How do we know that our interlocutor on the other side of the handset or chat is human? Can we always distinguish it from a machine? The development of artificial intelligence is already so dynamic that using chat rooms, instant messengers, hotlines of various companies and institutions, more and more often we will have a problem with determining whether we will be answered by a person or an effectively configured algorithm. Although the question about the distinction between a human being and a machine seems to be very “up to date”, it was already a question for scientists several decades ago. The Turing test, which is supposed to give an answer to this question, was established in the 1950s.
What is the test Alan Turing*, a scientist considered to be the “father of artificial intelligence”, was working on? Turing wondered whether the machine has the ability to “think”, simulating human abilities. In order to determine this, he developed his own tool, which we have been using to this day.
Turing test: judges from behind the wall.
How does the test work in practice? On one side there is a judge who is to determine whether the interlocutors, with whom he talks in natural language (not seeing or hearing them), are people or machines. It is presumed that one of the tested interlocutors is a human being, the other a machine. Additionally, it is assumed that the tested objects are obliged to tell the truth (we would not want the machine to be able to lie to us, would we?). Such a conversation takes place “through the wall”. Answers are given in the form of a text written in a computer font in order to eliminate “hints” from the voice. What is the expected result? If 3 in 10 judges are unable to answer the question of whether they talked to a machine or a human being, the test is considered passed.
Turing test: We’ve been waiting for the first optimistic result for… over 60 years!
Although many leading scientists have worked on artificial intelligence, algorithms have not been able to deceive judges for many years. People did manage remarkably well, indicating which interlocutor is a human being and which one an algorithm. Even though, some chatbots (ELIZA, PERRY) are said to be able to perfectly imitate dialogue with people, the test results were not officially confirmed. The first positive result of the test, although still not recognized by everyone, came as late as 2014, so we have been waiting for it for over 60 years! The research was organized by the University of Reading. An intelligent algorithm, named Eugene Goostman, “impersonated” a 13-year-old citizen of Odessa. The program, created by two engineers (a Russian and a Ukrainian), is well known to every enthusiast of the industry today, and its achievements have been widely publicized in the media. In spite of the fact that there were also some skeptical comments that the algorithm managed to pass the test only because it “pretended” to be a teenager and not an adult, who in addition was communicating in a foreign language, the event boosted the ambitions of engineers and entrepreneurs. The development of an intelligent robot that “communicates like a human being” became even more vivid and achievable. Stanusch Technologies also has aspirations in this area.
Eva – our chatbot with ambitions.
At Stanusch Technologies we have created and are constantly developing a chatbot called Eva, which this year for the first time was submitted for the Loebner Prize**. It is an annual competition held to determine which chatbot dialog will be considered by the judges to be the most similar to a dialog with a human being. The competition is inspired by the Turing test. More detailed information will be available soon! Meanwhile, if you are interested in this project, you can talk to the chatbot today. You can meet Eva at: https://chatwitheva.com/.
If not the Turing test, then what?
The Turing test, although known for years and still considered a determinant in assessing human-machine interactions, encounters some objections. More and more often it can be said that in order to create machines that “imitate human beings”, we need more than just balanced and context-sensitive answers to questions. The ability to talk for many researchers is not sufficient in itself to describe a machine as “capable of thinking” or to compare it to a human being. Interestingly, Turing himself agreed with the fact that it is virtually impossible to answer the question “Can a machine think?” Thinking is connected with consciousness. According to Turing, since the only being able to confirm its consciousness is the object itself (whether in the case of a machine or a human being), the attempt to answer this question by an external judge is unjustified.
Atom2Vec: a different approach to machine verification
In that case, more than 60 years after the creation of the Turing test, how do we attempt to determine the abilities of machines? Market giants (such as Google and Facebook) create their own tools (based on algorithms) to look at what to expect from artificial intelligence programs in a different way. An example of an alternative “machine test” is the Atom2Vec (Stanford University), based on Word2Vec from Google. According to the concept of scientists, the competence of an intelligent machine should be measured not so much by the ability to talk, but rather by the ability to make discoveries. And the first of them Atom2Vec has already achieved. After supplying it with some basic information it is able to e.g. discover properties of chemical elements and chemical compounds.
Machine tests: what else are we going to test?
With the development of the industry itself and the purpose (applications) for algorithms there will certainly be more ideas about how to determine the advancement of a machine. The question of whether the machine will ever be able to think, and if so, how to measure it, is still open.
* Alan Turing (1912-1954) – a British scientist considered to be the “father of artificial intelligence”, mathematician, cryptologist, creator of the Turing machine (an abstract machine capable of creating algorithms), as well as the Turing test. In 1939 he was employed as a cryptanalyst at Bletchley Park. There, based on the works of Polish scientists, he constructed the so-called Turing bomb, which enabled him to break the Enigma Cipher. In 1945, he was awarded the Order of the British Empire for his merits during World War II, but his achievements remained a mystery for a long time. The Turing test was developed and conducted long afterwards, in 1950. During this time Turing worked in Manchester, where he was involved in developing software for one of the first computers in the world.
** Loebner Prize – an annual contest to design a chatbot, the conversation with which will be considered the best simulation of a dialogue with a real person. The contest is based on the Turing Test. During the contest, the judges ask questions through the terminal to the software or the human being, not knowing who is on the other side. Based on their answers, they have to decide whether they have spoken to the software or to the human being.
The first edition of the competition was organized in 1990 by Hugh Loebner and Cambridge Center for Behavioral Studies. The main prize, which nobody has won so far, is an 18-carat gold medal and $100,000. It will be granted to a person whose program passes the Turing test. Once the main prize is awarded the contest will be put to an end.
Marketing Department Stanusch Technologies
Sources and recommendations:
- Poznaj Evę, naszego chatbota: https://chatwitheva.com/
- Noam Chomsky: Turning on the „Imitation Game” w: Turning on Parsing the Turing Test. Philosophical and Methodological Issue in the Quest for the Thinking Computer. R. Epstein, G.Roberts, G.Beber, Springer.
- Polskojęzyczna informacja prasowa o przejściu testu Turinga: https://www.tvn24.pl/wiadomosci-ze-swiata,2/komputer-po-raz-pierwszy-przeszedl-test-turinga-udawal-13-latka,437287.html