TURING TESTS

Turing tests are procedures to test the functional equivalence of people and computers. They generalize the thought experiment proposed by the British mathematician Alan M. Turing (1912–1954) in his pioneering 1950 paper, "Computer Machines and Intelligence," to answer the question, Can machines think?:

[T]he "imitation game" ... is played with three people, a man (A), a woman (B), and an interrogator (C) who may be of either sex. The interrogator stays in a room apart from the other two. The object of the game for the interrogator is to determine which of the other two is the man and which is the woman.

In order that tones of voice may not help the interrogator the answers should be written, or better still, typewritten. The ideal arrangement is to have a teleprinter communicating between the two rooms.

We now ask the question, "What will happen when a machine takes the part of A in this game?" Will the interrogator decide wrongly as often when the game is played like this as he does when the game is played between a man and a woman? These questions replace our original, "Can machines think?" (Turing 1950, pp. 433–434)

Turing's proposed test has been very influential in the philosophy of mind and cognitive science. Variations apply as well to some issues in the ethics of technology. Consider four cases.

A Moral Problem

First, Turing's original test had a moralizing aspect. A tricky game is needed to arrive at a fair test of human versus machine ability because humans are prejudiced against machine intelligence. Turing's blind test combats this prejudice. Note that in Turing's imitation game, computers serve two roles: as potential artificially intelligent interlocutor and as filtering media. This second role has become more significant with the spread of networked computer-mediated communication. As Turing noted, when people communicate only by typing, many cues drop out, and it is not immediately obvious who is male or female. Indeed, in a long-running Internet implementation based on Turing's original male–female game gender turns out to be very difficult to detect (Berman and Bruckman 2001). The spread of the Internet has made this filtering and uncertainty, which might be termed a "Turing effect" of computer-mediated communication, practically important. Its equalizing and liberating aspect is summed up by Peter Steiner's 1993 New Yorker cartoon caption: "On the Internet, nobody knows you're a dog." So too do age and rank in organizations drop away in chat rooms and e-mail, creating one of the moral risks of Internet anonymity: adults posing as children and vice versa. Indeed, the recent winners of the annual Loebner metals for best Turing test performance have been chatbots (Loebner Prize Internet site).

Machines with Moral Standing?

Second, and more speculatively, were a computer program to pass Turing's original test for intelligence, this success might have moral implications. For Roger Penrose, ownership of a device that passed the test "would involve us in moral responsibilities [because] to operate [such a] computer to satisfy our needs without regard to its sensibilities would be reprehensible." This could be morally equivalent to slavery. "Turning off the computer, or even perhaps selling it, when it might have become attached to us, would present us with moral difficulties" (Penrose 1989, p. 8). Of course, this argument assumes human-level intelligence sufficient for moral standing. A broader account of moral standing leads to an extension of Turing's test.

Third, there is the direct ethical extension of the Turing test. Instead of testing for intelligence, one could test for moral standing itself. Arguably, a computer program that could discuss ethically complex issues indistinguishably from a person should be granted moral standing (Allen, Varner, and Zinser 2000). Variations on this theme of testing for moral personhood via indistinguishability is common in science fiction. For example, in Ridley Scott's 1982 film Blade Runner, humans and computer-based "replicants" are indistinguishable by any nonphysical (invasive) Turing test.

Problems with Turing Tests

These Turing test applications disclose some of its problems: (a) The original version tests for communicative ability, but ethics (and perhaps intelligence) arguably requires the ability to act as well as to communicate. (b) Turing tests make playing a game (the imitation game) the criteria for intelligence or ethics, respectively. But the ability to deceive is neither necessary (think of naive but intelligent agents) nor sufficient (think of programmed con artists) for moral considerability. (c) More generally, experience with computers because Turing makes it obvious that people tend to overestimate the abilities of computer programs. Notwithstanding such problems, the Turing test remains ethically salient, invoking core moral ideals of fairness and the equivalence of the indistinguishable to challenge prejudice about the unique status of human abilities.

Human versus Machine: Chess

Fourth and again quite practically, there are indirect ethical questions about the human values challenged by machine performance of activities once thought to be open only to humans. The most noted example is the game of chess and the victories of IBM's Deep Blue computer system over grandmaster Gary Kasparov in 1996 and 1997. This can be considered, loosely, a real-world Turing test, whereby master level chess ceased to be a realm in which humans could be distinguishable from machines.

Predictably Deep Blue's success led to a strategic retreat, distinguishing easily (we say now!) mechanizable formal games such as chess from "really difficult" tasks embedded in thick human contexts. Subsequently the Internet search engine Google introduced automated news editing, and reviewers claimed that its editing service was indistinguishable from that of normal human editors. It remains open whether people will view these tests as raising the value of what machines can now do or lowering it. The initial reaction to Deep Blue's victory suggests the latter.

PETER DANIELSON

SEE ALSO Artificial Intelligence;Robots and Robotics;Turing, Alan.

BIBLIOGRAPHY

Allen, Colin; Gary Varner; and Jason Zinser. (2000). "Prolegomena to Any Future Artificial Moral Agent." Journal of Experimental and Theoretical Artificial Intelligence 12(3): 251–261. Thorough consideration of ethical aspects of the Turing Test.

Berman, Joshua, and Amy Bruckman. (2001). "The Turing Game: Exploring Identity in an Online Environment." Convergence 7(3): 83–102. Reports on a research platform for experimenting with online Turing Tests.

Millican, P & A. Clark, eds. (1996). Machines and Thought: The Legacy of Alan Turing. Oxford, UK: Clarendon Press. A collection of leading articles on the Turing Test emphasizing cognitive science and philosophy of mind.

Penrose, Roger. (1989). The Emperor's New Mind: Concerning Computers, Minds, and the Laws of Physics. Oxford: Oxford University Press. A well-known scientist considers one ethical aspect of the Turing Test.

Turing, Alan M. (1950). "Computing Machinery and Intelligence." Mind 59: 433–460. Turing's original article. See also under Internet Resources below.

INTERNET RESOURCES

The Loebner Prize. Available at http://www.loebner.net/Prizef/loebner-prize.html.

Turning, Alan M. (1950). "Computing Machinery and Intelligence." Available from http://www.loebner.net/Prizef/TuringArticle.html. Online reproduction of Turning's original article.

The Turning Test Page. Available from http://cogsci.ucsd.edu/~asaygin/tt/ttest.html#onlineref. Contains links to many Turning Test papers.