Friday, September 14, 2007

When a Gizmo Can Invent a Gizmo

WHAT'S NEXT; When a Gizmo Can Invent a Gizmo\

November 25, 1999 New York Times

IF Dr. Frankenstein's monster had published a best seller, who would have gotten the rights to that intellectual property?

His inventor up in the castle, of course.

Tough luck for the monster, but these are still early days for intellectual property rights for thinking machines. No one has seriously proposed that a computer should receive a share of the profits from an invention -- at least not yet. But other problems related to the ownership of items invented by computers are already being debated in preparation for the time, probably in about 10 years, when such inventions will be commonplace, said David E. Goldberg, an engineer and a professor at the University of Illinois at Urbana-Champaign.

Computers are already making inroads in the area of intellectual property as they design antennas, gas turbines and integrated circuits. Much of the work in this field of automatic discovery is preliminary and a lot of it is proprietary and therefore secret, but what can be seen provides tantalizing glimpses of a future in which computers work day and night -- no breaks for lunch -- to come up with original solutions with very little help from their programmers.

In Herndon, Va., for instance, a program by Derek S. Linden, chief technical officer for Linden Innovation Research, automatically designed a satellite communications antenna that was original enough to be patented by Dr. Linden. Programs developed at Engineous Software, in Morrisville, N.C., and sold to an aircraft manufacturer came up with improvements to a jet engine so marked that the aircraft company filed two patents on them.

Computers are still at the stage where they design components like filters, circuits and engines -- the whole plane is yet to come -- but their ability to come up with inventions automatically raises a host of philosophical and legal issues that will intensify, some people familiar with the field say, as computers grow more powerful and their discoveries more extensive.

It may be later than you think. ''We think of evolution as slow, as it took about 3.5 million years for natural systems to evolve,'' Dr. Goldberg said. ''Powerful computers will vastly accelerate the evolution of machine invention and the solving of hard problems. People thought that inventions were beyond the range of computers, but that's not so.''

Dr. Goldberg is an expert in genetic algorithms, math-based approaches that solve problems by mimicking the principles of natural biological selection. Such programs examine successive generations of solutions to problems and choose the best ones, whether the problem is scheduling workers in a factory or building an airplane wing.

These programs offer the promise of one day solving not only the problems at hand, but also future ones, long after the program has been sold and is out in the world with different owners. The issue of adjudicating these inexhaustible intellectual resources and the original solutions that result will probably prove quite complicated.

The first result will be a swamped patent office, predicted Jordan B. Pollack, a professor at Brandeis University who works in evolutionary robotics and teaches a course in intellectual property. A patent is denied when an invention is obvious to a person of ordinary skill in the art, Dr. Pollack explained, but applying that standard may become more complicated as more ideas come from computers. ''The whole idea of what's new and nonobvious will have to change,'' he said. ''Our definition of 'obvious' will need to be expanded,'' he added, to account for things that would be obvious to a computer program.

Patent infringement will also be an issue as computers grow smarter, Dr. Pollack said. ''These machines are going to go out on the factory floor, solving new problems for owner after owner,'' he predicted. ''They will be tripping over everyone else's patents all day long.''

The patent system itself, conceived to reward human innovation with a limited monopoly, will eventually have to be modified, Dr. Pollack said. ''We don't yet realize that machines are engaged in invention -- but this is already happening in many areas,'' he added.

John Koza, president of Genetic Programming in Los Altos, Calif., is an expert in evolutionary computing. ''The real trick in these evolutionary inventions lies in recognizing them,'' he said. ''It's actually easier to evolve circuits than to identify a big invention.''

Dr. Koza and his colleagues have been creating electrical circuits using evolutionary computing. ''We recognize when these circuits infringe upon existing patented circuits because we know the existing circuits as textbook inventions,'' he said. ''But in our hundreds of runs, we've probably already invented many other circuits but haven't yet spotted them.''

And what will happen to humans as computers become increasingly skillful at automatic invention? Dr. Goldberg predicted that people would quickly adapt.

He drew an analogy to the Industrial Revolution, when steam engines removed some of the heavy lifting of industrial work. ''Computers will give us the same kind of leverage in the intellectual range,'' he said.

Many inventions in the future will routinely be handled by computers. ''No one would think of building a skyscraper with thousands of workers,'' Dr. Goldberg said. ''Similarly, no one will think about solving a problem without getting the magnitude of intellectual leverage that is similar to the mechanical leverage of the steam engine.''

Dr. Goldberg is not worried about who will be in charge -- people will be. ''This development doesn't mean we'll be supplanted,'' he said.

People will still be involved in creative acts and worthy of their royalties, he said, but their role is going to shift.

''We'll become managers, directing the machines toward interesting problems and opportunities,'' Dr. Goldberg said. ''The creative act will be in mentioning the right problems.''

No comments: