What is the Turing Test? Definition and function explained

The Turing Test was developed by the mathematician, Alan Turing, in 1950 and is supposed to be able to prove the intelligence of machines through an experimental procedure. The supposed proof is provided by a question-and-answer game that is meant to prove human and artificial intelligence are indistinguishable because human interrogators are unable to distinguish between a human and artificial intelligence. Whether this is actually objective proof of machines with human-like intelligence remains disputed.

Is that a human or a bot? Anyone who spends a lot of time on social media or browsing the comment sections of online articles frequently asks themselves this very question. Social bots imitate human users as opinion bots, steer discussions and make automated comments. Often unintelligible from humans, they are based on algorithms that use artificial intelligence and machine learning to imitate human-like communication. This is precisely where the Turing Test, which is designed to determine whether we are dealing with humans or machines, comes into play.

The Turing Test was invented and developed by the eponymous British mathematician, computer scientist and logician Alan Turing in 1950. He intensively studied the problem of artificial intelligence and its criteria during his work on one of the first legendary tube computers called Manchester Mark I at the University of Manchester.

In his article “Computing machinery and intelligence”, published in the journal “Mind”, Turing outlined the basic features of an experimental set-up now famous as the Turing Test, but known at the time as the “Imitation Game”. Since artificial neural networks according to the principle neural network did not yet play a major role in the debate about artificial intelligence and objective scientific proof of thought processes was a long way off, observable analyses of communication with machines were used for this purpose. The goal was and is to be able to speak of artificial intelligence or machine intelligence in the case of machine communication behavior that is indistinguishable from humans.

The structure and procedure of the Turing test couldn’t be any simpler. The test uses a simple question-answer procedure between a human questioner and two anonymous answerers who are not visible to the questioner. The free, unspecified questions are asked by the human without any visual or auditory contact with the interlocutors via an input tool such as a keyboard or a screen. At the end of the test, if the human questioner cannot determine from the answers which of the two answerers is the machine, the intelligence of the machine can be defined as human-like.

To date (March 2022), no official examples can be cited of machines passing Turing tests. Nevertheless, the experimental setup is still relevant for the development of artificial intelligences today, e.g. in the context of deep learning, reinforcement learning and supervised learning, respectively. In the future, human-level machine communication based on neural networks will not only play a role on social media and in customer service. Fields such as medicine, diagnostics, agribusiness, security, surveillance, marketing, transportation and production will also be increasingly characterized by artificial intelligent communication.

It is still questionable whether the Turing Test can be used to provide credible or objective proof of artificial intelligence at all. Much of the criticism voiced about the test questions above all was whether the deceptively genuine imitation of human communication actually suggests an independent artificial intelligence or is rather nothing more than a deceptively genuine imitation. The observation of machine behavior, which suggests or apparently presupposes artificial intelligence, is not to be equated with objectively existing artificial intelligence. Intention and thinking ability could thus neither be depicted nor proven by the question-answer game of the Turing Test.

The machine learning test called Winograd Schema Challenge (WSC) is often mentioned as an optimized counter design. This uses a predefined question scheme that requires active knowledge application, general knowledge and rational thinking for correct answers. Based on Terry Winograd’s Winograd Scheme, answering the questions requires an understanding of context, human behavior, cultural background, and reasoning. Other alternatives include the Marcus test, which asks artificial intelligences about their understanding of a television show they “watched”, and the Lovelace Test 2.0, which examines AI’s potential creative abilities.

Despite all the points of criticism mentioned, the central idea of the Turing Test, the deceptively authentic imitation of human communication, still plays a major role in digitization today.

Three usage examples illustrate the unchanged contemporary significance of the Turing Test:

• Human Interaction Proof (HIP): The CAPTCHA query can be described as a negative Turing Test. As a human interaction proof, it is used to distinguish machines from humans as quickly as possible and to efficiently filter bots through automated text and image queries before they visit a website. CAPTCHA has the Turing Test in its name: Completely Automated Public Turing Test to tell Computers and Humans Apart.
• Bots: Bots are digital tools that offer positive or negative functions depending on how they are used. They are used, for example, as chatbots to efficiently automate customer service, but are also used as social bots or spam bots to spread fake news or malware. In both cases, forms of Turing testing are used to promote the development of the bots and make them as indistinguishable as possible from humans.
• Voice assistants: Voice assistants are probably one of the developments that come closest to Alan Turing’s basic idea. Voice-controlled, human-like assistants such as Alexa or Siri are based on the question-answer principle and are intended to automate everyday functions and user needs. Although none of the applications currently come close to passing the Turing test, the intelligent voice functions are constantly being optimized through machine learning and analyses of user behavior, making them more human-like.