Over the years, Artificial Intelligence is more and more present in society, and this, at different levels. Continuously improved, it is getting closer and closer to human intelligence. Several techniques are used to assess its capabilities. Among them, the Turing test is undoubtedly the best known and oldest evaluation. If the results obtained by this test make it possible to highlight a real improvement in Artificial Intelligence , the confusion with human intelligence remains to be put into perspective.
What is the Turing Test?
Principle
Developed and presented by mathematician and cryptologist Alan Turing in 1950 in his article “Computing Machinery and Intelligence”, the purpose of the Turing test is to measure the ability of Artificial Intelligence to be confused with human intelligence . To answer the question “can a machine think?”, This test takes place as follows: a human will dialogue with two interlocutors while knowing that one of the two is a machine. After 5 minutes of conversation, the man must succeed in determining which of his two interlocutors is the AI. In this game of imitations, the objective of the machine is not to answer the questions correctly, but to deceive the human interlocutor by answering in the most human way possible.
The first AI
It was in 1966 that an Artificial Intelligence program was able to pass this test: ELIZA . This program was able to analyze a text and search for keywords in order to respond coherently to its interlocutor. Having succeeded in convincing several of its interlocutors that he was a real person, the ELIZA program is considered to be the first to have passed the Turing test.
In 1972, another Artificial Intelligence was put to the test of this test: Parry . This time, this program imitated the behavior of a paranoid schizophrenic person. The assessment of his human character was then carried out by a group of psychiatrists. The results showed that, in 52% of the cases, the psychiatrists believed to be in dialogue with a real human being.
What are the issues ?
The test was developed to answer an ancient philosophical question about the ability of a machine to imitate human behavior. With Artificial Intelligence, the test appears to be particularly relevant and allows us to consider the possibility of a similarity between machine and human . Indeed, by their operation, programs can partially imitate human behavior. Today, AIs are more and more efficient and can be useful at different levels, in particular through machine learning . RNs constantly process statistical data and improve their reasoning and knowledge.
This constant improvement of Artificial Intelligence is a major asset for companies today. Indeed, these tools make it possible to process large amounts of data and transform them into real indicators to develop their strategy . The use of this technology has now become widespread and is used in many sectors of activity in order to optimize production and be closer to customer needs.Through the Turing test, the improvement in AI is noticeable and makes it possible to envisage an even more efficient use in the future.
The Turing Test: what weaknesses?
If this test appears to be particularly interesting to measure the performances and the improvement of Artificial Intelligence, it nevertheless presents certain weaknesses:
• The test does not assess intelligence but the ability of AI to mimic human behavior;
• Intelligent behaviors are not necessarily in line with human logic;
• Considering the reactions of a machine as equivalent to those of human beings is anthropomorphism;
• For some, the Turing test is only a distraction and brings nothing interesting for the development of Artificial Intelligence.
Despite its relevance, the Turing test therefore has many reservations about the importance of the results obtained. Indeed, if it makes it possible to assess the effectiveness of AI programs, it is difficult to speak of an intelligence similar to that of humans. Human intelligence responds to a completely different logic and is notably influenced by emotions. Thus, the results of this type of test should be taken with caution.
More and more human AI
Despite the criticisms and objections that can be raised against the Turing test, it is still a reference and is still used for several university studies. One of the last is that of the University of Reading in 2014 which allowed an RN to pass the Turing test. Indeed, the AI produced by Russian researchers succeeded in convincing 33% of the judges, thus passing the minimum bar of 30% set by Turing to confirm the success of the test. These results can show how much Artificial Intelligence can deceive humans by imitating their behavior.
Increasingly used, and in many sectors, Artificial Intelligence therefore represents an essential challenge in the future and is already considerably modifying the commercial strategies of companies . Constantly improving, this type of system and program tends to improve and to understand human behavior a little more. Faced with these forecasts, we can see interesting opportunities to facilitate his daily life but also risks of cybercrime. The highlighting, by the Turing test, of the quality of certain AIs to pose as humans makes it possible to anticipate these possible risks and to put in place the appropriate measures to avoid these drifts.
