Evolutionary Hardware Design
Merging Nature and technology



Christopher Altman
Pierre Laclede Honors College 


A Brief Introduction
Artificial intelligence is a controversial field that has overcome many obstacles since its inception in 1940s by renowned computer scientist Alan Turing, but it has a long way to go in order to live up to the expectations foreseen by its pioneers. The subject of machine consciousness has been a highly debated topic, made even more difficult by the fact that we have yet to pinpoint the mechanisms underlying consciousness in the first place.



 

A highly active point of current research lies in the study of field programmable gate arrays, chips that effectively program themselves, evolved to perform specific tasks set forth by the researcher. The concept of 'evolvable hardware' is being studied at many research centers across the globe: most prominently at the Advanced Telecommunications Research laboratory in Kyoto, Japan, and the University of Sussex in England. Evolvable hardware works by using chips that can be reconfigured by its users, literally rewriting the logic elements that perform the task at hand.

A key principle behind this technique is the concept that hardware can be evolved in a robot or computer in much the same way that nature works to evolve living organisms. The specific software instructions each chip utilizes can be viewed in much the same way as a chromosome works in nature. Random mutations can be incorporated into the chipÍs performance, then paired together to apply evolution into the chipÍs design.


Over a successive series of generations, FPGA based evolutionary hardware can become amazingly adept at performing the task at hand – better performing configurations are mated together, mutations are introduced, and individuals in the resulting generation are mated or discarded according to performance.

This method of hardware configuration has produced chips 10 to 100 times more efficient than human-designed chips built to perform the same task. Many of the chips under studies performed by AI researcher Adrian Thompson even evolved abnormal processes to maximize performance, such as electromagnetic coupling and feedback. The exploitation of natural processes to solve traditionally difficult problems shows great promise – as computational resources grow exponentially, this method may have significant impact upon the future of artificial intelligence.

Further Reading


A. Thompson. Evolutionary Design of Single-Electron Systems, in Second Annual NASA/DOD Workshop on Evolvable Hardware 2000.

University of Sussex Evolutionary Electronics Program

Adrian Thompson