|
| Deep in the heart of Belorussia, a group of programmers is trying to automate creativity and thus shape the future of innovation. Starting with a long-standing idea that invention has logical rules and principles that lead from problem to solution, the programmers are weaving a meta-researcher out of software code and Internet tools. Their goal is to enable engineers and scientists to come up with product ideas and new services much more quickly than the old-fashioned, lightbulb-over-the-head method. The name of the Boston-based company these programmers work for nicely sums up the concept: Invention Machine Corp.
A machine that can actually eliminate humans from the creative process is a ways off, but it is getting closer. And nobody knows that better than Corporate America. As high-tech and low-tech companies alike search for the new idea that will make them the next attention-grabbing growth company, the likes of Phillips Petroleum (P
), Medtronic (MDT
), Herman Miller (MLHR
), General Electric (GE
), and Motorola (MOT
) are among the 500 customers using Invention Machine's cutting-edge software to speed up the invention process. So far, these companies are primarily using the software to pull together and analyze huge amounts of data very quickly in new and innovative ways. Some results: stronger parts for Formula One racing cars and a new type of filter for oil drilling equipment.
The next step in the automated invention process--a giant step--applies biological principles. One approach, called genetic programming, creates thousands of different solutions to a given application. These virtual inventions fight for survival, with the fittest giving birth to another generation. Eventually, the ideal solution emerges with little or no human involvement. There is even a step beyond genetic programming: so-called evolutionary software, which redesigns itself, rejecting weak links and building on strengths until it evolves into the perfect process or product.
Improbably, today's pioneering work has its roots in the gulag of Josef Stalin. The logic stems from the Theory of Inventive Problem Solving, more commonly called TRIZ (its Russian acronym), which was first developed by Soviet scientist Genrich Altshuller while spending his six years in a Siberian prison as a dissident. TRIZ is a set of rules meant to take the trial-and-error tedium out of generating new ideas. Invention Machine's founder, Valery Tsourikov, studied Altshuller's theory, whose central pillar is a principle called value analysis: To improve a product, place a value on each component, then organize research according to those values. TRIZ gained considerable cachet in the 1990s, and engineers at organizations as diverse as Ford Motor (F
), Motorola, Boeing (BA
), Eli Lilly (LLY
), and NASA apply the rules to their research, according to the TRIZ Institute.
The Road to an Invention Machine
1940s: TRIZ
The Theory of Inventive Problem Solving, a set of universal rules of invention, developed by Russian scientist Genrich Altshuller.
1970s: GENETIC ALGORITHMS
Computer programs based on evolution, which solve a problem by pitting one solution against another until the best rises to the top, in Darwinian fashion.
1980s: GENETIC PROGRAMMING
Programs, coupled with very powerful computers, that can solve problems and design an entire product with minimal human input.
1990s: SEMANTIC PROCESSING
Artificial intelligence combined with Internet search tools and TRIZ concepts to analyze vast volumes of text, pulling out problems and solutions that a researcher may not have thought of.
2000: EVOLUTIONARY PROGRAMMING
Programs that actually improve on themselves, designing generation after generation until an ideal is reached.
|
Tsourikov set out to automate TRIZ and move beyond it. Rather than develop a way of thinking, he wanted software that, in some ways, can do the thinking itself. ``We have taken the best parts of TRIZ, perhaps 80%, and put it into a software tool and then added from there,'' says Sven Bjorkman, Invention Machine's senior technical marketing manager.
The resulting products are dubbed ``semantic processing technology'' by Invention Machine. Its software searches deep into the Internet and then analyzes large volumes of text from technical journals, scientific forums, and other sources, including a complete index of U.S. patents. By breaking down sentence structure, the software reorganizes the content of documents into a format of problems and solutions. What would have taken a human researcher days is done in minutes, and the results often contain information an engineer may not have thought to look for, says Bjorkman.
Dow Corning, for example, is using Invention Machine's Techoptimizer, Co-Brain, and Knowledgist programs in its search for a new adhesive. It wants something to bond very rigid, brittle materials when exposed to high vibration--an important industrial application. ``Techoptimizer worked us through four different ideas that we hadn't really thought about on our own,'' says James W. Mentele, senior information scientist for Dow Corning. These ideas included making an adhesive of nonuniform density and even using a rubber screw to fasten the surfaces together. All the solutions were ``very nice, very logical, but had not been proposed by our engineers,'' says Mentele.
SCI-FI SOMETIME SOON. Genetic programming and evolutionary software, concepts out of science fiction, go beyond semantic software, but have yet to produce commercial results. Genetic programming is derived from genetic algorithms, mathematical procedures for solving problems based on chromosome-like behavior. These algorithms were first developed by John Holland at the University of Michigan in the 1970s. They are commonly used for scheduling in manufacturing plants, but are limited to problems whose solutions can be expressed as mathematical formulas. Genetic programmers, by contrast, are searching for an evolutionary process without any predetermined formulas. It is the computer version of primordial ooze, according to John R. Koza, a computer science professor at Stanford University and the inventor of genetic programming.
Koza's approach got a big boost last year with his publication of a breakthrough paper in the first issue of the journal Genetic Programming & Evolvable Machines. Koza described 10 potential products that were created by genetic software from nothing more than the basic parameters of a problem posed by the programmers. As a test, Koza's team asked the program for a device for receiving television signals, giving only some general size and performance requirements. The resulting ``invention'' turned out to be the same as the familiar Yagi-Uda antenna, the ladder-shaped TV antenna designed by two men in Japan in the 1920s and still in use. ``We have proven that genetic programming can produce products of commercial value,'' says Koza. ``I think it's only a matter of time before something useful and patentable and totally new is created.''
It's not a huge leap of logic to imagine software that can invent its own products, without human involvement. That's the dream of evolutionary programmers, and they, too, can point to a big leap forward last fall. Two scientists at Brandeis University announced that they had developed the first robots that could design and build other robots. In an effort called the Golem Project, after the mythical Jewish creature created from clay, researchers Hod Lipson and Jordan B. Pollack programmed robots with evolutionary software. Attached to a machine that makes plastic models, the parent robots produced devices that could crawl on their own.
Genetic programming and evolutionary software are still found primarily in university labs, but companies looking for the Next New Thing are taking notice. Koza says that the proportion of corporate attendees at genetic programming meetings has grown from zero a few years ago to half. It's not so hard to imagine, then, that an engineer at a hot growth company may someday have a lab partner made up of a mass of blinking lights. |
Copyright 2001 , by The McGraw-Hill Companies Inc. All rights reserved.
Printer-Friendly Version