We go forward, generation after generation, trusting to luck for that we will realise a satisfactory outcome - a reasonable and responsible member of a society. In reality the spread of outcomes is very wide and mostly acceptable. But there cannot be a parent alive who has not had reservations about some feature of his or her offspring's behaviour.

A most sobering experience is to see your own faults reflected in your child. At best, parents can take corrective action, but the final mature outcome is still uncertain and attempts at correction may even be counterproductive.

So far carbon life systems have evolved a range of sensors, emotions, communication strategies and rule sets that promote the survival of individual species. On the average, love and kindness overpower hate and meanness, care overpowers neglect, collaboration is better than isolation, and we choose to take sensible risks, fight for survival and be benevolent. Our natural tendency is not to kill but to nurture life.

A large proportion of this appears to be fundamental to our nature: we are just born that way - if we were not, there would be little hope of surviving very long. The downside of all this is the constraints it places upon the way and speed we can evolve. This process of change and adaptation turns out to be very slow and measured, which is not always a desirable feature.

In silicon systems we are still writing the initial rules, discovering the opportunities and adapting the technology because we have new found degrees of freedom. Why ape the limited world of carbon life when there are new opportunities? When we consider software or robotic reproduction in the context of children, there are no moral constraints.

A "child" may now be born of parents in a system that coldly judges its performance of this new entity. If it does not measure up to requirements, does not represent a sufficient advance, it can be terminated. The observed attributes it had - good and bad - can be fed back into the parents to modify or steer the genetic features of the next "child". So, between mature generations a raft of "virtual children" can be tried and tested, sequentially refined, and used to create the best or nearest fit to requirements. This sees each generation of new parents capable of making giant evolutionary leaps in ability.

In telecommunications and computer networks today we are witnessing a rapid evolution of peripheral equipments and services. These include millions of fax machines, pagers, mobile phones and online PCs in less than a decade, and there is much more to come. Conventional software production techniques for super reliable and adaptive networks and service provision are looking increasingly difficult, and by 2005 may be impossible. But it might just be that terminating software children for not living up to our expectations could be the answer.

Peter Cochrane holds the Collier Chair for the Public Understanding of Science & Technology at the University of Bristol. His home page is:
http://cochrane.org.uk