "decision making routine" which had a small window on the knowledge base (equivalent to human short term memory). The approach takes  incomplete, fuzzy  and missing information in its stride, and for many tasks results were obtained by the decision making routine without anything that looked a bit like a task specific program.
    So why haven't you heard of CODIL, which was the name of the symmetrical information language.. It's a sad story which I describe on my blog, Trapped by the Box, but basically, exceptional claims need exceptional proof, which in turn needs exceptional funding to provide. The problem was that funding is provided by an establishment who knows that the stored program computer must be the only possible way forward (look how much money and careers depend on the technology) and where by now all the population under retirement age will have been taught (brain washed?) at school that writing programs is the way forward.
    At a deeper philosophical level, there is another difficulty. Humans are the most intelligent animals we know, and so there is a danger that we put ourselves at the centre of the "intelligence universe" in the same way that our ancestors put the earth at the centre of the physical universe. As we are "so clever" the mechanism that makes us intelligent must "of course" be very sophisticated and hard to find - and so all simple solutions must be rejected.
    In fact all my research does is to move the focus of"intelligence" from the processing algorithms (which are very simple) to the communication language - with the concept of recursion (which can easily be mapped onto a network model) playing an important part. In as far as one can identify "intelligence" it is in the way that statements in the communication language interact with each other. Of particular interest the Decision Making Unit algorithm is probably simple enough to be looked at in evolutionary terms. The approach also suggests that intelligence as we see it, and distinguish it from other animals, is a result of the development of an effective communication language.
    Having abandoned the research many years ago (following a family suicide and the failure to get research grants) I recently decide to look online to see what had happened in the intervening years. In case anyone is interested I am in the process of setting up a blog, Trapped by the Box, which discusses the research, and includes information on publications and a working demonstration system.

1 comment:

  1. That's a fascinating point of view. I think there is no stopping the progress of modern technology, and it's only a matter of time before we find artificial intelligence to be a crucial part of our machine's functions.

    ReplyDelete