The Information - James Gleick [119]
“When the machine was turned off,” he said, “the relays essentially forgot everything they knew, so that they are now starting afresh, with no knowledge of the maze.” His listeners were rapt. “You see the finger now exploring the maze, hunting for the goal. When it reaches the center of a square, the machine makes a new decision as to the next direction to try.”♦ When the rod hit a partition, the motors reversed and the relays recorded the event. The machine made each “decision” based on its previous “knowledge”—it was impossible to avoid these psychological words—according to a strategy Shannon had designed. It wandered about the space by trial and error, turning down blind alleys and bumping into walls. Finally, as they all watched, the rat found the goal, a bell rang, a lightbulb flashed on, and the motors stopped.
Then Shannon put the rat back at the starting point for a new run. This time it went directly to the goal without making any wrong turns or hitting any partitions. It had “learned.” Placed in other, unexplored parts of the maze, it would revert to trial and error until, eventually, “it builds up a complete pattern of information and is able to reach the goal directly from any point.”♦
To carry out the exploring and goal-seeking strategy, the machine had to store one piece of information for each square it visited: namely, the direction by which it last left the square. There were only four possibilities—north, west, south, east—so, as Shannon carefully explained, two relays were assigned as memory for each square. Two relays meant two bits of information, enough for a choice among four alternatives, because there were four possible states: off-off, off-on, on-off, and on-on.
Next Shannon rearranged the partitions so that the old solution would no longer work. The machine would then “fumble around” till it found a new solution. Sometimes, however, a particularly awkward combination of previous memory and a new maze would put the machine in an endless loop. He showed them: “When it arrives at A, it remembers that the old solution said to go to B, and so it goes around the circle, A, B, C, D, A, B, C, D. It has established a vicious circle, or a singing condition.”♦
“A neurosis!” said Ralph Gerard.
Shannon added “an antineurotic circuit”: a counter, set to break out of the loop when the machine repeated the same sequence six times. Leonard Savage saw that this was a bit of a cheat. “It doesn’t have any way to recognize that it is ‘psycho’—it just recognizes that it has been going too long?” he asked. Shannon agreed.
SHANNON AND HIS MAZE (Illustration credit 8.2)
“It is all too human,” remarked Lawrence K. Frank.
“George Orwell should have seen this,” said Henry Brosin, a psychiatrist.
A peculiarity of the way Shannon had organized the machine’s memory—associating a single direction with each square—was that the path could not be reversed. Having reached the goal, the machine did not “know” how to return to its origin. The knowledge, such as it was, emerged from what Shannon called the vector field, the totality of the twenty-five directional vectors. “You can’t say where the sensing finger came from by studying the memory,” he explained.
“Like a man who knows the town,” said McCulloch, “so he can go from any place to any other place, but doesn’t always remember how he went.”♦
Shannon’s rat was kin to Babbage’s silver dancer and the metal swans and fishes of Merlin’s Mechanical Museum: automata performing a simulation of life. They never failed to amaze and entertain. The dawn of the information age brought a whole new generation of synthetic mice, beetles, and turtles, made with vacuum tubes and then transistors. They were crude, almost trivial,