Click here to load whole tree
NITAAI-Veda.nyf > Soul Science God Philosophy > Science and Spiritual Quest > Section 2 Machine, Mind and Consciousness > MOLECULAR INTELLIGENCE > 4. Computations Cannot Evoke Intelligence

4. Computations Cannot Evoke Intelligence


Although intelligence remains an elusive concept to the laws of physics and biology, still researchers in the field of Artificial Intelligence (AI) are managing to make ever more powerful and  'intelligent' machines. For example, Deep Blue [2], the chess-playing computer developed by IBM defeated the world chess champion, Garry Kasparov, in 1997, and this year (2007) researchers  at the University of Alberta developed the perfect checkers playing program, Chinook [3].If intelligence is not physical then where does it come from in these computer programs?  Could   it  be   that  complex  computations  can  give  rise  to  real intelligence? Or, is it that playing games like chess and checkers does not involve real intelligence?But this does not mean that a machine  also uses insight and understanding while playing these games.We think that most of our readers will agree that playing chess or checkers does   involve   a   lot   of  intelligence,when humans  play the game. But this does not mean that a machine also uses insight     and    understanding     while playing these games. Take the example of chess. The rules of this game are very

well defined, so that a machine can easily be programmed to play it correctly.


Let us explain how a computer plays chess. Once the rules have been programmed into the software, the computer tries to find the best next move. It simulates performing each possible move  and the next move and so on for some N moves into the future. For each possible outcome, it evaluates the board position. The winning board positions can easily be encoded into the program.  For other board configurations, different figures of merit can be defined based on the opinion of human chess masters. Then, the machine simply has to select the move that leads to the best  outcome. This algorithm is called the min-max algorithm. Several variants of this basic approach are used by most chess playing programs.Now suppose that at each step in the above  algorithm, there are M possible moves that the machine can chose from. Then the search space for the machine becomes MN, a number that grows exponentially with the number of steps N.


We use our past experience and understanding of the game to narrow down the search space to only 10-20 different outcomes.


Those of us who play chess know very   well   that   humans   do   not evaluate so many moves to decide the next move. We use our past experience and understanding of the game to narrow  down the search space    to    only    10-20    different outcomes.   Furthermore,   even   to judge the board position requires understanding, an understanding that was programmed into the  computer, based on the intuitions of chess grand masters.So playing chess against a computer means playing against these chess masters, now equipped with a gigantic calculator that can  explore a much bigger search space than what is possible with the human brain. This should explain why we find it so hard to defeat even a PC chess program. Now imagine a parallel computer  with 256 dedicated chess processors cranking out 200 million chess positions per second—the Deep Blue.Playing chess requires both understanding and computation. The understanding is  smuggled into a chess-playing computer as board evaluation metrics taken from human chess masters. All what the machine really does is computations based upon these algorithms.

Let us now explore two extremes of the computation-understanding spectrum: fully computable games and computationally intractable games. Checkers is an example of a fully computable  game. Because of its simplicity, all possible sequences of moves can be exhaustively enumerated. Then, a computer that is fed with this database can easily choose the winning move for every  possible board position. This is exactly what researchers at the University of Alberta achieved by running a checkers simulation for twenty years! They produced a program, called Chinook,  which guarantees that it won't lose the game. At worst, the game will end in a draw. Their paper was published this year in the prestigious journal, Science [3].


When computers perform tasks normally associated with human intelligence, they do not use intelligence or understanding to execute these tasks. Rather, they rely on sheer computations  without any understanding of what they are doing. Even we human beings sometimes perform tasks without possessing any understanding of what we are doing, like walking, driving, or following  directions. Prof. John Searle formalized this concept in his "Chinese Room Argument" [4],