The Age Of Spiritual Machines Essay Research — страница 2

chess — is run, creates a “tree of possibilities.” In the context of chess, the computer examines the board and consults its database of possible moves. It looks ahead a half a move at possible moves made by its opponent after its own possible move, limited only by the speed of the computer, which is dramatically faster than the human brain. When it sees that the next half-move is mutually damaging, it reaches a “termination leaf,” understanding that either it has won or lost. Of course, the limitation of such a routine is its inability to discriminate, to find patterns, and to know which branches he should ignore altogether. While Deep Blue has trumped human-kind at it’s game, he has done so through a very unhuman process. That doesn’t make it any less of a defeat,

but realizing that helps to put it in context. Let me elaborate. Deep Blue was meticulously programmed how to choose a path at each branch in the road. Presented with a slightly different or new situation, it would be unable to adapt. Change the rules a bit, and Deep Blue is helpless. That is where the next form of processing takes over. This second form of advanced processing is the Neural Net. As its name would indicate, it is based upon the neuron operations that occur in the human brain. A Neural Net is comprised of simulated software or hardware neurons that are randomly assigned elements of an input such as a picture. Kurzweil uses the example of a banking machine that can identify a human face: like the human brain, it remembers different parts of a person s face in each

of neurons. Since we use human brains, we do not remember an acquaintance s appearance in one lump area of the brain; our memories are randomly dispersed (in the case of visual memory the brain randomly stores color, shape, detail, etc.) to access that information by bringing it together upon input, or seeing the acquaintance. The computer recognizes a person s face when each neuron with the familiar information fires and triggers an output neuron on top of it. The output neuron then gives the command to display “This is James Bond, Agent 007.” What separates Neural Nets from say, a disparate Recursive system, is that it is not programmed in a linear fashion. Instead the Neural Net is taught and conditioned. When it is introduced to the face, it remembers the information in

arbitrary neurons and then remembers that the information is to be associated with James Bond. If the face is presented again, and the information returned is false, then the net is conditioned to respond more efficiently the next time. The neurons which fired more accurate detail are strengthened, while the neurons which reported incorrect data are weakened. The third and most efficient form of automated problem-solving is the Evolutionary Algorithm . In Junior High School, we learned in algebra that we could solve the equation “6 + x = 10″ could be solved by subtracting six from either side and arriving at the answer “x = 4.” Eventually, we learned more complex reasoning. The Evolutionary Algorithm is used when there are problems with hundreds of variables,

possibilities with different outcomes, that would take long expanses of time and many human hands to arrive at the best possible outcome. It is a simulated software organism. Kurzweil draws an example from a computer using an Evolutionary algorithm to invest in stocks. A group of simulated software organisms are fed data and statistics reflecting the behavior and fluctuations of the market and the stock. The organisms then hypothetically invest. Each organism is programmed to have a different mode of operation when playing the market. The organisms that underperform are killed off, and the ones that succeed evolve to the next level, sharing their better traits amongst them. Eventually, we get generation after generation of increasingly better investors. The applications of

Evolutionary algorithms are more widespread than we might think, ranging from thumb print identification, to the designs of jet engines, to even the nanokernels which our computer operating systems are based on. Although all three of these technologies are incredible achievements for humans, they are just that: achievements for humans. Despite the massive amounts of information that computers learn and calculate, the learning process must be programmed by an external source. Kurzweil defines intelligence as “the ability to to use optimally limited resources, including time, to achieve a variety of goals.” If this holds true, then yes, we might able able to classify modern machines as intelligent. They do manipulate the very scarce resources they are given in order to arrive