杏吧原创

Machine Nature by Moshe Sipper, McGraw Hill, $24.95, ISBN 0071387048 Reviewed by Peter Bentley

EVOLUTIONARY computation merges biological ideas, such as natural evolution, with those of computer science. In Machine Nature, Moshe Sipper explores this young science, beginning his investigation with a round-up of the usual suspects: genetic programming, artificial neural networks, fuzzy logic, robots and evolvable hardware. He also touches briefly on even newer developments such as DNA computing, artificial immune systems and how ants optimise resources in building up colonies.

He devotes the last section of the book to a discussion of where such research is heading, with ideas of open-ended evolution, biologists learning from computer models and robots becoming embodied. Sipper is at his best when describing robotics and his own 鈥渟elf-healing鈥� electronic circuits.

All good stuff and very readable because Sipper鈥檚 style is informal, interspersed with slightly odd quotes, from everyone from William Shakespeare to William Gladstone. But I did find the book a bit disappointing. Sipper shows little understanding of biology or complexity theory and manages to get some of his terminology wrong. Worst of all, though, are some serious omissions. For example, he talks of 鈥渃ellular computing鈥� without mentioning cellular automata at all, and evolvable hardware without explaining the fabulous intrinsic evolution performed by Adrian Thompson. Perhaps less time spent quoting playwrights and more citing scientists in the field would have helped.

But as a fellow scientist and writer in this field, I was pleased 鈥� and surprised 鈥� that we share common themes, such as my pet hate of the misuse of the word 鈥渁rtificial鈥� in labels such as 鈥渁rtificial life鈥�. Convergent evolution, perhaps?