American Inventor, Eat Your Heart Out.

There is a new up and coming inventor on the block. Stanford Professor John Koza has combined the fields of Artificial Intelligence and Genetic Algorithms to create a machine that is capable of invention. The "Invention Machine" as it has come to be named is a Beowulf Cluster of now dated home computers. The hardware is unremarkable, it's the software that makes the magic happen.

But first a little background on Genetic Algorithms. A Genetic Algorithm is essentially a computer program that uses evolution as a way to find answers. A Genetic Algorithm examines a very large set of possible answers. It then culls the absolute worst from the lot, and spawns some new ones based on the best characteristics of those that remain. The new generation of possible solutions is ever so slightly better than the previous one. Most Genetic Algorithms also have code to introduce random "mutation" into each new generation. Creating new possible solutions the same way nature does, with the occasional minor change made at random. Mutiply that process by a million generations, and the best possible answer begins to evolve. Eventually through the culling, breeding and mutation of solutions, only the fittest survive. This is how Genetic Algorithms are able to derive answers to often complex problems. John Koza has taken the Genetic Algorithm process a step further. Through Genetic Programing he has given his machine the ability to genetically evolve it's own programing.

For example; Lets say that you need a some new optics designed for a telescope. They have to to have specific characteristics, but other than that, you have a clean slate. Proffesor Koza puts the target specifications into the Invention Machine and sets it to work. With a typical Genetic Algorithm, the machine would spawn a billion possible optics, and then through a few million generations evolve the best one for the job. But Koza's Invention machine actually creates a myriad of possible optics creation programs. And puts those programs through the genetic evolution process. By examining their output, efficiency, and various other factors, the machine treats each program like an organism. Some are free to breed and thrive,and some are mutated. The least fit are given the old yeller treatment. Eventualy, not only does Koza end up with a spectacularly functional optics design, he also ends up with an evolved computer program optimized for designing that optics. The results of this non-hypothetical are a confirmed success. In addition, The Invention Machine was given a design to avoid. Patented optics that Proffessor Koza didn't want to infringe on. The machine made a successful end run around the off limits design, and came up with something exactly as functional, but completely different. Machine 1: Patent lawyers zero.

Genetic Programming has been a success all around. New inventions are being created constantly now, taking between a day and a month to evolve. An antenna designed from genetic algorithms is now floating some where out in space. NASA picked it over it's human designed competitors. It out performed them all, and took less time to design. In fact it's design was so counter intuitive, that I'm hard press to envision a human ever coming up with something like that. The invention machine has since come up with new circuits, radios, and even parallel designs with things that humans had already invented. Most successfully, the Invention Machine was issued a patent for a process to increase the efficiency of a factory. Machine 2: Patent lawyers 0. That means that some where, there is a room with 1000 computers in it, that is a registered and legal protected holder of intellectual property rights. A cluster of computers that is legally recognized as creative. Thier methods may be unusual, but we have officially entered the age of creative computers.

This presents us with an interesting question and an interesting opportunity. First I am left to wonder if a computer can actually produce intellectual property. patent and copyright exist (in theory) to protect the rights of creators. But by definition, a machine has no rights to protect. You cant sue a computer for infringement, nor can you pay it royalties. A computer is in fact incapable of even recognizing it's work as creative. And yet here we are. So to fix it, do we change the patent law? Either that or grant the Invention Machine partial human rights. Or perhaps we just need to redefine creativity to include both the result and the method that gets you there. My suggestion would be to simply assign any patents granted to machines directly into the public domain. That way everyone could benefit from this amazing technology. And I'm sure the machines wont mind.

Which brings me to the opportunity. Right now Genetic programing is rare. It exists in a few scattered machines around the world. But this isn't by necessity. The hardware that powers the Invention Machine is far from state of the art. I propose a distributed computing network resident invention engine. Genetic programing run in bits and pieces on computers all over the world. If something like that caught on, it would put the Invention Machine to shame. Solving in minutes what it takes the Invention Machine months to do. You could sign on like you would with protein folding or SETI at home, and ad one more piece of evolving code to the mix. Tens of millions of multi-GHz computers, instead of a mere thousand aging first generation Pentiums. Members could suggest lists of new problems to solve or devices to create. A universal database of materials, existing designs, and manufacturing processes could be constructed and added to wiki-style. And again, all creations would get patented and piped right into the public domain. A disruptive technology to end disruptive technologies. Eventually you could work in garage fabricators, or something like RepRap. But that's a post for another time.

Suffice it to say that I'll gladly donate any of my spare processing cycles to inventing new hybrid motors, water purification devices, solar panels, computer chips, or jet packs. Proffesor Koza, the ball is in your court. How about it? [Inspired and Via Popular Science]



"He's the guy who patented using genetic algorithms to patent everything they can permutate from an initial
description of a problem domain – not just a better mousetrap, but the set of all possible better mousetraps." - Accelerando