HUMAN INTELLIGENCE AND THE LAWS OF PHYSICS :
SAMEER GAURIA (I.I.T. BOMBAY)
Intelligence what is it? What is it about intelligence that is hard to understand and replicate artificially? Why is that despite leaps and bounds in all other fields of knowledge this field is virtually virgin ground and stays that way?
The seat of human intelligence is the brain; the human brain is a veritable beehive of activity- sifting through vast amounts of data every waking moment. The bulk of of the activity is just computational - converting the data to information, sifting through it rejecting what is unnecessary, responding to what is important. But all this work is essentially algorithmic. Algorithms are deterministic by their very nature, thereby implying that the brain is essentially algorithmic; and all its activity can be accounted for thus. That is, the brain is just an immense machine that executes some huge algorithm. Algorithms can account for speech, pattern recognition, vocalisation, reflexes and repetitive motor activity.
Computationally the brain out-performs any neural network by several orders of magnitude. But it is inconceivable how algorithmic activity could explain solution by insight- the so-called Aha experience. All of us have experienced these at one time or another. These essentially bypass all algorithmic activity and present us with a solution straight away. An interesting feature of these experiences is that there is near complete surety about the solution when it comes. One knows that it is right even before verification. In fact the verification has a feeling of redundancy about it.
We are thus inevitably led to the solution that the explanation to intelligence lies elsewhere. There must be a non-algorithmic pseudo-random aspect of intelligence. Pseudo-random in the fashion that we can direct our thoughts but not control them or predict them. An important point is that the brain esoteric as it might seem, is well and truly a part of this world and must obey the laws of physics just as surely as rolling balls and moving blocks. It is worthwhile to consider whether the present framework of physics can give rise to a theory of intelligence and free will.
The thrust of progress of physics has been the conviction that there are universal laws that govern the universe and these laws can and eventually will be discovered by man. There is also the belief that these laws will be remarkably 'aesthetic'. In fact Dirac and Maxwell said that it was their 'aesthetic' sense that guided them in the formation of theories. Einsteins Theory of Relativity cannot be paralleled in the simplicity of its assumptions and the extent of its predictions. It is close to perfect. By contrast Quantum Mechanics does not have much of this beauty or aesthetic justification. But it works (and how!!?) In any real system it is impossible to specify the initial state precisely. And the initial impreciseness increases as time passes by and very soon the state of the system is as good as unknown, thus the evolution of the system though determined by the laws of physics is not computational.
But the laws of physics are deterministic. These laws precisely determine the evolution of any system. For a given initial state the future of the system is completely determined by these laws. But this evolution is non-computable. There is actually one physical process that is not deterministic. In fact it is completely random and that is the process of making a measurement on a quantum system. The latter may exist as a superposition of several states but when we make a measurement it is forced into one of these states.
What are the implications of this for intelligence in general and artificial intelligence in particular? On one hand we have laws that are completely deterministic and rule out any possibility of free will and the pseudo-random aspect of intelligence. And we have another law that is so completely random so as to tender it quite useless from this point of view.
Clearly we shall need changes in the fundamental laws of physics before we can come up with any theory of intelligence. We will need a theory that unifies these laws and manages to bring about a compromise between the randomness and determinism of various laws. It should be able to incorporate a certain sort of feedback through which a system can determine it's own evolution. Such a grand unified theory would possibly require a truly quantum jump in concepts; an absolutely new way of thinking about how the universe functions and how we ourselves are a part of it.
Back to main.