Last Modified: ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?



Homepage / Publications & Opinion / Archive / Interviews (pre 2002)

NEW FORMS OF SOFTWARE AND SYSTEMS FOR A NEW AGE
Press Briefing April 1994, Peter Cochrane

It is interesting to contemplate that, whilst man has continually made his systems increasingly complex, and usually for realising relatively simple functions, Mother Nature (God!) has gone in entirely the different direction. There are numerous examples of extremely simple life forms exhibiting phenomenally complex behaviour. These include ants and insects who have social orders and structures that are incredibly robust, exhibit tremendous resilience and survivability, but founded on a remarkably few lines of software and an unbelievably simple computer. Similarly, life forms such as jellyfish and pelola worms are able to organise themselves as colonies and act in unison without the benefit of a centralised processor of any kind.

Contrast the above with today's computers, IT and telecommunications networks. These are phenomenally complex - in fact, the most complex entities ever constructed by humankind - and yet their functionality is remarkably simple. We already see complexity far in excess of human understanding resulting in massive brownouts of networks in the United States, and at a human level, we are consuming vast amounts of resource in terms of the number of people devoted to writing and structuring software and systems. Both of which gives some indication that our present trajectory may be unsustainable.

All of this heralds a growing danger and difficulty - and probably constitutes a fundamental limit to our technological progress. Without a complete view and understanding of the entities we are constructing, it is hard to see how we will avoid brownouts and catastrophic failures. To date, the efforts of man in understanding software have been somewhat limited and the biggest single problem we face in computer science and engineering is a total lack of science and engineering! There are formalisms, there are rules, there are green-fingered approaches to constructing software, but none of these constitutes an ounce of understanding or a sustainable trajectory in the long term.

Are there any alternatives? Looking at nature, we might suspect that there are and the series of developments at BT Laboratories have confirmed that this may indeed be the case. For example, in previous briefing on the activity concerning ants has now led to the demonstration of complete network restoration algorithms that have reduced the software component from 1.6M lines of code to only 800 lines, with the net result that the network operates more efficiently and effectively, is more resilient, and perhaps most important of all, is understandable by a human being.

More recently novel solutions to the travelling salesman problem have also been found and we currently hold the world record for the fastest algorithm. This is not just an interesting academic exercise but an increasingly important aspect of future telecommunications where the periphery will become far more mobile and a major task will be to locate people and find a route to them in the shortest possible time. In its present realisation we are able to find the best route linking 1,000 cities in under 3 minutes on a modest computer. Alternative solutions require super computers and take considerably longer. This scheme has also been extended to hyper space and we are able to optimise routings and linkages in N-dimensional space in super short times. Interestingly, the departure from an optimal solution is less than 4% - and indeed, is a constant error - over the range 1,000-30,000 cities.

Using dumb agents (ants) to solve problems and realise solutions to the network control and management situations is one extreme of a spectrum that can be made increasingly complex yet understandable by the human. An alternative is to use smart agents which have the ability to carry information and interact between themselves in a dynamic way. This leads to the notion of breeding, evolution and artificial life. The difficulty here is that God has had 500,000,000 years to get it right, but in the engineering context we don't have that length of time to wait! Such agents exhibit interesting properties on an organisation scale and are not only able to manage network problems on a rule based series of algorithms, but exhibit learning properties and a degree of cognitive skill. Such entities have been used in trial systems for a range of problems spanning work force allocation and optimisation, network routings, finding information, database searching and organisation, and performance extension of complex systems that exceed the normal failure criteria.

All of the above work is of an exploratory nature, but in each case convincing demonstrators have been constructed and tested. At this briefing you will have an opportunity to gain hands-on experience of both dumb and smart agents in a number of environments and see for yourself their effectiveness and limitations. This is an experience not to be missed !

All materials created by Peter Cochrane and presented within this site are copyright ? Peter Cochrane - but this is an open resource - and you are invited to make as many downloads as you wish provided you use in a reputable manner. Website maintained, hosted and designed by USP Networks Events Review