Creativity is often considered the last bastion of human superiority. While machines have surpassed us in all types of physical tasks, most people presume that they can never become creative.

The Free Dictionary defines “creative” as, “having the ability or power to create; characterized by originality and expressiveness.” Most of human progress originates in creative thoughts, whether useful inventions or new ways of looking at the world.

Can machines do this? It depends on whom you ask. When Deep Blue defeated Garry Kasparov, most of the grandmasters present agreed that some of its “ideas” were truly original and groundbreaking.

Those “ideas” were developed by brute force trial-and-error; hundreds of millions of scenarios were compared to find the optimal ones. This does not accord with most notions of the creative process.

Yet Thomas Edison found the first viable light bulb filament after a reported 10,000 trials. My friend Morgan Westerman “invented” his website www.theinterviewwithgod.com (a creation that has been viewed by 50,000,000 people) by considering and discarding 3,000 images to select 16 worthy of his online slide show.

Is that so different?

Judgment is involved, but clearly machines display sound judgment, at least in limited domains. Computers have been programmed to perform medical diagnosis better than 90% of physicians. You’d do better “hiring” such a program than selecting a doctor at random from the yellow pages.

Now comes a program that may finally put to rest the notion that creativity is exclusively human.

Popular Science reports that programming expert John Koza has built a next-generation “thinking machine.” It doesn’t just solve puzzles; it invents new things.

How new? Some have passed the ultimate test for novelty. They’ve been awarded patents.

Koza has taken a radically new approach to programming. To understand this, we need to first understand genetic algorithms.

Conventional genetic algorithms work on a random evolutionary basis. A set of rules is established to govern changes in a virtual “space.” Objects are introduced that “behave” in certain ways — each different from the rest.

In a God-like fashion, the programmer defines optimal behavior and those objects that behave closest to this ideal survive into the next “generation,” where the process is repeated again and again, generation after generation.

In each generation, the “best” algorithms of the previous generation are matched against a new set of “mutations.” Just as in nature, the existing “genotypes” usually survive — but occasionally a superior accident arises and displaces its predecessor. This is “evolution.”

On a practical level, this has been used to optimize trading rules. For example, 100 different moving averages might be introduced to a “space” consisting of S&P 500 price data for the past 25 years. A set of buy/sell rules would apply to all of them.

Optimal performance would obviously include profit maximization. It might include secondary factors such minimal drawdowns and a high percentage of winning trades.

Of the 100 initial candidates, the best performing moving average would survive to the next generation, where it would be matched against another group of randomly generated moving averages — or other kinds of trading rules.

Over thousands of generations of trial and error, some pretty sophisticated and effective trading rules can be evolved this way. Billions of dollars are now invested as a result. Some of your own money may even be allocated to rules that were “artificially evolved.”

Koza’s genetic algorithms take this process a step further. He likens it to intentional breeding, “If you’re trying to breed a better racehorse, you could take a herd of horses out into a field and irradiate them and hope that, through random mutation, you get a better horse,” Koza explains. “Or you could mate a champion with another champion.”

As opposed to starting with raw electronic components, his computer was programmed to design circuits based upon the attributes of established circuit designs. It quickly “invented” a host of viable new circuit designs. He knew they were viable because they infringed patented designs — in effect, the computer was rediscovering what others had invented.

Koza reportedly said, “That’s when we began to see that genetic programming could be human-competitive,” he says. “If you remake a patented circuit, you’re doing something that people consider inventive.”

On January 25, 2005, a U.S. patent examiner unknowingly evaluated an invention submission from a computer. In addition to novelty, patents must be non-obvious — admittedly subjective yet a hallmark of true creativity. Patents have now been awarded for computer inventions.

Jason Lohn, a student of Koza, has applied his work to design of antennae for NASA. According to Lohn, antenna design has traditionally been dependent on human intuition.

He used Koza’s approach to antenna design. After entering the characteristics of known successful antennae, he let the software run. After several hundred generations, it output a strange design. It was crooked, and resembled a bent paper clip.

Lohn could find no comparable antenna design anywhere in the literature. Yet this unique design met NASA’s requirements for a wide bandwidth and a wide beam…and it’s going up into space.

In the view of Popular Science, the power of computers to create will shift the human contribution from the experimental side of invention to the conceptual side. This may be the ideal match of human gifts and computer abilities.

My friend, the inventor Jerry Smith (who had invented some of the most secure locks in the world as well as a new type of magnet, among hundreds of other things), opined that invention is mostly a matter of fully understanding the problem. This understanding may be the true last bastion of human creativity.

Today, scientific researchers are realizing more than ever the power of questions. By asking the right questions of their artificial assistants, important discoveries can arise.

Our Transformational Technologies Portfolio holdings Symyx Technologies (SMMX: Nasdaq) and Accelrys (ACCL: Nasdaq) offer tools that accelerate the discovery process for new compounds and materials. These tools can perform millions of virtual “experiments” in the time a human might perform one. Some wonderful new materials and drug compounds have been discovered.

The results are so solid that scientific journals accept “virtual discoveries” made this way even though they have no physical existence. (Consider the most recent discovery of a superconducting material that I reported to readers: It has yet to be made in the lab, yet was accepted for publication in the journal Physical Review B).

Many Fortune 1,000 companies involved with chemical and biological processes are relying on these tools to accelerate their discovery process, reduce dead ends and gain a competitive edge.

Artificial creativity is rapidly becoming a very big business.

To your profitable future,

Jonathan Kolber
May 17, 2006

Artificial Creations Can be Very Valuable was originally featured in the Penny Sleuth.