Turing test: what it is, how it works, advantages and limitations

When we talk about artificial intelligence, what do we mean? Experts in this matter have various points of view on the matter, but one of the most shared is that a computer can be considered intelligent when it is capable of thinking like a human being.

Checking to what extent a machine “thinks” is complicated. Before getting machines to think, we should first understand how we think, a matter that remains a great mystery to this day.

Be that as it may, of all the ways to check if a machine has artificial intelligence, the Turing test is surely the most famous. Next we are going to discover what it consists of and what its implications have been.

• Related article: "6 examples of artificial intelligence applied to society"

What is the Turing test?

Turing's test is an artificial intelligence (AI) inquiry method to determine whether or not a computer is capable of thinking like a human being. This test is named after its author, Alan Turing (1912-1954), one of the most famous experts in theoretical computer science, cryptanalysis, theoretical biology, and mathematics. Turing proposed that to say that a computer has true intelligence, it must be able to imitate the responses that a human would give to specific conditions.

The original Turing test required three separate terminals. One terminal would be operated by a computer, while the other two would be operated by humans. During the test, one of the humans has the task of asking questions, while the other human and the computer must answer them.

The questions will be on a specific topic, using a specific format and context. After having done the interrogation, the human asking the questions must decide which respondent was the human and which was the computer.

The test is repeated as many times as desired. If the interrogator is wrong more than half of the time in deciding who the machine is, the computer is considered to have artificial intelligence., since he has behaved in such a human way that he has cast doubt on the one who asked the questions.

• You may be interested: "Computational theory of mind: what does it consist of?"

test history

As we have mentioned, this test is named after its inventor, Alan Turing. This English mathematician pioneered computer research and machine learning during the 1940s and 1950s. Turing introduced this test in his 1950 paper called "Computing Machinery and Intelligence" when he was working at the University of Manchester. Without Alan Turing and his particular proposal, current AI programs such as chatbots and office assistants would be science fiction.

His article began: "I propose to consider the question, Can machines think?" This question was accompanied by a peculiar hypothetical exercise: The Imitation Game. In its original conception, this did not involve a computer but three human participants in three separate rooms, connected to each other by a screen and a keyboard. In one room there was a man, in another a woman, and in another there was the person who was going to act as judge, of irrelevant gender. The woman had to try to convince the judge that she was the man.

After explaining this exercise with humans, Turing proposed it again, but this time putting a computer as a participant whose mission was to convince the judge that he was not a computer, but a being human. The task of the interrogator is to decide who is the Artificial Intelligence and who is the human being, as we have commented in the previous section.

Since the Turing test was conceptualized, it has been used countless times, being used as a useful tool for determining that a computer or cyber program possesses intelligence artificial. We have some resounding examples with the ELIZA program (1966), by Joseph Weizenbaum, capable of posing as a psychotherapist, and PARRY (1972) developed by the psychiatrist Kenneth Colby who pretended to be a person with paranoid schizophrenia.

• You may be interested: "The 4 main types of science (and their fields of research)"

Limitations of the test when studying artificial intelligence

The Turing test is not infallible. His main criticism has to do with the way it has traditionally been applied, usually in the form of very limited so that the computer, with the technology that was available at the time, could exhibit an intelligence similar to the human.

Many of the occasions in which the test has been applied, the questions that were asked to the computers were closed, of the type that are answered with a “yes” or “no”. Sometimes these questions were so easy for computer programs to answer that they could hardly be taken as a demonstration of their AI.

On the other hand, when the Turing test was applied with open or conversational questions, the response of the computer programs could become very chaotic. Although today's computer programs are more intelligent, it can be said that depending on which open questions they make answer in a way that makes it obvious to his flesh and blood interlocutor to discover that he is talking to a machine.

For many researchers, the question of whether or not a computer can pass the Turing test no longer matters today. Instead of focusing on seeing how he is able to convince someone that they are human instead of computer, the important thing would be to see how to make human-machine interaction more intuitive and efficient. For example, using a conversational interface.

• Related article: "17 curiosities about human perception"

Alternatives to this test

Since Turing lived, the world of computing and computing has evolved greatly. Today there are many alternatives to the original Turing test, which allow us to determine more clearly what an AI is. Between them we have:

1. Reverse Turing test

In this case, it is the human trying to convince the computer that it is not another computer. We have an example of this in the popular CAPTCHA, very recurrent in web pages where you have to enter a password.

2. Total Turing Test (Total Turing Test)

here the interrogator can test perceptual abilities in addition to checking the ability to manipulate objects of the questioned subject.

• You may be interested: "Turing machine: what it is and how it works"

3. Minimum Intelligent Signal Test (MIST)

This would be the most minimalist version of the Turing test, in which only true/false or yes/no closed questions are asked.

4. TheMarcus Test

This test is very interesting. Allows you to evaluate to what extent a computer program is capable of "seeing" a television show and understanding it, checking it by asking him questions about the plot or characters of it.

5. The Lovelace Test 2.0

This test is made to detect artificial intelligence through his ability to create art.

• Related article: "Ada Lovelace: Biography of this Mathematician and Programming Pioneer"

6. Winograd Schema Challenge

Multiple choice questions are asked in a specific format.

How is the Turing test currently used?

Despite there being variations of the Turing test, its original version is still used today.. We have an example of this with the Loebner Prize, a contest held annually since 1990 in which a decoration is given to those computers that manage to convince the jury that grants it that they are human, using the Turing test to define it. The competition follows the standard rules of the Turing test.

In a competition organized by the University of Reading to commemorate 60 years since Turing's death in 2014, a chatbot named Eugen Goostman who pretended to be a 13-year-old boy managed to pass the Turing test, managing to fool 33% of the judges.

In 2018, Google Duplex managed to book an appointment with a hair salon over the phone while a crowd of 7,000 watched. The receptionist did not know that she was not having a conversation with a real human being. This is considered by some to be one of the most recent and interesting cases of passing the Turing test, even though it does not follow the original format proposed by Alan Turing.

We have another case with GPT-3, a natural language processing model that some consider to have the highest probability of passing the test in its original format. However, despite its highly advanced abilities, many have criticized this technology because it can be easily tricked into asking meaningless questions and causing her to have trouble passing the test Turing.

Although there is debate about the relevance of the Turing test to define that a computer program has intelligence, this proof is still used today and widely commented on from a philosophical point of view to discuss and investigate about AI. As artificial intelligence advances and a better understanding of how the human brain works, the Turing test will continue to be used as a tool to establish when a machine is capable of thinking like a human being.