(29.05.2001,12:00)

Artificial intelligence: Trial and error

Can there be computer, which is more intelligent than humans? Which sounds like Science Fiction, soon reality could become.

Science Fiction fans know the scene: Act ion hero Arnold Schwarzenegger destroys himself as terminator T-800 first a chip, then. Technology, which - at least in James Camerons terminator films - to which megacomputer Skynet permits the suppression of mankind, today actually in scientific laboratories developed: Hardware, which is able to change and develop further according to the Darwin evolution theory. However still no speech can be from a control of humans - the researchers promise themselves today above all savings with the design of electronic chips. Computers are to get the ability to develop to learn by trial and error to optimize itself - perhaps even problem solutions to find, which are situated beyond the human conception.

The technology is ready
Evolvable hardware (EHW) - itself changing hardware - the glossary word is called. The target: to create itself the environment and the request themselves adapting systems. That sounds still like an idea from a Science Fiction film, however among scientists already since beginning of the 90's-years by the availability to simply reconfigurable hardware (for example Field Programmable gate array, FPGA - those are simply reconfigurable circuits) increasingly attention found.

These chips are combined now with evolutionary software algorithms: Procedures, which are based on the Darwin evolution theory. By imitation of principles such as reproduction (reproduction) and mutation (coincidental, small modification) these procedures modify their own parameters. Following the Darwin " Survival OF the Fittest " it is ensured that only well adapted individuals survive or at least a larger number of descendants have.

The technology is interesting therefore also for space travel and armaments industry. The US space research authority NASA organizes regularly Workshops to this forschungsgebiet. Jason Lohn of NASA research center in California explains: " of EWH one promises to adapt to machines, which program themselves, and develop can. " NASA is interested for example to develop systems in weltraumfahrzeuge which can repair themselves, if somewhat inclined goes.

Cash save with the design
An interesting area of application for changing hardware is the Design of similar circuits. One finds everywhere digital devices in the today's time, but must these naturally also with the real world communicate - and those is still similar. Problem: Enviromental influences as for example temperature and air humidity can change the electrical characteristics in microchip of integrated resistances and condensers strongly, which is expressed in signal distortions. The probably largest research establishment in this area is the Evolvable of system lab of the Electrotechnical Laboratory (ETL) in Japan. It is under the direction of Tetsuya Higuchi. He succeeded in optimizing with the help of genetic algorithms of chip designs in such a way that they correspond complete with the Design specifications. As the algorithm constantly checks the performance of each chip, it develops an architecture, which is to a large extent insensitive for fluctuations with resistances and condensers.

Further applications of the changing hardware see themselves Higuchi and colleagues for example in autonomous robots, which can reprogram themselves on a new task. Them it constitutes also nothing, if suddenly a sensor (for instance an eye) does not function any longer - they get themselves the necessary data over their environment then from other sensors or by trial and error. Also the Japanese researchers designed a hand prosthesis adapting to their user. Their advantage: While it takes one month normally approximately, until the owner of the prosthesis can control these contently placing, the model equipped with changing hardware is already in minutes operational.

Refined solutions owing to evolution
Also Adrian Thompson of the university of Sussex operates on evolutionary circuit designs. It created a circuit, which was by use of a genetic algorithm as well as a reconfigurable chip able to differentiate between two frequencies. Which the chip carried out, few attention excited - was more exciting, as the result came to status: It turned out that even cells, which were electrically not at all into the circuit merged influenced its performance. When Thompson made itself to analyze the reason for this he found a circuit, which was so complex nested that he could not explain it: " it functioned, but I did not have notion, like the chip make. " The solution of the evolution is besides so refined that no human developer would have come ever on it.

Some companies plan already commercial applications of the EHW technique. In the US state the Colorado resident company Genobyte operates to time on a wireless communication technology with the name gene TIC silicone, which is to come still in this year on the market. More than the imaginative names of the product wants to betray the President of Genobyte, Michael Korkin, however yet. Still the plans continue to go from Hugo de Garis, Leiter of the research group artificial brains at the star lab research center with Brussels. De Garis wants to simulate artificial intelligence not over software, but build a computer, which resembles the brain in structure and structure. Target is it to design machines which produce brain-like structures in the form of microchip. De Garis developed the Cam Brain machine in such a way specified together with Genobyte - principal purpose: the development of a brain with a billion of artificial nerve cells.

Left
Japanese researchers develop changing hardware
www.etl.go.jp/etl/divisions/~ehw /
Projects of the company Genobyte
www.genobyte.com/projects.html
The third NASA Workshop over Evolvable hardware (EHW)
cism.jpl.nasa.gov/ehw/events/nasaeh01 /
Professor de Garis and its Starbrain project
foobar.starlab.net/~degaris /
Research to evolutionary electronics
www.cogs.susx.ac.uk/users/adrianth/ade.html

De Garis is natural enthusiastically by its projects lightup and sees themselves changing hardware already as the basis of future brain technical designers: " I bet that it will give a billion-dollar market for brain-based computers within approximately 15 years world-wide. " In approximately 30 years de Garis wants to build artificial brains, which are time more efficient one million than the human thinking organ. And that, although it warns expressly of dangers, how it admits so far only from Science Fiction films like " terminator " is: " I am convinced that machines will far overhaul us in things intelligence. That could finally become us as species the calamity. "

In addition, some researchers warn for practical reasons before to large enthusiasm. So for example pure hard stone of the specialist area computer science of the University of Kaiserslautern: " like at that time when arising the term Artificial Intelligence, then some representatives of the Evolvable hardware scene create covered expectations concerning the feasibility and the time horizon to probably approach in order better to NASA funds. Disappointments in some years are programmed. "

Kirsten Kuecherer, Matthias Matting