What power is required to emulate the functioning of the human brain?

    According to the research of Kwabena Boahen, an IT scientist at Stanford University, a computer that fully emulates the functioning of the human brain will consume at least 10 MW of energy, which is comparable to the power generated by a small hydropower plant. Thanks to the work carried out by a group of scientists, engineers and programmers who have developed a small supercomputer, the power consumed by the human brain emulator can be reduced to 20 watts, and this is a completely different qualitative level.

    The new computer chip, Neurogrid, is based on completely different principles than traditional microprocessors and microcontrollers. But this new approach, with all its advantages, can produce erroneous results. At the moment, the probability of an incorrect result is about 30%, which, quite naturally, is an unacceptable value that significantly limits the prospects for the practical use of this technique. It is quite appropriate to note here that the human brain is also capable of making erroneous conclusions and illogical actions, probably not with a lower percentage of probability; people tend to make mistakes.

    Kwaben Bochen and his team managed to create a chip of such a supercomputer, containing a million emulated neurons. For comparison, earlier silicon chips had only up to 45 thousand silicon neurons “on board”. The Neurogrid processors, which are planned to be manufactured in 2011, will contain 64 million neurons, which corresponds to the brains of small animals such as mice and rats.

    Replacing the clear logic and accuracy of calculations with the chaotic nature of neurons can push the development of a completely new field of computer science - creative calculations that are performed according to unpredictable patterns of brain activity. Of course, from the point of view of modern computer technology, such supercomputers will never replace the usual ones in terms of accuracy and speed of calculations. But, only on such neural supercomputers will the first systems likely to have real artificial intelligence be created.

    Taken here

    Also popular now: