The brain is due to be completed in March and its developers hope it will for the first time allow a robot to interact with stimuli in its environment to develop the sort of intelligence seen in animals, becoming one of the first superstars of artificial life.

"Observers won't need a PhD to appreciate that there is a brain behind it," says Hugo de Garis of Advanced Telecommunications Research in Kyoto, Japan, the brain's principal architect. De Garis's Cellular Automata Machine (CAM) brain is being built under contract by Genobyte, a company in Boulder, Colorado.

"Real" artificial neurons

Dr Hugo de Garis's colleagues thought he was "nuts"

The brain will consist of 37.7 million artificial neurons, compared to the few hundred normally deployed in artificial neural networks. Also different is the fact that the artificial neurons are real electronic devices not simply software simulations.

Each neuron is composed of transistors grouped together in cells. These cells can also simulate the axons and dendrites that connect neurons.

Programming by evolution

Neural networks need to "learn" to perform particular tasks, but no human programmer could write the software needed to teach a network as complex as the CAM brain. Instead, this will be generated using an approach that simulates biological evolution, says New Scientist magazine, which announced the development.

The "genetic material" that describes the structure and connections of the network will be evolved over many generations of random mutations and breeding to get the optimum design.

Roboneko, being designed by Dr Korkin, means "robot child cat" in Japanese

A special computer chip inside which the connections can be continually reconfigured is needed to achieve this evolution. A new and exceptionally robust chip of this type has been made by Xilinx, a company in San Jose, California. This now makes de Garis' project a practical proposition - when he first conceived the idea many of his colleagues thought he was "nuts".

Bigger but not better

Some experts are still sceptical that building ever bigger neural networks will deliver new discoveries in understanding how our brains create consciousness. "The point is that these puzzles are not puzzles because our neural models are not large enough," argued Igor Aleksander, a neural systems engineer at Imperial College London in New Scientist.

The robot cat, Robokoneko, will not be built until tests have been completed on a computer simulation. The CAM brain's developers admit that they cannot predict exactly how it will perform when it is linked to Robokoneko.