‘Man vs. Machine’ is the most favorite subject of Hollywood moviemakers. You might have enjoyed a large number of movies on this theme. Many science fiction stories and books also sell the same concept. They all portray the idea of using computers to replace humans in areas of sophisticated decision-making. It has been argued, at times with more emotion than logic. The reasons are easy to understand but very hard to justify in a logical manner. So, to explain the concept in a simple way, I’m taking the example of Chess, the most sophisticated mental game ever developed in the history of mankind.I think these concepts can also be applied to almost all other scenarios.
As we all know, the task of programming a computer to play great chess has assumed a high priority in the field of Artificial Intelligence. Let’s first try to apply the finite state theory to these two thinking entities. The human brain and the computer both can be viewed as finite state systems. The number of brain cells or neurons is limited, probably 235 (two raise to the power thirty-five) at most and the state of each neuron can be described by a small number of bits. If so, then finite state theory applies to the brain.
However, the number of states is so large that this data is unlikely to result in useful interpretations about the human brain. In the computer also, the state of the central processor, main memory, and auxiliary storage at any time is also in one of a very large but finite number of states. So, from the hardware point of view they can somewhat replace each other.
Now see the difference; the human method of chess calculation puts heavy emphasis on pattern recognition. His experience of all past games, whether played or read, is processed in a manner, which leads to form his chess concepts. The silicon monster plays chess in a totally different way. Since the game is finite, each move has a limited number of options and the length or the game has an upper bound. A sufficiently fast processor should be able to calculate it all out, through to checkmate or a draw, just by analyzing the given position.
However, this approach is very over optimistic. Computers can indeed perform large numbers of calculations in a short time (can calculate several million moves per second), but the possibilities involved in Chess quickly become too high.
The programmer should first introduce some means of limiting those variations considered. He has to give the computer some idea of what a fine move looks like. On the other hand, the human player, by his experience of the game, develops an almost instinctive feel for a good move. This limits the calculations to a few hundred moves.
These are primary limitations in the ability of silicon processors to simulate human thought process; the two halves of the brain have distinct modes of thought. In right-handed adults, the left cerebral hemisphere controls logic and appears to face overwhelming difficulties when posed tasks of which the resolution does not admit any verbal description. The right side of the brain performs pattern recognition, like recognition of faces, shapes and colors. These may be performed with ease by right hemisphere, though the left cerebral hemisphere is unable to describe in words the process of recognition.
Digital computers may be intelligent enough to simulate the processes of the left cerebral hemisphere, but visual-spatial capabilities of the right side are very hard to be reproduced by motorized means. After the machine has been taught to identify all legal moves, it needs an algorithm with which to select its moves. The digital giant can only ‘think’ in terms of numerical terms, so at the nucleus of its chess understanding must lie a Positional Evaluation Function (PEF), a system of putting numerical values to the features of a position and then joining them in a manner which shows who stands better and by how much.
A human player considers many positional features while forming his judgement of the position, so the computer should also consider all these positional features. For example, control of the center, control of the rest of the board, material, king safety, piece mobility, attacking possibilities, pawn structure, time pressure and so on, all these positional features deserve numerical values. The machine then must build its tree of analysis to decide which positions should be assessed in this manner.
By using accurate ‘tree-pruning’ techniques. the computer can save its precious time in calculation. In this technique,irrelevant branches of the tree may be chopped off and overlooked. The programmer’s main responsibility at this stage is to grow the right tree and assess all the positions accurately. But this task is not as easy; let us consider the typical problem that he faces.
All computers have very strong Arithmetic and Logic Units, so the direct calculation of forced variations is a cakewalk for them, but even there remain severe problems in this area of computer chess thought. What is required is a workable and effective rule for deciding when a line of analysis has proceeded sufficiently deep to be considered terminated. A fixed depth of search is dearly insufficient. It might terminate its calculation in the middle of a tactical fight or just before checkmate.
On the other hand, a human player knows by his experience and intuition that when to stop or when to calculate more deeply in the given position. The computer suffers from horizon effect, a shortsightedness that stops it from looking any further.
The main problem lies in a correct definition of quiescence. It is a state in which nothing tactically significant is happening. When the position is quiescent, analysis of forced variations stops. However, a mechanical definition of that type of state is very tough to find. This preserves its Tree of Analysis at a controllable yet effective size.
As we have seen, brute force rules in the world of computer chess. It just calculates moves based on its Positional Evaluation Function and wins because of its raw power of number crunching.
This situation is very unfortunate because it doesn’t contribute anything valuable in our knowledgebase of human thought processes.
It seems that computers would finally manage to beat the best human players, and If they continue to play using present style, it would definitely increase our understanding about algorithms but not about our mode of thinking.