downtoearth-subscribe

Back to nature

Back to nature "THE living. and breathing realm of nature offers a huge)ibrary of design metaphors," says John Riles, president of Thinking Tools, a California-based concern which specialises in developing software based on natural analogies.

The human immune system offers a .wide range of possibilities to ward off a virus of an altogether different kind: that which invades a computer. Stephanie Forr~st, professor of compU- ter science at the University of New Mexico, USA, Al~SPerelson, animmu- nolo gist at the us-ba:sed Los Alamos National Laboratory and reserrchersat Interval Research Incorporated of Paolo Alto, California, us, have come up with an anti-virus software based on the con" cept of'negative selection' adopted by a certain class of protein-attacking ce!ls in man, the T-ce!ls.Those T-ce!ls which are a threat to the body's own proteins get destroyed almost i immediately after birth. The surviving T-ce!ls become highly s~nsitive to foreign bodies and on detecting attacking viruses, immediately signal the immune system and set it into operation. The researchers have designed their software inspired by the functioning of theseT-ce.!ls. The preliminary test of the"software - comprising combinations of computer code~ -exposed a large number of computer viruses, some of them unknown. As soon as a virus was detected, a window popped up on the screen, with a warning sil:ying"a change has been detected" and the software then displayed the infected file. David Liddle, co-founder of Interval Research argues that nature has the ability to tackle and test a wide range of solutions while computers with their very pointed and logical approach sweep thr,ough only a narrow range of solutions.

Hiles has developed a programme taking the c~e from the deadlytubercu- losis (TB) bacterium (MycQbacterium tuberculosis) which constantly under- goes spontaneous mut,ltions and ends up developi~g antibiotic-resistant strains. Software inspired by such behaviQur constantly manipulates the information and repeatedly challenges the user. The person using the software 'better learn' the programme thorou- ghly because .it can ruin the operation if the same mistake is repeated. This pro- gramme will be commercially available from mid-1996.

Andrew Pargellis, a biophysicist and software expert at the AT&T Bell Laboratories has led a team to build the digital equivalent of the 'primordial soup' -the hotchpotch of chemicals from which the earliest form of life is believed to have originateQ. The pro- gramme, aptly dubbed' Amoeba', coi1- sists of a thousand rectangles of diffe- rent colours, each containing a part of a much larger and complicated mathe- matical instruction. They are then sub- jected to a simple command -"copy thyself'. Soon, the rectangles start filling up the screen in a haphazard manner - but suddenly one may for instance find more blue rectangles than red.,This implies that the blue ones are more 'evolved' than the red ones, in the sense. that they have reorganised the code to reproduce themselves in fewer steps than the rest. The point is that the software, long after being given instuc- tions, spontaneousl~ seeks out better ways to implement them. Though the software for the time being solves only simple arithmetic problems, Pargellis is optimistic that with future develop- mcnts it will be possible ',to define a problem and let the software evolve solutions to it".

The idea of harnessing the notion of evolution to generate software solutions stems from the concept of 'genetic algo- rithm' pioneered by John Holland of the University of Michigan in us in the '50s. The way zeroes and ones are woven together in a binary program resembles the manner in which genes are strung in a chromosome. The mutant genes are passed on to descendants resulting in the 'natural selection' of the type more inclined towards survival. "Three billion years of evolution is the most powerful algorithm that ever can be," stresses David Goldberg, professor of computer engineering at the University of Illinois at Urbana-Champaign, us. The genetic algorithm software has been put to a rather unusual use -to reconstruct faces of possible criminals based on the description of witnesses. Each trait of a human face is coded in the fornlof digi- tised information and a likely image is constructed from thousands of random faces, most of which are eliminated by a process analogous to 'natural selection' in spirit.

Computer scientists have perhaps barely leafed through the first few pages of the most complete and consistent handbook ever: nature. Undoubtedly, there is a lot more left to read.

Related Content