Reliability and reproducibility are key achievements of the industrial age realised through incremental improvements from one generation to the next.

This is a powerful and directed evolutionary process in which we stand on the shoulders of previous generations and employ cumulative experience and knowledge. It is unlike any mechanism found in nature, and results from hierarchical and structured thinking. Modular design and construction definitely work. If you can make a lathe from a bow and arrow, then you can make a better one with the parts you produce.

Similarly, if you can make transistors you can make circuits, and then integrated circuits, and machines that make better transistors. And so positive feedback accelerates the process as machines beget better machines. Nature never uses such directed evolution because it requires intelligence.

So what of future software systems? Can we expect new engineering processes to emerge? You might think so, but perhaps not, as most human thinking is tempered by direct experience of things physical. Also, our limited ability to grasp non-physical experience might be a key constraint. Just how do you understand two million lines of high-level, and very abstract, code? Our mathematical tools and thinking are useless for systems of such scale and complexity, and so we resort to modular designs precluding optimisation and directed evolution. All of this leads to slow response times and poor reliability.

Nature understands nothing, and so would not do it this way. It uses blind evolution through natural selection and chance mutation over millions of years. But we need solutions in weeks. Fortunately, the speed of machines allows us to accelerate the evolution of software entities so a million years take less than a week.

To be really smart seems to entail enhanced attraction through the mechanisms inherent in sexual reproduction. In software, love has nothing to do with the process; it is cold and calculated with the same regard for survival of the fittest seen in nature. But there is now a new card to play. Software can adapt and adopt the right number of sexes to meet the needs of a particular problem. Interestingly, the smartest software seems to come from three sexes, not two.

So when we try to predict the future we should remember the lathe and the transistor, and contemplate the feedback impact of rapidly evolving software. Machine intelligence will speed up the process further and will not be hampered by emotion, or for that matter, the mating incompatibilities and sexual limitations of carbon life. But, like us, future machines will be concerned with information and disorder processing - only they will be faster and better at it.

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