The Turing Test Tautology

• Published in: Journal of information ethics Vol. 34; no. 1; pp. 11 - 29
• Main Author: Hartemann, Frederic V
• Format: Journal Article
• Language: English
• Published: Jefferson McFarland & Company, Inc 01.04.2025
• Subjects: Algorithms; Artificial intelligence; Ethics; Human-computer interaction; Imitation; Language modeling; Medicine; Queries; Questions; Racism; Semantics; Social media; Syntax; Test validity and reliability; Validity
• ISSN: 1061-9321; 1941-2894
• DOI: 10.2307/JIE.34.1.11

Artificial intelligence, or AI, is a subject of considerable interest that is discussed daily in the news media. Large-scale language models, in particular, have been hailed as now routinely passing the Turing test, sparking serious debate about AI emergence and the so-called singularity, especially in view of the existential threats these may pose to humans. Because the Turing test plays a central philosophical role in assessing the potential intelligence of computer algorithms, we first ask three questions: how does the test work; what is its validity; and where could intelligence derive from in syntactical machines? The original concept for the test was Turing's Imitation Game, proposed in 1950; it is essentially identical to its modern version: an AI algorithm imitating a person, and a human compete to convince a third-party individual, who must determine their actual nature. The current version paraphrases Turing's argument: a program that can convincingly and consistently hide its true kind from humans through extensive queries passes the test. This assessment is so central to AI that many generations of machines have been assembled and programmed with this unique purpose: in 1966, Joseph Weizenbaum created ELIZA; in the 1970s, PARRY was used by psychiatrists; in 1991, Hugh Loebner launched his namesake prize for the annual winner of a Turing test competition. Anecdotally, Eugene Goostman, a chatbot, is often believed to be the first to truly pass the Turing Test, in 2014. The validity question will be examined in terms of the reverse test, where the questions are asked by an AI, and shown to lead to interesting logical contradictions; I call this important experiment AI introspection. The third question is analyzed along two different lines of reasoning: can a syntactical machine gain semantic knowledge, and how does intelligence manifest itself within computer algorithms? Following Searle, the first query will be answered by the negative; the second question will be tied to the contradistinction between intelligence and the generation of new information in computer algorithms. These discussions are refined in conceptual terms and focus on intelligence, convergence, emergence, syntax, semantics, and ethics. The latter is a key focus of this paper; there is an extraordinarily wide variety of ethical concerns when it comes to AI, covering virtually any human activity where the technology may be deployed: economy, industry, defense, medicine, social media, virtual reality, arts, and science to name a few. An equally diverse series of institutions have begun to ponder the ethics of AI. For example, in 2016, the World Economic Forum delineated the following list: "Unemployment. What happens after the end of jobs? Inequality. How do we distribute the wealth created by machines? Humanity. How do machines affect our behavior and interaction? Artificial stupidity. How can we guard against mistakes? Racist robots. How do we eliminate AI bias? Security. How do we keep AI safe from adversaries? Evil genies. How do we protect against unintended consequences? Singularity. How do we stay in control of a complex intelligent system? Robot rights. How do we define the humane treatment of AI?" In 2021, a UNESCO pamphlet outlined the following problems: biased AI; AI in the Court of Law; AI creates art; and a life and death dilemma for an autonomous car. These and so-called doomsday scenarios are briefly discussed.