But when computer scientists actually got down to creating such machines, they realized the problem was much larger than they had imagined. Conventional computers find it very difficult to do basic things that people do. When we look at the picture of a Hyundai Accent or a Tata Indica, we have no trouble identifying it as a car. But even such a mundane task is extremely complex, and computers struggle to even recognize it as a car, let alone recognizing the make or model. Tasks which are second nature to us are, as researchers learnt to their dismay, a series of very complex transactions between neurons, and simulating this task on a computer is very difficult. If you have ever tried to use a voice recognition program, you’ll know how computers struggle to identify speech.
There are primarily two reasons why the evolution of Artificial Intelligence hasn’t kept up with expectations.
The first is the lack of computational power. Today, some mobile phones pack more power than the supercomputers of the 1970’s. Although computing power of processors continues to increase rapidly, it falls woefully short of what is required for an AI program. Supercomputers like “Deep Blue” which are capable of performing billions of calculations per second, still cannot win at chess convincingly. If we want to simulate “thinking” today, we’ll need a computer the size of Earth. Something revolutionary like room-temperature super conductors or quantum computers is needed to satisfy the power required by AI applications.
The second problem involves the understanding of “intelligence” itself and the techniques involved in writing “intelligent” programs. It is widely believed that we lack a computer language to express “intelligence”. Several languages were created with AI as a primary design goal, but all these have failed to come up with anything substantial. The problem seems to be more fundamental. Is it possible at all to express human intelligence in a mathematical form, which can be simulated by a computer? This question can evoke a wide variety of answers depending on who you ask.
Some researchers now believe that human intelligence cannot be simulated by machines at all, and say that machine intelligence is fundamentally different from human intelligence. These people are now focusing their energies on practical applications like industrial robots, where the “intelligence” required by a machine is in a very specialized area, and is readily available and well understood. They are creating specialized machines that are extremely good in their assigned task, but cannot do anything else.
Despite all the problems, some other researchers refuse to give up the dream of creating a general-purpose intelligent robot. They are now trying out new techniques. At the forefront of these new technologies are neural networks. These networks try to simulate the functioning of the human brain, which is a complex web of interconnected neurons, eventually hoping to simulate human intelligence. They are also trying to get computers to learn, so that the machine can discover intelligence itself.
Whether or not such machines become a reality is debatable, but the field of Artificial intelligence promises exiting action in the years to come.