I did spend more than 10 years doing research on designing microchips - called artificial neural networks (ANN) - which mimic how the human brain processes information. I published research papers on the subject and in 1994, I published my book Artificial Neural Networks Engineering. It was a fascinating subject and those research years were most enjoyable. ANN consists of an interconnected group of transistors on a microchip that make up artificial neurons that implement a function - like voice or image recognition - based on external or internal information to the microchip proceeded by a “learning" phase whereby the microchip learns what it is actually trying to recognise. And “training" an ANN essentially means to guide it to do that function through smart algorithms - for example to recognise a certain voice or an image from among 100,000 - with the highest accuracy, with the minimum battery energy and at the highest speed. Designing ANN microchips is closely related to the modeling of biological neural systems which intimately related to the complex area of cognitive and behavioral modeling. The human brain has some 100 billion neurons which are interconnected in 3D unlike the 2D connections in the case of the microchip. These neurons are massively interconnected to each other via synapses. The interconnection map in the human brain is constantly changing unlike the interconnections between the processing and the memory cells on a microchip which have relatively simpler interconnection pattern and are almost static. While the neurons communicate in an analogue fashion, the cells on a microchip communicate usually using digital signals. For a given function, the energy needed to operate these biological neurons is extremely small and it is physically impossible to reach that limit with any one or more microchips of today. Moreover, considering the complex functions it performs and the massive information it stores the human brain is too light; it weighs only 1400 grams (2% of the total body weight in adults). The brain can be treated as a biological computer in the sense that it acquires information from the surrounding world, stores it, and processes it. The operations of the individual neurons are somewhat understood today, but exactly how the brain works, how it stores massive amount of information, how it does certain sophisticated functions like voice and image recognition with little energy and with such high accuracy and high speed remains a mystery. And how the brain expresses and stores, in the words of Hippocrates “joys, delights, laughter and sports, and sorrows, grief, despondency, and lamentations" or in short how the brain generates the human mind are all research areas that still beyond the reach of science today. Also the brain's control of behaviour based on a complex sensory input system and its use of information-integrating capabilities are research areas yet to be studied. The human brain's location in the head is no accident; it is close to our sensory organs - to maximise the signal to noise ratio - for such senses as vision, hearing, balance, taste, and smell. Today to use digital memory on microchips to store even few results of these senses during the life time of an adult as the human brain does - images, voices, tastes and smells - is prohibited in terms of cost, size, weight and battery energy needed. You can also question if microchip designers will ever emulate Brain Plasticity which refers to the changes in brain function, activity and structure in response to experience and learning or the “Flynn Effect" which reflects how general intelligence is progressively increasing with time. I am not as pessimistic as Canadian computer science professor Dr. A. K. Dewdney when he wrote, "Although (ANN) do solve a few toy problems, their powers of computation are so limited that I am surprised anyone takes them seriously as a general problem-solving tool." Dewdney in effect said that for ANN to implement complex functions - like voice and image recognition - much processing and storage resources need to be committed which would in turn increase the cost of the hardware to achieve it at high accuracy and at high speed. There are progress made in the last 20 years in the area of ANN since I wrote my book but one fact still remains unchallenged by any microchip designer: the human brain is designed optimally to process signals which are necessary to our survival on this planet earth - thank God for that.
