Intelligent AI programs do not need algorithms. They require methods in order to solve problems in a non-computational manner. How AI will achieve this will decide the fate of machine intelligence in the future.

The human mind, the seat of intelligence, is a great mystery which science has not yet been able to fully fathom. But intelligence it not unique to human beings; there are many phenomena in this universe that reflect intelligence. Science has been trying to unravel these mysteries through careful observations and experimental investigations. Scientists have painstakingly unlocked many secrets of nature and are now trying to develop artificial intelligence resembling nature according to the results of their experiments. They have already enjoyed a good measure of success.

The artificial intelligence programs being pursued by many scientists all over the world seek to simulate human problem solving and learning abilities. Inherent to this approach is a certain conceptual distance between artificial intelligence and biological or organic intelligence. Though scientists may be able to get more or less the same result, it still remains only a simulation of real intelligence. Therefore, artificial intelligence bears no resemblance to natural intelligence whose many dimensions are beyond the grasp of science.

The success so far has only been in the simulation of intelligence; not the creation of true intelligence. And to that extent, artificial intelligence is a misnomer as those dedicated to developing a true intelligent system are yet far from the real thing. The difference appears to arise because in artificial intelligence there is only a simulating system, which seems to embody the hypotheses about the nature of intelligence.

The question is in what manner to scientifically approach artificial intelligence in order to be able to come very close to organic intelligence. For this, it is important to elucidate phenomena of the brain that dictate intelligent behavior in humans. These show up as intellectual functions that have to be performed in order to complete a task. Since any scientific understanding must be enunciated in the form of a theory, those attempting to mimic human intelligence must first put up in theory the phenomena of the brain. This would certainly make the starting point adventitious. And it is this complexity that explains why different scientists start at different points - simply because the subject matter is so comprehensive.

What is intelligence? Intelligence characterizes the way human perform tasks. Intelligent behavior implies successful performance of a task. This implies that a capability is necessary to perform a task. But, all humans have capability and still not all are able to do a particular task. Sometimes this recognition reflects degrees of ability – all humans have more or less difficulty in doing a task and accordingly exhibit more or less intelligence.

Can there be a credible theory of intelligence? A good theory specifies a set of principles or prepositions or equations so that a scientist skilled in the art can use them to make a prediction on the basis of a theory. Such a theory would have to specify how intelligence operates and what constitutes intelligent behavior.

But these points cannot be applied to theories to define other aspects of the mind. Intelligence is only a functional capability of the brain. What about other things like emotion, pain, dreams, extra sensory perception, which are also functional capabilities of the brain? To predict a functional capability is to be able to produce the function by other means than the original, but, nevertheless, to produce the function. The theory to explain how humans perform their tasks will be a system for performing a task intelligently. The phenomena enumerated are themselves functional capabilities, from perceiving to extracting implicit knowledge. So, the system will predict that human exhibit these phenomena by themselves exhibiting them. The development of artificial intelligence has brought into focus the diverse nature of intelligence. But, a theory that merely highlights functional capabilities of intelligence will not serve the purpose. It should also conform to human intelligence. Definitely not in ways that are far from human intelligence, yet call it intelligence.

Intelligence in reality points out that the problem solving approach may not be perfect when it is situation oriented. Weather forecasting is one such example - we know all the forces governing various climatic factors, their variations et cetera. However, predictions often deviate widely from the actual observations. Thus any intelligent approach should be as flexible as possible. This will leave room not only for various possibilities but it can be modified according to the situation.

The neural system is one such system, which has gained considerable popularity in the recent years. This system works the way the human brain works. The electronic implementation of the idea of brain plasticity (the brain’s capability to modify the strength of connections among neurons to improve performance) makes it possible for this system to use its own procedures. Therefore it will be different from biological procedures used by the brain to vary the strength of connections among neurons.

A basic feature of the neural network, distinguishing it from the conventional algorithmic programming, is that rather than having a previously given algorithm – that is specifically provided to solve some particular problem or a class of problems – we are instead provided with merely a loosely connected family of units (the electronic neurons) where the strength of the various connections are continually changing in order to maximize the quality of the output. In this way, the system continually learns, improving the output all the time. The action of such a system is still algorithmic since it can be manipulated and implemented on an ordinary general-purpose electronic computer. However, there is a crucial difference in the underlying algorithm. Now the algorithm – according to which the system acts at any time – is not the one that was fed initially to provide a specific solution to a pre assigned problem, but it is the one that is gradually evolving on the basis of a bottom up approach, which implies continuous improvement.

But the procedure, as a whole, constitutes an algorithm, provided that the judgment – as to the quality of the output at any given stage – is made according to some algorithmic criterion. This would indeed normally be the case if one is thinking of using the neural network to solve problems in pure mathematics; its action would be entirely algorithmic. That is why we need a non-algorithmic approach to our problems with a little bit of algorithmic calculations that can be manipulated in a non-computational manner. Only then can we unravel the mystery of the thin line that divides man and machine.