Tuesday, April 6, 2010

Inherent Variability: A Major Difference Between Man and Machines

Image (a)
Image (b)
Image (a) shows my attempt to write alphabet 'a' the same way four times. Image (b) shows how a machine does the same. Obviously, we can see the machine is precise in doing this. A loud applause for the machine!

Image (c)

Image (c) shows a hand written word, in two different ways. The word can, obviously, be recognized by any human reader. How good is the machine in doing the same task? This is how a captcha differentiates man and machines (though it is not exactly written by human beings, but some algorithm that can generate and display words with some distortion, which is difficult to be recognized by machines and can be easily recognized by humans).

Any written alphabet, spoken word, action performed by a human being is a stochastic process; i.e., it is an output of a random function. Apart from the variability in doing things, we are built with an inherent ability to recognize and learn from the variability.

Though both the images on the right and left represent the same great person, we can easily recognize the variability and if we try to draw the same image again and again, we will end up in producing a different image every time.

This shows  our inherent ability to produce and recognize the variability. This is very, very difficult for a machine to recognize this as a same person.

Watch a 6-months-old baby (almost a fresh brain, of course, though trained for a few tasks and far better than the best known robots), give her a toy and she feels happy for some time, and then searches for a new one.

Have you ever wondered why we don't feel happy watching the same movie more that a few times?

It is all the part of the human intelligence architecture: the brain. Our brain (still we are not in a position to exactly know its architecture) contains 1011 neurons and 1015 synapses. Each new activity changes the whole network. Each time you watch a movie, you are watching it with a new neuron-synapse network. Each time we watch, we look at some new things which we missed in the other part. The watched part goes automatically into the sub-conscious mind, and the selective mechanism is now focused on the new variable part which goes into the conscious mind. Each time our mind expects some learning. When this learning saturates or if we couldn't find any interesting variability any more, we feel boredom. So if any movie is enjoyable for quite a number of times, it has a lot of likable-variability in it.

There are a few generalization algorithms that are used by machines to generalize identification of objects.  For e.g: Asimo, one of the worlds best humanoid robot can generalize few objects based on its characteristics. Obviously, not as good as humans. Humans work on very high dimensional space to find  or understand the variability and use A-priori knowledge, which makes the difference in identification.

Even the motor control in human beings has the variability in every action we do compared to the machines which can just do any action based on the limited controlling parameters.

The reason why television, computers are still interesting is the variability of shows in the  TV  (compromising multi-dimensional parameters into a few dimensions) and computer being one device having capability of running different applications providing variability. Overall, the inherent variability makes our lives very interesting, which is very difficult to achieve by the machines.