What is this? This info-mural is one of seven Can The Turing Test Determine 2 Whether Computers Can Think? Is failing the test decisive? 38 Ned Block, 1981 Judges may discriminate too well. Overly discerning or chauvinistic judges might Is the test, behaviorally or operationally construed, a legitimate intelligence test? 71 Jack Copeland, 1993 SUPERPARRY. An advanced alien How old are you? 73 Jack Copeland, 1993 1998 72 Anticipated by The machine is prepared fail intelligent machines civilization could write a program that, How old are you? Jack Copeland, 1993 for repeated sentences. solely because of their like Colby's PARRY program (see How old are you? “argumentation maps” in a is The repeated sentence The aliens would design the machinelike behavior. "PARRY," Box 35), would contain ready- The History and Status of the Debate — Map 2 of 7 disputed maneuver. An program to anticipate is supported by made responses to questions. This by interrogator could trick is repeated sentences and to SUPERPARRY would be better prepared SUPERPARRY by typing disputed reply with "increasingly for the test, however, because it would the same sentence over and by irritable responses," contain responses to all possible questions. series that explores Turing’s An Issue Map™ Publication 39 Jack Copeland, 1993 Intelligent machines could fail the test. An intelligent machine could fail the Turing How old are you? Whatever ... SUPERPARRY could therefore pass the Turing test by brute force. But it is clearly not thinking; it is just dumbly searching a long list of conversations (see "The All- is disputed over again. 74 Anticipated by ultimately terminating altogether. question: “Can computers think test by acting nonhuman, by Possible-Conversations Machine," Box Jack Copeland, 1993 36 Ned Block, 1981 by is 75 Jack Copeland, 1993 is supported by being psychologically 90). The substitution trick. An Failing the test is not decisive. It is possible to fail the disputed The machine is prepared for unsophisticated (though interrogator could trick Turing test for intelligence and still be an intelligent being. by substitution tricks. Because Start Here intelligent), or simply by being bored with the proceedings. SUPERPARRY by typing, "Let's write *!= in place of there are only a finite number of "substitution tricks," SUPERPARRY and/or will they ever be able Pass Fail 'horse'; now tell me whether a If a simulated *!= has a tail." could be built handle all of them. is PHOTO # 4 is supported by disputed Refer to photocopy for precise cropping Pass Fail by instructions. 40 Jack Copeland, 1993 to?” Argumentation mapping is is supported by Ned Block Some intelligent beings 70 Jack Copeland, 1993 76 Jack Copeland, 1993 1 Alan Turing, 1950 intelligence passes, would fail the test. I believe that at the end Chimpanzees, dolphins, and The black box objection. The Turing test The beefed-up Turing test. The addition of a design criterion to treats the mind as a "black box." Internal of the century ... one prelinguistic infants can the original Turing test ensures that the Yes, machines can processes are left hidden and mysterious; all a method that provides: think but would fail the will be able to speak of 41 Jack Copeland, 1993 Turing test. If a thinking that matters is input and output. But internal system being tested has the right kind of internal design. processing should not be ignored, because it is machines thinking Nondecisive tests can be useful. It is true that (or will be able to) is it intelligent? is animal could fail the test, is is supported by failing the Turing test is not decisive—it is not a crucial to determining whether or not a system then presumably a thinking disputed without expecting to be is supported by disputed by definitive litmus test for intelligence. But that doesn't machine could too. by really thinks. think. A computational - a method for portraying major contradicted. make it a bad test. Many perfectly good tests are nondefinitive but useful. For example, fingerprint tests are nondefinitive (they can be foiled by gloves) but Output Input system can possess all 22 Jack Copeland, 1993 remain useful nevertheless. 43 Anticipated by Input Jack Copeland Output philosophical, political, and Some simulations are duplications. Two kinds of 44 Robert French, 1990 important elements of 21 Anticipated by Jack Copeland, 1993 Simulated intelligence is not simulation can be distinguished. (1) Some lack essential features of the object simulated; for example, theatrical Pass Robert French, 1990 We can make the test fair by excluding subcognitive A purely cognitive Turing test is not an intelligence test. We 77 Alan Turing, 1950; human thinking or is real intelligence. Just as simulated diamonds are not real death lacks essential physiological features of real death. Such simulations are not duplications. (2) A second type questions. A machine's inability to answer subcognitive cannot distill a level of purely cognitive questions to make an James Moor, 1987 78 Jack Copeland, 1993 pragmatic debates disputed captures the essential features of the object simulated despite Fail unbiased test. To exclude all is Knowledge of internal Turing in a car crash. So, did I ever tell diamonds, a machine that simulates 42 Robert French, 1990 questions (e.g., questions about disputed understanding. by intelligence to pass the Turing test is not really intelligent. In general, is disputed being produced in a nonstandard way, for example artificial coal. Simulations of this kind are duplications. The Turing test is species biased. Failing the Turing test proves nothing about general is disputed by ranking associative pairings) may show that machines can is disputed by subcognitive questions would be to exclude all questions involving analogy and categorization, which by processes is unnecessary. Inferences about thinking are not based on knowledge of internal is Suppose you and Alan Turing are driving to a party and your car crashes. He is gashed up, but there is no blood. Instead, wires and silicon you about this test I designed? a simulated X is not a real X. by intelligence because only humans can pass it. never match human responses, operations. We generally infer that disputed chips stick out of his skin, and you wonder if he I'm Intelligent nonhuman entities (such as intelligent would render the test useless as a by - a summary of an ongoing, but we can make the test someone thinks just on the basis of is just an elaborate puppet. In such a case you Alan Turing dying!! machines, aliens, etc.) could never pass the test, test of what we call intelligence. unbiased by excluding such outward behavior. "Not even brain might modify your judgment that Turing is a because they would not be able to answer questions. scientists examine the brains of thinking being based on this new knowledge of subcognitive questions. Such questions can only their friends before attributing his internal constitution. So, internal processes is supported by be correctly answered on the basis of a specifically thinking to them" (Moor, 1987, p. are relevant to judgments of thinking. major philosophical debate of human associative network of representations. Robert French 1129). Simulation Duplication is supported by of death of coal is supported by the 20th century is supported by 79 John Searle, 1980b 23 John Searle, 1980b 46 Robert French, 1990 Overt behavior does not demonstrate understanding. Paradise Island Simulations are not Rating games reveal the subcognitive is The Chinese Room argument shows that a system can engage in 24 Lawrence Carleton, 1984 underpinnings of human intelligence. duplications. A disputed Chinese-speaking behavior yet fail to understand Chinese. The Simulations are duplications if the inputs and Rating games provide an experimental basis for simulation of a fire will by Chinese Room is a system that could pass the Turing test without outputs are of the right kind. If a simulation uses - a new way of doing intellec- not burn. A simulation formulating subcognitive questions that only humans thinking in Chinese. the same kinds of inputs and outputs as the simulated 45 Robert French, 1990 could answer. A rating game asks a subject to judge of metabolism in Note: Also, see Map 4. phenomenon uses, then it is also a duplication. The The seagull test is species biased. digestion will not the strength of an "associative pairing" between 2 be • hav • ior • ism: A school of psychology that takes the overt behavior of a system is Chinese Room, for example, uses the same kinds of The seagull test—an imaginary test to determine whether a machine terms. For example, in the "neologism rating game" nourish. A simulation of to be the only legitmate basis of research. We can distinguish philosophical (or logical) In one room ... a machine 1 answers questions posed by the The Turing Test In another room ... a human 2 control answers question posed an automobile engine disputed inputs and outputs (i.e., symbol strings) to model speaking can fly—shows how a Turing-type test is species biased. Judges a subject is presented with a new term, like "Flugly," behaviorism from methodological behaviorism. Philosophical behaviorism is the by compare 2 radar screens that track a seagull and a putative flying 69 tual history. as humans use when they speak, hence it is a duplication and asked to rate it as a name for various things: position that mental states are reducible to sets of input-output correlations (or 2 Alan Turing, 1950 interrogator. by the interrogator. will not get anywhere. The Turing test provides a as well as a simulation. Computer simulations of fire, machine, respectively. A machine passes if the judges can't (a) a child's favorite teddy bear, The Turing test is an adequate Similarly, a simulation of behavioral/operational definition of "behavioral dispositions"). Methodological behaviorism is a research program that digestion, etc., are different. Because they don't use the distinguish it from the seagull on the radar screens. Failing the (b) the name of a glamorous movie star, or ---starring-- restricts itself to the investigation of behavior and its observable causes. test of thinking. The Turing test is understanding is not intelligence. The Turing test defines thinking 80 James Moor, 1976, 1987 seagull test tells us nothing about flying in general, because only What does it actually understanding. right kinds of input and output, they are mere simulations. (c) the name of character in a W. C. Fields movie. Narka Flugly a test to determine whether a Supported by seagulls (or machines behaving exactly like them) could pass it. in terms of overt, measurable behavior. It The behavioral/operational interpretation is vulnerable to Only humans could answer such questions plausibly, offers a behavioral/operational definition of machine can think. If a computer "The Chinese Room Is More Than A Simulation," Non-seagull flyers (such as airplanes, helicopters, bats, etc.) could counterexamples. Whether behaviorally or operationally on the basis of subcognitive associations. our ordinary concept of thinking. can persuade judges that it is human Box 55, Map 4. never pass the test, because they would produce a distinctive radar interpreted, the Turing test is vulnerable to cases where unthinking Disputed by via teletyped conversation, then it signature. is machines pass the test or thinking machines fail it (see the "Is passing op • er • a • tion • al • ism: A philosophical passes the test and is deemed able to "A Box of Rocks Could Pass the Toe-Stepping the test decisive?" and "Is failing the test decisive?" arguments on disputed position asserting that scientific concepts should think. Game," Box 7. this map). But, if the Turing test is interpreted as a source of inductive by be defined in terms of repeatable operations. For Notes: Interrogator is interpreted as Note: A similar debate takes place in the "Is evidence, then such counterexamples can be accommodated. example, intelligence can be defined in terms of • Shown here is a standard passing the test decisive?" arguments on this Supported by the ability to pass the Turing test. interpretation of Turing's test; map, which deal with the question of what "The Inductive Evidence Interpretation," Box 108. many others are possible. Turing's can be demonstrated by successful simulation. original version of the Turing test Unmapped Territory contain? takes the form of an imitation game (see "The Imitation Game," Machin trol Additional Judges an con Other Turing test arguments Box 4). e cont Referee Hum behaviorism/ estant • Authors differ in whether they take operationalism the Turing test to establish "thinking" 48 Charles Karelis, 1986 arguments 81 Anticipated by Ned Block, 1981 In a third room ... the interrogator engages in teletyped (computer) conversation with the 82 Ned Block, 1981 or "intelligence." 3 contestants as judges look on. If a machine can trick the interrogator and the judges into thinking it Nonstandard human The operationalist interpretation. An Players in the Turing Test The operational interpretation is too 92 Anticipated by controls can mislead a intelligent system is operationally defined 93 Ned Block, 1981 is a human, then that machine has passed the Turing test. rigid. If thinking is operationally defined in Ned Block, 1981 is supported by judge. An unintelligent machine as one that passes the Turing test. By Some machines can terms of the ability to pass the Turing test, All intelligent machines could pass the Turing test if it definition, systems that pass are intelligent; exhibit their own then systems that pass are necessarily intelligent exhibit the intelligence of 25 Anticipated by Alan Turing, 1950 were compared with a stupid, is interpreted as systems that fail are unintelligent. intelligence. The all Judge: a person who is and systems that fail are necessarily unintelligent. their designers. There is Machine Human contestant, or ESP would confound the Turing flippant, or in general is possible-conversations interprets the conversations 26 Alan Turing, 1950 disputed But, this is too rigid. Intelligent machines could is nothing unusual about the contestant: the control: a human whose test. Extrasensory perception could anomalous control. disputed disputed machine, which dumbly between interrogator and is Perform the test in by fail the test (see the "Is failing the test decisive?" all-possible-conversations by machine that is conversation is compared with invalidate the Turing test in a variety telepathy-proof rooms. by searches a list of contestant. Often it is disputed in • tel • li • gence: arguments on this map) and unintelligent machine. It may be said of any attempting to pass that of the machine. of ways. If a competitor had ESP she To prevent ESP-related conversations, only exhibits assumed that an by 49 Charles Karelis, 1986 The ability to machines could pass the test (see the "Is passing intelligent machine that its the Turing test. The control is sometimes also could "listen in" on the judges and problems, the test would the intelligence of its interrogator is at the same is A zombie or an unconscious Pass pass the Turing the test decisive?" arguments on this map). intelligence is really just that of its programmers. A machine 3 Alan Turing, 1950 This is the Gallup Poll. called a "confederate" or gain an unfair advantage. If a judge disputed have to be carried out in Fail test. Supported by designers. Do you think that time a judge. machine could pass the equipped with general The question "can machines think?" is "player" or "foil." had ESP he could easily discern by "telepathy-proof" rooms. "The Neo–Turing Test Conception of machines can think? test. A behaviorally mechanisms for learning, too vague. The question of whether machines human from machine by clairvoyance. con • trol: A comparison case for what is being sophisticated but unconscious Intelligence," Box 88. problem solving, etc. would can think is too vague. It cannot be decided by tested, to ensure that test results are not just artifacts As Turing put it, "With ESP anything Referee: a is supported by zombie could pass the Turing exhibit its own intelligence asking what these words commonly mean, of the test procedure or some other factor. In the Interrogator: a may happen." person who test, but it would not be thinking. as well as the intelligence of because that reduces the question of machine Turing test the responses of human controls are Altogether the seven maps: watches to make person who asks Similarly, a behaviorally its designers. is supported by intelligence to a statistical survey like the Gallup compared with those of machines. Without such 84 Ned Block, 1981 sure interrogators questions of the sophisticated but unconscious Poll. Hence, the question "can machines think?" controls we would never know whether observed contestants. Human judges are unreliable. Even though conversational ability was a result of actual ask no unfair 28 James Moor, 1976 computer might pass the Turing should be replaced by the more precise question test without thinking. the behavioral disposition argument avoids the conversational ability or of some other factor, such questions. 27 Michael Apter, 1971 The Turing test does not 95 Ned Block, 1981 of whether they can pass the Turing test. problems associated with the operational as the interrogator's bias, gullibility, etc. Turing assumes that the brain is a The Turing test depend on the assumption A dated system can still - summarize over 800 major interpretation, it still has the problem that it 94 Anticipated by Ned Block, 1981 machine. When Turing claims that machines doesn't assume that the brain is a machine. The machine is dated. The machine be intelligent. that! relies on human judges. Some judges may is can be taught in much the same way that If a machine were to pass the 50 Charles Karelis, 1986 chauvinistically reject intelligent machines (see would not be able to answer questions Intelligence does not require is supported by disputed children are, he assumes the brain is a machine. Turing test, it wouldn't tell us "Judges May Discriminate Too Well," Box 38), about current events. knowledge of current by Consciousness is essential to thinking. No matter how is But whether or not the brain is a machine is part events. For example, the all Does the imitation game determine anything about the relation complex an entity's behavior is, it's not thinking unless it's whereas others might be overly liberal in passing disputed moves in the debates threaded of what's at issue, because if brains were between brains and machines. It 83 Anticipated by Ned Block, 1981 cleverly designed but unintelligent machines by possible-conversations is conscious. Proposed counter- examples (e.g., problem solving What do you think of machines then obviously some machines would only show that some The behavioral disposition (see "Human Judges May Be Fooled Too the latest events in the machine could be disputed in a dreamless sleep) are just deviant uses of language. (brains) would be able to think. So, Turing begs by machines are capable of interpretation. A system is Easily," Box 52). Middle East? programmed to simulate Note: For more on consciousness, see Map 6. is interpreted as is the question of whether machines can think—he intelligent if it is behaviorally disputed Supported by Robinson Crusoe, who is thinking, whether those is assumes part of what he's trying to prove. disposed to pass the Turing test. In by "The Neo–Turing Test Conception of disputed intelligent even though he machines are like brains or not. into claims, rebuttals, and whether computers can think? 51 John Barresi, 1987 this interpretation, neither passing Intelligence," Box 88. by can't answer questions about Merely syntactic machines could pass the the test nor failing the test is recent events. Turing test. Purely syntactic machines could be Sentence conclusive, because intelligence 7 Keith Gunderson, 1964 doesn't require passing but only a Turing test Specific intellectual developed that would pass the Turing test. But A box of rocks could pass the toe-stepping 29 B. Meltzer, 1971 proficiency is supported by without a full naturalistic semantics it is doubtful VP disposition to pass. 96 Ned Block, 1981 The Turing test misleads AI research. The tasks counterrebuttals game. There is an analogy between the imitation game NP NP that such machines will ever think in the human Is passing is The machine could be and a toe-stepping game, in which an interrogator puts Turing test motivates AI researchers to accomplish a disputed sense. A naturalistic semantics would provide a updated. Programmers his foot through an opening in the wall and tries to misleading goal—the general but arbitrary form of by global model of situational context that could allow I have long hair Part I In one room ... a man (X) and a woman (Y) each try to convince an interrogator that he In another room ... the interrogator tries to determine whether a man or a woman has stepped on is disputed intelligence involved in carrying on a normal flexible interaction with the world as well as could periodically update the machine's list of 4 Alan Turing, 1950 his foot. There is a version of the toe-stepping game in by conversation. This kind of intelligence is arbitrary in AI or she is a woman. determine whether X or Y sustained self-perpetuation. 85 Gilbert Ryle, 1949 responses. - 97-130 arguments and rebuttals The imitation game. is which a box of rocks hooked up to an electric eye could that it is relative to particular cultures. AI researchers is the woman, based on Supported by Philosophical (or the test Turing's original formulation Interrogator: disputed win the toe-stepping game by producing the same effect should focus their efforts on specific intellectual tasks their responses. "The Cyberiad Test," Box 68. logical) behaviorism. of his test takes the form of Will X please tell by as a human foot stepping on an interrogator's toe. that can be aggregated to form more generally is supported by an imitation game, which intelligent systems, such as the ability to recognize What we call intelligence is 97 Anticipated by Ned Block, 1981 me the length of However, a box of rocks cannot duplicate real human takes place in 2 stages. patterns, solve differential equations, etc. simply patterns of potential The all-possible-conversations his or her hair? toe-stepping. Similarly, the imitation game just shows 98 Ned Block, 1981 per map is and actual behavior. The machine redefines intelligence. X: My hair is that a system can produce the outward effects of disputed Sure, it can The machine is all echoes. think ... external behavior of a system We normally conceive of intelligence conversation, not that it can duplicate real human decisive? is supported by shingled and the 30 Bennie Shanon, 1989 by 52 Ned Block, 1981 is enough to determine 86 Hilary Putnam, 1975a relative to input-output capacity, not Intelligence is not being redefined, longest strands intelligence. is Super Spartans. We can imagine a because it is part of our normal The Turing test assumes a is supported by Human judges may be fooled whether or not it thinks. relative to internal processing. By are about 9 inches representationalist theory of mind. too easily. Human judges can disputed race of Super Spartans who have been stipulating an internal processing conception of intelligence that is by is - 70 issue areas in the 7 maps long. By confining itself to teletyped trained to suppress any behavioral is input-output capacity can be be fooled by unintelligent but disputed disputed condition on intelligence, Block disputed Y: I am the woman, 9 John Stevenson, 1976 conversations, the Turing test assumes a cleverly programmed machines. indication of pain. They exhibit no pain merely redefines the word misleading. Someone who plays by by by Part II don't listen to The imitation game is an representational theory of mind, according to Supported by is behavior whatsoever, even when they "intelligent." grand-master chess by copying him! all-purpose test. Playing the which thinking involves computational "The ELIZA Effect," Map 1, disputed are in excruciating pain. So, another grand master would not be To see whether a machine can think, replace the man (X) in the imitation is operations on symbolic representations. But by philosophical behaviorism fails to give in•tel•li• deemed a grand master. Similarly, imitation game is a "second order" 47 Ned Block, 1981 Box 106. - 32 sidebars history and further game with a machine. If the machine can successfully imitate the person, ability that presupposes many other disputed critics have pointed out various phenomena that Passing the test is not decisive. Even Pass an account of what pain is. gence the all-possible-conversations then we say the machine can think. by Fail machine, which merely echoes its abilities. Hence, it is misleading to cannot be accounted for in this framework. 31 Justin Lieber, 1989 if a computer were to pass the Turing test, characterize the ability to pass the Hence, the Turing test begs the question of this would not justify the conclusion that it programmers, would not normally Turing was not committed to imitation game as just one example or whether machines can think because its very was thinking intelligently. 53 James Moor, 1987 be considered intelligent. representationalism. Turing leaves OUCH! is property of thinking. 11 James Moor, 1987 procedure assumes that a representational system open the question of whether the Note: Because the issue of passing the Turing A serious judge would background mental I feel no disputed The customer could infer like a computer is capable of thinking. (Also representational account of intelligence test is closely tied with the issue of what can not be easily fooled. A This court is 87 Roderick M. Chisholm, 1957 states is is 6 Jack Copeland, 1993 by is see the "Can symbolic representations account critical judge would not not fooled. Philosophical behaviorism is pain! disputed disputed A man simulating a disputed that the vacuum was all is valid. In fact, in some passages he be inferred from external behavior, this region purpose. The buyer would for human thinking?" arguments on Map 3.) seems to reject the representational contains arguments that are similar to those lazily interact with a circular. Any attempt to define mental by by woman is not a woman. by 10 Keith Gunderson, 1964 If a man were to win the not have to actually watch view in favor of connectionism (where in the "Is the test, behaviorally or operationally competitor in the Turing states in terms of behavioral dispositions imitation game, he The vacuum salesman test, but would be focused relies on further mental states. But, this which are the vacuum cleaner do he discusses learning machines). construed, a legitimate intelligence test?" are would only have example. If a vacuum-cleaner on distinguishing the is a circular definition of mentality. For defined in 99 Anticipated by Ned Block, 1981 8 Keith Gunderson, 1964 everything in its repertoire to is supported by Moreover, Turing would have denied arguments on this map. defined in succeeded in simulating salesman demonstrated an human from the computer. example, if the desire to eat chicken is terms of Assuming we are better than the One example cannot explain infer that it was a good is that an "autonomous" account of terms of a woman. He would all-purpose Swish 600 by sucking Such a judge could not defined as a disposition to eat chicken machine is chauvinistic. A system 100 Ned Block, 1981 thinking. The imitation game vacuum. Even if the disputed intelligence could be abstracted from obviously not be a up dust from a carpet, the easily be fooled. when chicken is present, then this that processes information differently than Intelligence requires woman. Therefore, the attempts to explain thinking in salesman only demonstrated by the underlying mechanisms of customer would be unsatisfied. = The argumentation maps: 5 Kenneth Colby, F. Hilf, S. Weber, and H. Kraemer, 1972 the vacuum cleaner picking intelligence, for "he was a hopeful presupposes a belief that chicken is edible, we do may not be intelligent in our sense, "richness" of The imitation game is flawed. The imitation game is imitation game is not an terms of one example: a man's He or she would want to see it behavioral but we may not be intelligent in its sense adequate test. up dust, if the buyer saw that mechanist through and through." 54 John Searle, 1980b which presupposes a belief that chicken information a weak test for several reasons. ability to imitate a woman. But suck up mud, straw, paper, and cat is nonpoisonous, and so forth. dispositions either. To assume that human intelligence it came with various The Chinese Room processing. • The concept of "womanlikeness" is too vague. this is just one of many different hairs from couches, tight corners, attachments, he or she could passes. The Chinese is better than machine intelligence is The all-possible- • Control subjects may be inexperienced at deception. examples of mental activity, such is supported by and so forth. One example is not is reasonably infer that they Room thought experiment chauvinistic. conversations machine • The test fails to specify what to do when a computer 32 Bennie Shanon, 1989 is supported by - arrange debate so that the cur- as wondering, reflecting, enough to demonstrate the disputed worked. The Turing test makes an autonomy involves a system that is isn't unintelligent simply fails to imitate a man imitating a woman. Is the deliberating, writing sonnets, and all-purpose Swish 600, just as one by is Supported by claim. In assuming a representational theory of passes the Turing test by disputed because it processes so forth. Thinking is "all purpose," computer a successful imitation of a man failing to example is not enough to "The Inductive Evidence mind, the Turing test endorses an autonomy au • ton • o • my (of speaking Chinese but disputed by information differently imitate a woman? Or is the computer a poor imitation of whereas playing the imitation Keith Gunderson demonstrate thinking. A: Phonology by Interpretation," Box 108. claim to the effect that intelligence can be theories): Two theories A fails to understand than we do. Its style of game uses just one ability. Great! Lousy man? Or what? and B are autonomous if Chinese. information processing is rent stopping point of each studied solely in terms of representations and B: Syntax computations, without recourse to the brain, the the phenomena of A can Note: Also, see Map 4. How are you? Hi what's at issue. In be accounted for without Everything in A can John Searle particular, it lacks the body, or the world. But there is good reason to recourse to the be explained without Hello richness of processing think the autonomy claim is false, and that in conceptual framework of reference to B and capacity that is associated studying the mind we should also study its total 56 Joseph Rychlak, 1991 B. vice versa. with intelligence. debate thread is easily seen Can the Loebner Prize stimulate the study embodied context. is supported by 55 Joseph Rychlak, 1991 The anthropomorphizing objection. Our natural tendency to anthropomorphize The reverse Turing test. Humans tend to treat machines as fellow humans, making quick and unwarranted ascriptions of intelligence. We say that computers "want to do" things or "will not allow" certain alternatives. This tendency could be tested for using a "reverse Turing Test." Instead of testing to see if a machine is disputed Is the neo–Turing test a - identify original arguments by machines should caution us against calling can trick a human into thinking it is intelligent, we test to see if a by of intelligence? them intelligent. We are easily fooled machine can avoid being treated as intelligent. We ask: "Can a legitimate intelligence test? is supported by into thinking that a machine that engages machine interacting with a human being avoid being in conversation can think (see "The ELIZA anthropomorphized?" (p. 60) 15 Stuart Shieber, 1994a 17 Stuart Shieber, 1994a Effect," Map 1, Box 106). But, passing Note: Also, see "The ELIZA Effect," Map 1, Box 106. 102 Ned Block, 1981 over 380 protagonists world- Loebner's version of the Turing The Kremer Prize. In 1959, British engineer the Turing test isn't enough to establish 101 Anticipated by Ned Block, The machine would be Have any is supported by test lacks a clear goal. By restricting conversation to limited Henry Kremer established a prize for the first intelligent. Admitting that the intelligent thought. Introspection is also 1981 13 topics and using referees to ensure that no unfair questions are asked, human-powered flight over a half-mile necessary, and the Turing test doesn't 58 Daniel Dennett, 1985 What if we are like the all- all-possible-conversations the "open-ended, free-wheeling" nature of real human conversation is supported by figure-eight course. The Kremer Prize was Joseph Rychlak The quick probe assumption. machine would be intelligent if The Loebner version of the Turing reveal it. TU is possible-conversations test. This is lost. As a result, contestants win by exploiting crafty tricks rather successful (it was awarded in 1977) for 2 Passing the Turing test implies RIN machine? Suppose humans is it were like us is compatible with wide over 40 years is is supported by G 90 Ned Block, 1981 disputed version of the Turing test adds several than by simulating intelligence; hence, the goal of understanding reasons. (1) It had a clear goal: to provide indefinitely many other abilities, TE 88 Anticipated by Ned Block, 1981 by process information in the same disputed claiming that it is not in fact disputed ST The all-possible-conversations machine. A machine could be machines measures to make it practical. human intelligence is lost. incentive to advance the field of for example the ability to The neo–Turing test conception of 89 Ned Block, 1981 way that the all-possible- by intelligent. It's like asking an by Fail Pass • Test questions are restricted to human-powered flight. (2) The task was just understand jokes, write poetry, intelligence. The neo–Turing test conception built to engage in sensible conversation by searching a database atheist what he would say if he is Psychologism: Internal conversations machine does. beyond the reach of current technology: all the discuss world politics, etc. It is a containing all possible lines of conversation in a finite Turing test. turned out to be God. He might predetermined topics (see "Restricted disputed of intelligence is a beefed-up version of differences matter. Psychologism is Would that mean humans are essentials of human-powered flight were by "quick probe" of general behaviorism that avoids classical problems The machine would be intelligent according to the neo–Turing test, admit that if he turned out to be Turing Tests," Box 14). the doctrine that whether or not behavior unintelligent? - make the current frontier of 12 • Referees watch to make sure judges 16 Stuart Shieber, 1994a is supported by available in 1959, they just had to be creatively is intelligence. associated with behaviorism as well as problems is intelligent depends on the nature of but because of the way it processes information, it is clearly God, God would exist, but deny The Loebner Prize. A The Loebner Prize is too far beyond combined. disputed unintelligent. It merely echoes the intelligence of its programmers. associated with the Turing test. It does not rely the information processing that produces that God in fact exists. passed the test? $100,000 award, funded by stick to the topics and ask fair the reach of current technology. by on human judges (it merely requires "a sensible it. If 2 systems behave in exactly the Note: Block simply calls this the "unintelligent machine." philanthropist Hugh Loebner, is questions without "trickery or guile." Computer, • Prizes are awarded using a novel Because full-scale simulated conversation is beyond the reach of sequence of verbal responses to verbal stimuli") same way, one might be intelligent while offered for the first software to write me a current technology, contestants in the Loebner competition are 59 James Moor, 1976, 1987 debate easily identifiable is supported by scoring method. Judges rank the is sonnet ... and the test avoids problems associated with the other is stupid, because of differences pass the Turing test in an annual disputed forced to exploit trickery to win the annual prize. The AI community 18 Stuart Shieber, 1994a Narrowness objections play a misleading contest started in 1990. Whether contestants relative to one another, dispositional analysis by substituting "capacities" is in the way each system processes is supported by That machine isn't by should not focus on winning Loebner's "parlor game" version of the The da Vinci Prize. An imaginary prize is set 57 Let's go talk is numbers game. It is misleading to characterize disputed a system has passed the test is rather than just passing or failing disputed for dispositions. information. Behavior by itself isn't intelligent. Its Turing test, but should engage in basic research that will up in 1442 (da Vinci's era) for the highest The test is too narrow. The Turing test is limited in scope. to the the Turing test as just one test, because the test can by 103 Anticipated by Ned Block, 1981 determined in a competition them. by Note: Block sets up this position in order to enough; internal differences matter. intelligence comes • The highest scoring computer in incrementally lead the way toward artificially intelligent behavior. human-powered flight. We imagine that, like computer. be used to conduct many different kinds of tests. The Combinatorial explosion makes the where computers face off with PHOTO # 31 the Loebner Prize, a yearly competition was held Implemented Model It narrowly focuses on one ability—the ability to engage in attack it, so that he can show that even a beefed- from the all-possible-conversations machine 104 Ned Block, 1981 - provide summaries of eleven Refer to photocopy for each contest is the yearly winner, human conversation—and is therefore inadequate as a test of interrogator can test for the ability to tell jokes, speak up version of behaviorism will fall prey to Great! Lousy programmers who humans in a special version of precise cropping and a prize was offered each year. At the time 33 34 Joseph Weizenbaum, 1976 a foreign language, discuss urban renewal, or for any coded in all the impossible. Programming the machine to engage The all-possible- and if a computer's score is equal to general intelligence. psychologism. conversations the Turing test. Even if entrants instructions. the prize was instituted, the requisite technology The Turing ELIZA. Written circa 1965, ELIZA emulates a Rogerian of a wide range of activities. How are you? Hi conversations. in an hour-long Turing test is impossible, because it or greater than the average score of Supported by fail to pass the Turing test and for human flight (in particular, the airfoil) was test has psychotherapist by exploiting a few simple strategies. would give rise to a "combinatorial explosion." The machine is logically Hugh Loebner a human contestant, then it passes "One Example Cannot Explain Thinking," Box 8. Intelligence (or, more accurately, win the $100,000 prize, an beyond the reach of current technology. As a been Although written in just 200 lines of BASIC code, ELIZA conversational intelligence) is Hello programmer would have to code more strings than possible. The all- this version of the Turing test and possible-conversations major philosophical camps of annual award is offered for the is supported by result, competitors won by attaching springs to passed. managed to fool many people, including professional the capacity to produce a is there are particles in the universe. program that makes the best wins the $100,000 prize. psychotherapists, into believing it was intelligent. (See 60 James Moor, 1987 disputed is machine is logically their shoes and jumping. Twenty-five years Existing AI sensible sequence of verbal showing. is supported by "The ELIZA Effect," Map 1, Box 106.) is Linguistic evidence is sufficient for good responses to a sequence of verbal Output by What's disputed possible even if the later no basic developments in human-powered programs have 61 Jerry Fodor, 1968 Input disputed inductive inference. By considering only linguistic stimuli,whatever they may be How are happening? What's by machine is empirically flight had been achieved, but the trickery of passed the is supported by The Turing test only provides by behavior, we can make an inductive inference that a Swell. $ is supported by (p. 18). Hi with impossible. Because spring-assisted flight had been highly optimized. Turing test for Patient: Well, my partial evidence of system can think. Of course, we may revise this Ned Block you? the protagonists (or schools of intelligence. Passing the Turing you? the neo–Turing test intelligence. boyfriend made me Programmers $ $ come here. test does not guarantee successful inference as further evidence comes to light, but that does not mean that all such evidence must be gathered Pretty good. Ciao. makes a claim about $ ELIZA: Your simulation of human behavior. the concept of $ $ before a justified inference can be made. Otherwise, intelligence, it is refuted boyfriend made you Any given Turing test provides Terrible. $ $ come here. only a finite sample of a test taker's "scientists would never gather enough evidence for any 91 Robert Richardson, Either Or Fine. by the logical thought). hypothesis" (p. 1128). 1982 How's it possibility of the all- be • hav • ior • al dis • po • si • tion: A 14 is is supported by Patient: He says Hello. $ I'm depressed much behavioral repertoire. Therefore, Supported by is The neo–Turing test Internal states are allowed as part Internal states are not allowed to going? possible-conversations $ $ Restricted Turing tests. A kind of Turing test that restricts disputed by $ $ $ of the time. passing the Turing test only "The Inductive Interpretation," Box 108. James Moor disputed fails to improve on of the definition of intelligence play a role in the definition of specification of input-output correlations that can Great! Tired. machine. conversation to one topic, such as "women's clothes" or $ $ ELIZA: I am sorry provides partial, inductive by through the notion of a capacity, intelligence, be used to define mental terms. Here is an example Not bad. $ behaviorism. "automotives." Because restricted Turing tests are much easier to to hear that you evidence of intelligence. of a definition of hunger for chicken, stated in OK Lousy. How do I get a Block's neo–Turing test $ $ pass, they provide a feasible short-term goal for AI researchers to $ are depressed. is subject to the in which case in which case terms of behavioral dispositions. work toward. $ 63 Daniel Dennett, 1985 Joseph Weizenbaum Sense organs are an following dilemma. $ $ Implemented Model the test is not purely behavioral as we have no reason to believe that If I see chicken ... $ 35 Kenneth Colby, 1981 unnecessary addition Block claims. Block's conception fares any better then I eat the chicken. $ 62 Peter Carruthers, 1986 $ PARRY. PARRY is a system that The sense organs to the test. Because the than classical behaviorism. (See 20 Stuart Shieber, 1994b $ is supported by is Turing test indirectly tests "Super Spartans," Box 86; 19 Hugh Loebner, 1994 passed an extended version of the objection. The Turing test 105 John Barresi, 1987 The Loebner prize money 37 Robert Abelson, 1968 disputed sense-based capacities "Philosophical Behaviorism Is Mozart's backside. After a coach ride to Vienna Turing test. PARRY simulates a neglects perceptual aspects of by The all-possible-conversations machine cannot should not be spent on The extended Turing test. An (see "The Quick Probe Circular," Box 87.) in 1781, Mozart complained that his buttocks were so paranoid subject by using a natural PARRY: ... I think intelligence. A machine might pass the Cyberiad test. The Cyberiad test avoids the Today's Topic: "useful spin-offs." Whether extended version of the Turing test Assumption," Box 58), the tired that he could hardly ride on. Had a da Vinci language parser and an pass the test by producing DINOSAURS is the Loebner prize money should "interpretation-action" module that he is behind the facilitates its use for scientific purposes. addition of sensory be • hav • ior • al ca • pac • i • ty: A specification problems posed by Block's all-possible-conversations Prize advanced the state of spring technology in earlier gangsters who are Its innovations are: competent verbal behavior but In Either Case machine because the Cyberiad test tests for more than just disputed be spent on generating "useful devices is unnecessary. It of input-output correlations that includes a specification • The judges are not informed that a centuries, Mozart's coach might have boasted a tracks a range of emotional states and out to get me. fail to understand how its words printed copy? by spin-offs" is dubious. The money doesn't make the test any of internal states (beliefs, desires, etc.). Here is an linguistic ability. The Cyberiad test tests for survival in a smoother suspension, and his backside might not have applies a set of paranoia-specific Interrogator: Did relate to the perceptual world. The neo–Turing test fails to offer a plausible behavioral conception is best spent toward advancing you see things that computer is involved in the test at all. more demanding than it Daniel Dennett example of a definition of hunger for chicken, stated in natural environment, and hence linguistic competence is been so sore. The Loebner Prize might turn out to strategies for dealing with An adequate test would have to of intelligence. not enough to pass it. artificial intelligence. others don't seem They are told simply to compare the already is. terms of behavioral capacities. be useful in unexpected ways, much as the da Vinci shame-induced distress. PARRY was provide the machine with some is Supported by to see? performance of 2 entrants with Prize might have. tested against control subjects sort of sensory apparatus. disputed "The Cyberiad Test," Box 68. PARRY: No, I don't. respect to some dimension, for example If I see chicken and I believe chicken is edible suffering from clinical paranoia, and I'm not like the by womanliness, paranoid behavior, etc. 64 Jack Copeland, 1993 and I desire chicken ... • The computer does not participate on the transcripts were judged by other patient here. is supported by Sense organs were not prohibited by then I eat the chicken. professional clinical psychologists. Interrogator: You is its own, but (unbeknownst to the is supported by 66 Steve Harnad, 1995 Turing. The Turing test does not prohibit didn't answer my Legend question. PARRY: Maybe it judges) shares time with a human subject. The 2 are switched off The Turing test underdetermines the goal of creating humanlike robots. By focusing on a particular subset of disputed by machines with sense organs from entering. In fact, Turing thought the best way to prepare a wasn't worth periodically throughout the test. human behavior (symbolic capacity), the Turing test computer for competition would be to give it • The judges score the human/computer Is the test, as a source of Focus Box: The lowest-numbered box in each issue area is an introductory focus box. answering. underconstrains the engineering problem of creating robots sensory abilities and an appropriate education. The arguments on these maps are organized by links that carry a range of meanings: The focus box introduces and summarizes the core dispute of each issue area, sometimes pair after every question, so that is supported by with human capabilities. Because of this, machines that separate scores can be generated for pass the Turing test will not be easily extensible to machines as an assumption and sometimes as a general claim with no particular author. each. The point of this method is to that have other human capacities. Arguments that uphold or defend another claim. Examples include: force judges to focus on the specific supporting evidence, further argumentation, thought experiments, Arguments with No Authors: Arguments that are not attributable to a particular source inductive evidence, a is supported by dimension being evaluated. extensions or qualifications, and implemented models. (e.g., general philosophical positions, broad concepts, common tests in artificial intelligence) are listed with no accompanying author. is 106 Robert Richardson, 1982 is disputed A brute list searcher cannot flexibly respond to changing contexts. The all- 67 John Barresi, 1987 disputed 65 Jack Copeland, 1993 by possible-conversations machine could not adapt its speech to novel and changing contexts. You can order artist/researcher A charge made against another claim. Examples include: Citations: Complete bibliographic citations can be found in the booklet that accompanies You can't fool Mother Nature. Because by the Turing test only measures the ability of Understanding can be tested without A machine that searches a list to produce responses to verbal stimuli might be able to find legitimate intelligence test? is disputed logical negations, counterexamples, attacks on an argument's this map. a machine to trick judges, it cannot test for sensory interaction. Understanding can be some response that would make sense in some contexts, but it could not coherently adapt by emphasis, potential dangers an argument might raise, thought the flexible adaptations needed to interact tested through verbal quizzing, even when sense to the contingencies of an ongoing conversation. experiments, and implemented models. Methodology: A further discussion of argumentation analysis methodology can be found organs are lacking. Indeed, there are certain signed copies of all seven maps in the real world, where Mother Nature is in the booklet that accompanies this map. the judge of intelligence. In the real world kinds of concepts that have to be tested for intelligence is best indicated by an ability 68 John Barresi, 1987 through verbal quizzing because of their to survive through coordinated perception The Cyberiad test. A race of machine nonperceptual nature (e.g., "square root," 109 Douglas Stalker, 1978 is interpreted as Anticipated by Where this phrase appears in a box, it identifies a potential 110 James Moor, 1978 and action. The Turing test neglects such "intelligence," and "philosophy"). 107 Charles Karelis, 1986 How many computers from www.macrovu.com for A distinctive reconfiguration of an earlier claim. attack on a previous argument that is raised by the author people ("cybers") is judged intelligent if 108 James Moor, 1976, 1987 The inductive Competing explanations does it take to screw in a abilities and biases us to think of intelligence they can do whatever humans can do in The inductive evidence interpretation. The Turing test provides interpretation is really can be compatible. An Bantering zombies can't think. Bantering so that it can be disputed. in an intellectualist way. zombies, using processing mechanisms far lightbulb? a natural environment. They must be able inductive evidence that a system thinks. Such evidence may be weak an explanatory explanation of a computer's is to replace humans in their social roles, or strong, but it is never deductively certain. The claim that a system interpretation. The behavior in terms of its disputed superior to those of the all-possible conversations is interpreted as machine, would still be unintelligent. No matter As articulated by Where this phrase appears in a box, it identifies a reform- perpetuate their species for as long as the $500.00 plus shipping and han- thinks, based on its passing the Turing test, remains subject to revision inductive evidence thinking may be compatible by Unmapped Territory ulation of another author's argument. The reformulation human species could, and maintain an how much we enrich the information processing This icon indicates areas of argument that lie on or near the You can't as further evidence comes to light. interpretation assumes that with an explanation in terms of of the all-possible-conversations machine, it still -1 boundaries of the central issue areas mapped on these maps. is different enough from the original author's wording to fool me ... evolving language. a computer's passing the its underlying mechanisms. A Additional That provides further is is can't think unless it's conscious. warrant the use of the tag. This phrase is also used when Pass Fail Turing test should be computer that prints payroll arguments It marks regions of potential interest for future mapmakers inductive evidence that disputed explained in terms of its disputed Supported by dling. by checks can be explained both and explorers. the original argument is impossible to locate other than in they think, but we don't thinking. But, thinking is by in terms of its physical "Consciousness is Essential to Thinking," its articulation by a later author (e.g., word of mouth), or know for sure. Box 50. is supported by not the best explanation of mechanisms and its abstract to denote a general philosophical position that is given a a computer's ability to pass program—these are special articulation by a particular author. More tests the test. A better compatible explanations of the One of 7 in this Issue Mapping™ series—Get the rest! explanation is a mechanical payroll computer's behavior. one, in terms of the structure Similarly, mental and The Issue Mapping™ series is published by MacroVU Press, a division of MacroVU, Inc. The remaining 6 maps in this Issue Mapping™ series can be ordered with MasterCard, VISA, check, of the computer, its mechanical explanations of a © 1998 R. E. Horn. All rights reserved. MacroVU is a registered trademark of MacroVU, Inc. Issue Map and Issue Mapping are trademarks or money order. Order from MacroVU, Inc. by phone (206–780–9612), by fax (206–842–0296), program, and its physical computer that passes the Version 1.0 of Robert E. Horn. environment. Turing test may be compatible. or through the mail (Box 366, 321 High School Rd. NE, Bainbridge Island, WA 98110).