The Turing test today is considered the litmus test of artificial intelligence. Devised and exposed in 1950 by the mathematical scientist Alan Turing, The test aims to answer the question of whether or not a machine can truly think like humans. The test in simple words consists of placing a human to chat with a machine, without knowing if he really does it with a machine or a person. After 5 minutes of talk, the judge must decide if the talk was held with a human or with a machine. The Turing test is considered passed, if the computer is able to convince at least 30% of the judges with whom you chat.
On June 7 and commemorating the 60th anniversary of Turing’s death, the annual tests were carried out, in which programmers from all over the world submit their programs to the famous Turing test to achieve the recognition of true artificial intelligence. This year and for the first time since the famous test was created, a computer program was able to pass it by convincing 33% of the jurors that it was a human.
It is about Eugene Goostman, a chatbot or computer program designed to behave like a 13-year-old Ukrainian youth when conversing with him. Its creators, a Russian research group, designed it in 2001 and since then they have perfected it little by little to pass the test of talking for 5 minutes in the most natural way possible. “I feel like I have passed the Turing test in a simple way. Nothing original ”, said Goostman himself after asking him how he felt after having overcome the challenge.
One of its programmers, Vladimir Veselov, says that one of the objectives that they set themselves when creating Goostman was to make him say that he knew everything, but that his young age gave him away as a young man who still has much to learn. To achieve this, they endowed Goostman with a picaresque sense of humor, which convinces the jurors that it is actually a 13-year-old boy, when they talk to him through the chat. The chatbot combines language processing techniques, keyword matching, and database access. The result is an intelligent conversation agent that can interact with one or more human users through auditory or textual methods.
This is not the only time that Goostman faces the Turing test, already in 2001, 2005, 2008 and 2012 he participated in the event and was very close to achieving it. In his attempt in 2012, he convinced 29% of the jurors, being below the 30% established by Turing 60 years ago as the minimum value to be considered human, however it was the best score obtained that year of the other 5 participants.
In this page You can read a conversation that a mathematical scientist had recently with an online version of Eugine, the machine, easily managing to unmask Goostman. It is assumed that in this link You can talk to Goostman but until the date of publication of this note the site is down, probably due to the number of people who want to try their luck.
The rain of criticism
Despite the remarkable achievement of the Goostman development team, and the iconic milestone of passing the famous Turing test, the scientific community is divided in its opinions, among those who celebrate the scope of this achievement, and among those who question the veracity of the results obtained.
Among the strongest arguments is the one that says the fact that that a machine manages to fool a human only proves that the machine is capable of imitating intelligence, and not that it actually possesses it.
Others believe that Eugene’s programmers concocted the Turing test by assigning a specific age to their program and “disguising” their responses with a layer of humor that ultimately confused examiners. Which dismisses its true intelligence and puts it in the category of a show that misleads the person it is talking to.
Finally, Other scientists even question the validity of the Turing test itself after 60 years and warn of the need to update it (with a visual test) or with a different and more rigorous way of certifying the advances in artificial intelligence. After all, they argue, We are not facing the feat of a super machine, but rather the ability of a program designed strictly for conversation, but not for thinking.
The Turing test
Exhibited in 1950 by mathematician Alan Turing in an article (Computing machinery and intelligence) for Mind magazine, and it remains one of the best testing methods for proponents of artificial intelligence. It is based on the positivist hypothesis that, if a machine behaves in all respects as intelligent, then it must be intelligent.
Basically the test consists of a challenge in which the machine must pretend to be a human in a conversation with a man through text communication in chat mode. The subject is not warned if he is talking to a machine or a person so that if the subject is unable to determine whether the other party in the communication is human or machine, then the machine is considered to have reached a certain level of maturity : is smart.
In 1990 a competition was started, the Loebner Prize, an annual competition between computer programs that follows the standard established in the Turing test. A human judge is faced with two computer screens, one of them under the control of a computer and the other under the control of a human. The judge poses questions to the two screens and receives answers. The prize is endowed with 100,000 US dollars for the program that passes the test, and a consolation prize for the best monthly program.
The first time a judge mistook a machine for a human was in 2010, when Bruce Wilcox’s robot Suzette passed the test. One of the applications of the Turing test is the control of spam or junk mail. Given the large volume of emails sent, spam is usually sent automatically by a machine. Thus, the Turing test can be used to distinguish whether the email was sent by a human sender or by a machine (for example, by the Captcha test).