Computing Machinery and Intelligence. The Imitation Game. Criticisms of the Game. The Imitation Game. Machines Concerned in the Game

Transcription

1 Computing Machinery and Intelligence By: A.M. Turing Andre Shields, Dean Farnsworth The Imitation Game Problem Can Machines Think? How the Game works Played with a man, a woman and and interrogator The interrogator is held in a separate room from the other two and asks them questions trying to determine who's the man and who's the woman The Imitation Game How the Game Works Questions and answers must be typed so that voice doesn't give anything away It is proposed to replace the man with a machine and see how it affects the interrogators accuracy in guessing Criticisms of the Game Weighted to Heavily against the machine If a man tried to be a machine, he would have a poor showing, wouldn't the same happen to a machine trying to be a man If the machine is constructed properly, no Machines Concerned in the Game Only allow digital computers to take part in the game Digital Computer Can only do operations that a typical Consists of three parts: Store Executive unit Control Universality of Digital Computers Digital Computers are discrete state machines Given initial state and inputs, it can be predicted d what the futures states t will be Digital Computers are universal because they can mimic any discrete state machine 1

2 The Problem Can Machine's Think? Or CanaDiscreteStateMachinedoWellin a in the Imitation game? Author Believes in 50 years (2000) machines will be able to do quite well in the game Theological Objection God did not give inanimate objects souls, therefore machines cannot think Heads in the Sand Objection Consequences of man thinking would be too dreadful to think of. Mathematical Objection Gödel's theorem says any finite state machine has limits There must necessarily be questions it cannot answer For example if asked if a machine like itself would ever answer yes - would not be answerable Mathematical Objection In response, states that no human can answer every question either. No single question would reliably trip up all computers. Believes that people with this objection are likely to at least consider the test valid with some debate, unlike those with the previous objections. Argument from Consciousness The chance fall of symbols cannot result in pleasure, grief, anger, depression. Claiming to feel emotions is not the same as really feeling them No machine is ever capable of emotional states. 2

3 Argument from Consciousness response No way to disprove this argument Cannot become a machine and feel self thinking, and no one would believe you even if you could. A logical debater would step back and accept that imitating emotions accurately is close enough. Arguments from Disabilities Computers cannot ever do certain things Know beauty Have sense of humor Fall in love Do something really New Make mistakes Arguments from Disabilities These arguments not given any evidence Based mostly on previous experience with computers and scientific induction Same argument as previous (Consciousness) but disguised Given the correct resources, computers can simulate all these actions Lady Lovelace's objection Computers cannot originate anything, only do what we are capable of programming them to do Early attempts to set up learning machines did not seem to result in original behavior Lady Lovelace's objection Attempts at learning machines only recently started No one claimed learning machines could not do new things, only that they had not Computers often surprise people, and will do so more often as they develop Question revisited later in article Arg. from Continuity in Nervous System Nervous system not a discrete state machine Impossible to perfectly mimic one with the other. 3

4 Arg. from Continuity in Nervous System Even if true, not helpful to person taking the test. Limits of accuracy too small compared to wide variety of human answers Argument from Informality of Behavior Humans do not have definite rules of conduct and therefore cannot be machines Human behavior unpredictable in a way a machine could never replicate Argument from Informality of Behavior Even with modest storage computers can come up with unpredictable responses Extremely difficult to use this in the actual test Argument from Extra-sensory Perception Computers unable to use telepathy, clairvoyance, precognition, etc... Unfortunately, neither can humans that we can tell. Human development relies on somewhat reliable factors Initial state of mind at birth Type of education subjected to Type of environment subjected to Rather than imitate adult mind, possible to simulate child mind and then put it through same education/environment? This approach should result in very simple requirements for machine, and no more difficult education than the child needed Experiments with recreating a child mind can be done using simulated evolution or natural selection. 4

5 True that machines do not have same variety of senses available as child But also true that Helen Keller was capable of learning without full use of normal human senses. Goal for future computers is that they behave like a child when presented with positive/negative stimuli Using a random element in the machine will both enhance speed of learning and the humanity of its behavior. 5