The Turing test was created in the late 40s of the last century. English mathematician Alan Matheson Turing tried to understand whether robots can think. This is what pushed him to the invention.
The history of the creation of the Turing test
English mathematician Alan Matheson Turing is known as a unique specialist in the field of computer science, computing and cryptography. It was he who created the prototype of the modern computer (Turing computer). The scientist had many other achievements. In the late 40s of the last century, a mathematician began to wonder what kind of electronic intelligence can be considered reasonable and whether a robot can approach human behavior so much that the interlocutor will not understand who is actually in front of him.
The idea of creating a dough arose after the Imitation Game became popular in England. This fashionable game for that time involved the participation of 3 players - a man, a woman and a judge, in the role of which a person of any gender could be. The man and the woman went to different rooms and handed notes to the judge. By the style of writing and other features, the referee should have understood which notes belonged to a player of one gender or another. Alan Turing decided that one of the participants could be replaced with an electronic machine. If, in the process of electronic remote communication, the experimenter cannot determine which of the interlocutors is a real person and who is a robot, the test can be considered passed. And this should be the reason for the recognition of the intelligence of artificial intelligence.
Taking the test
In 1950, Alan Turing formulated a system of questions that could convince people that machines can think.
Over time, the test was modernized, and not machines, but computer bots began to act more often as objects of testing. During the entire existence of the test, only a few programs managed to pass it. But some experts questioned this success. Correct answers can be explained by coincidence, and even in the best cases, the programs were able to answer no more than 60% of the questions. A complete coincidence was never achieved.
One of the programs that successfully passed the Turing test was Eliza. Its creators endowed artificial intelligence with the ability to extract keywords from a person's speech and compose counter questions. In half of the cases, people were unable to recognize that they were communicating with a machine, and not with a live interlocutor. Some experts questioned the test result due to the fact that the organizers set up the subjects in advance for live communication and the participants in the experiment did not even realize that the robot could give answers and ask questions.
Successful can be called the passing of the test by the program compiled by the Odessa citizen Yevgeny Gustman and the Russian engineer Vladimir Veselov. She mimicked the personality of a boy at the age of 13. On June 7, 2014 it was tested. It was attended by 5 bots and 30 real people. Only 33 out of 100 juries were able to determine which answers were given by robots, and which were real people. Such success can be explained not only by a well-designed program, but also by the fact that the intelligence of a thirteen-year-old adolescent is somewhat lower than that of an adult. Perhaps some of the jury was misled by this circumstance.
Opponents of the recognition of the result are also supported by the fact that Zhenya Gustman, who created the program, wrote it in English. During testing, many judges chalked up the machine's strange responses or avoiding answers not only to the age of the alleged interlocutor, but also to the language barrier. They considered that the robot, which they took for a human, did not know the language well.
Since the creation of the Turing test, the following programs have also come close to passing it successfully:
- "Deep Blue";
- "Watson";
- "Parry".
Loebner Prize
When creating programs and modern robots, experts do not consider passing the Turing test a paramount task. This is just a formality. The success of a new development does not depend on test results. The most important thing is for the program to be useful, to perform certain tasks. But in 1991 the Lebner Prize was established. Within its framework, artificial intelligences compete with each other to successfully pass the test. There are 3 categories of medals:
- gold (communication with video and audio elements);
- silver (for text correspondence);
- bronze (awarded to the car that achieved the best result this year).
Gold and silver medals have not yet been awarded to anyone. Bronze awards are presented regularly. Recently, there are more and more applications for participation in the competition, as new messengers and chat bots are being created. The competition has many critics. A quick glance at participant protocols over the past decades shows that a machine can be easily detected with less sophisticated questions. The most successful players also cite the difficulty of the Lebner competition due to the lack of a computer program that could conduct a decent conversation for five minutes. It is generally accepted that the competition applications are developed solely for the purpose of receiving a small prize awarded to the best participant of the year, and they are not designed for more.
Currently, the Turing test has received several modern modifications:
- reverse Turing test (you must enter a security code to confirm that the user is a human, not a robot);
- the minimum intellectual test (assumes only the options "yes" and "no" as answers);
- Turing meta-test.
Disadvantages of the test
One of the main disadvantages of the test is that the program is tasked with deceiving a person, confusing him in order to make him believe in communication with a real interlocutor. It turns out that one who knows how to manipulate can be recognized as thinking, and this can be called into question. In life, everything happens a little differently. In theory, a good robot should imitate human actions as accurately as possible, and not confuse the interlocutor. Programs designed specifically for passing the test evade answers in the right places, cite ignorance. Machines are programmed to make the correspondence look as natural as possible.
Many scientists believe that in fact the Turing test assesses the similarity of speech behavior between humans and robots, but not the ability of artificial intelligence to think, as stated by the creator. Skeptics claim that the orientation towards such testing slows down progress and prevents science from moving forward. In the last century, passing the test was a great achievement and even something fantastic, but nowadays the ability of a computer to "correspond like a person" is not supernatural.
