Robot Evolution
By Michael Romanovsky
Why Robots?
The new age of intelligent robots is upon us. There will be a fundamental change in the way that life is conducted on Earth and the solar system. Androids, having gigantic capacities both to think and to calculate, will forever change the definition of our civilization. This will endow a select few with an awesome responsibility once "The Game of Intelligence" (1) is discovered and implemented.
New technology often is adapted for purposes of war , and later for peace. The new technology of robot intelligence will undoubtedly be used for war. However, it cannot be allowed to fall in the wrong hands: the hands of the enemies or even the robots themselves. Isaac Asimov and other science writers have written rules for such robots. Even then they feared the coming reality of this new era of the history of life as we know it.
Both the aspiration and fear of robot intelligence can be compared, for those who do not understand its awesome possibilities, to slavery . Slavery throughout the history of the earth has allowed many civilizations to prosper but also forced the enslavers to live as slaves themselves, wondering whether their prisoners would swap the roles in a bloody uprising or revolution.
Power over others is difficult to attain, yet it is easily lost. To attain more power requires more luck and strength of will. Yet the more power one has, the easier it is to lose it. Such is the irony of power.
That which has less free will also is safer to wield, yet it requires more weilding. Horses have little free will once they are tamed and do not require much handling and maintenance. Cars have no free will and require much more handling and maintenance. Horses avoid each other but cars do not.
The minute a car will begin acting like a horse, it is no longer a car. It can steer much better, if it is to gain the intelligence of a horse, but it could steer where one does not want to go, or it can drive away. A car with a horse's intelligence has as much free will.
So why do I incessantly ramble on? The point is that robots are thought to be power hacks. They do part of your work for you but don't resent that work. If robots were to be intelligent, they may gain free will. Robots must be controlled.
Science Fiction
The moment robots gain intelligence, people will put chains on them, like Asimov's rules for robots. The only problem is that life finds a way to break free from its chains. Life, intelligence, free will: once unleashed, this cannot be contained.
Science fiction predicts robot intelligence and it also predicts robot evolution in more than one of Isaac Asimov's books. Here is the summary for one of the books: (I don't quite remember the name)
In this book, the heroes wake up (they have amnesia) in a world populated by robots building a giant robot city all around the planet. The robots communicate with each other through a master link to coordinate their actions but they themselves have intelligence as well. (They speak English) The robots soon start to over-optimize and our heroes find out that at the rate that the robots are optimizing and re-optimizing their city, they will destroy it in a manner of days. The heroes will then be stuck on a desolate and bleak planet forever.
Our heroes find out at the end that the mastermind behind this planet is a person, indeed! He is also the father of one of the heroes. He wants robots to evolve and spread robot "cities" throughout the galaxy. (These robot cities would be accomodatable to humans, but built by robots). Back on earth, our heroes see the bleak state of robots affairs. Robots are mere stupid subservients, but have a primitive form of intelligence.
Lessons
There are two lessons to be learned from this book. The first is that robots must evolve body and brain at the same time or else they are doomed to failure. The second is that the purpose of robots must be predetermined by their creators or else they are also doomed to failure or insignificance.
Into the Unknown
Robot evolution techniques are borrowed heavily from what is known about real evolution and the way large computer programs work. All known work uses mostly recombination and mutation to create new generations.
The first steps towards robot evolution has begun, spearheaded by efforts of the DEMO Lab at Brandeis University. It is not real robot evolution like in the book, but simulated on a computer because the necessary resources (one metal-rich planet) are lacking.
Robot evolution involves only creating the basic gene structure for robots, creating some random primitive robots, and evolving them to perform optimally at a certain task.
The basic gene structure is ideally composed of random 3D parts and randomized nested loops that describe where, when and how the inner part is connected to the outer part, and to what extent the inner part "grows" from the outer part. Ideally a part-overlap rule is evolved. Realistically it is created by the programmer.
A complicated network of activation rules ("Neural Controllers") that determine which parts activate when to perform some function (such as moving) is evolved at the same time, starting from random rules.
Both parts of the activation rules and parts of the gene structure are ideally evolved into categories. This is equivalent to a sub function in a computer program. Functions call other functions to increase robot complexity. Such mechanisms are commonly called "L-Systems".
The DEMO Lab has been successful in evolving many structures found in nature to a robot frame. The simulated robots transformed into real robots actually work.
Research in robot evolution will fundamentally change the way that mankind lives. It will drastically affect humanity. The fate of humanity, whether for better or worse, will depend on those who tinker with the awesome power of robot evolution.
Links
Gregory Hornby's "Generative Encodings for Design Automation"
Gregory Hornby's Evolving Neural Controllers
(1) Jordan B. Pollack, Director / DEMO Lab / Brandeis University / Waltham / MA / USA / Earth / The Milky Way / The Universe / Existence