View Full Version : Artificial Life

27th January 2011, 11:21 AM
The development and use of brainlike technology in computers is a form of biologically inspired computing - researchers and inventors adapt knowledge gained from biology and neuroscience into their computational devices. Continuing in the use of biology for inspiration, computer scientists may go even farther, creating some form of artificial life to go along with a "brain" consisting of artificial neural networks.

A common avenue of research in artificial life is the simulation of biological systems. In this kind of research, computer programs simulate essential aspects of life such as nutrition and genetics, and researchers study the evolution of their simulated life forms. An entire "world" unfolds as the computer program runs. This research can be a lot of fun, and in the 1990s British computer scientist Stephen Grand and his colleagues created a popular game called Creatures. Players of this computer game taught little creatures called Norns how to eat, work, and survive the threats of dangerous creatures such as Grendels. The program simulated biochemical reactions along with genetics, by which the creatures changed and evolved, and a brain made of neural networks. Several versions of Creatures were released, and people continue to play and enjoy it.

Robotics is another area of research that sometimes employs neural networks and other biologically inspired techniques. For example, Florentin Worgotter at the University of Gottingen in Germany, along with colleagues at the University of Stirling and the University of Glasgow in Scotland, recently developed a robot capable of walking on two legs over rough or uneven terrain. The robot, called RunBot, has an "eye" - a sensor that detects infrared light - to watch where it is going.

Controlling the motion is a computer that runs programs to simulate neural networks, which consist of simple Hopfield-like neurons. But these networks are different from the artificial neural networks described earlier, since they are organized more like the networks in the brain. The brain generally consists of levels, or tiers, of connected networks, each of which perform a certain function and then pass the results along the chain. RunBot has this sort of hierarchical organization - a structure composed of tiers - and uses Hebbian principles and adjustable synapses so that it can learn how to walk up slopes. (Climbing an incline is an extremely difficult task for robots because it involves a change in stride and center of gravity.) Poramate Manoonpong, Worgotter, and their colleagues published their paper, "Adaptive, Fast Walking in a Biped Robot under Neuronal Control and Learning," in 2007 in Public Library of Science Computational Biology.

Intelligent behavior, such as learning how to move, is one of the goals of AI. Researchers have made progress but have yet to construct a device even remotely comparable to the human brain. And most of the "brainy" technology is based on computer simulations instead of hardware that emulates brain activity.

But these first steps are necessary. As researchers learn more about neural networks, the computer simulations become more complex and realistic. Implementing these networks in hardware, such as building a machine with a "brain" composed of neuronlike computational elements, is a much more expensive and elaborate procedure. Yet this is already occurring - even as long ago as 1960, when pioneering researcher Frank Rosenblatt built the Mark I Perceptron - and will continue to improve. This gives rise to profound questions about the nature and experience of these machines, including the possibility of some sort of artificial consciousness.