How can you Detection of a liar? In the past, the main way to detect a liar was Analysis of your non -verbal language or even your nervousnessField in the end, then the famous polygraph analysis cardiovascular and respiratory activity, as well as skin conductivityAll of them are factors that can vary significantly when a person is lying, although there are professional liars who practice to make their lies inconspicuous. Examination in our life artificial intelligence You can express it a little easier, although it can also be straight. And although there are numerous articles on the Internet that indicate that they are the most common words among liars in accordance with artificial intelligence, this is not an easy problem.
In general, these are words that do not indicate as such A liar, but they can be abundant among people who lie. Over the past month, numerous media indicated a list of 5 types of words that may justify that people are lying. However, despite the fact that everyone seems to be drunk from the same source, I could not find anything that mentions it. Everyone explains that this is the result of an analysis of artificial intelligence, but nothing else is said. Thus, logically, we cannot take it seriously.
Therefore, I made my own analysis, primarily with Quick review of scientific literature And then with some Questions aimed at the GPT chatThe field I went through real fragments of whatsApp conversations. Some truths and other false ones. But to what extent can you find a liar? Let’s see.
This is a liar dictionary
In these articles without a source that spread on the Internet, a link to five ordinary formulas in the liar dictionary. On the one hand, the link is made on Absolute termsThe field of words is how “Never” or “always” They seem common among liars. According to artificial intelligence, there are no gray for them. Only white or black.
It is also noted that these are common introductory words, such as “honestly”, which emphasizes the sincerity of the statement. As Latin speech says, The justification is not Petit, the accusation of the manifesto.
Thirdly, we find a confrontation of those absolute terms that we have seen before. Some liars prefer reduce confidence With words, such as “I believe”, “I suppose” or “maybe”.
It can also be given ambiguity With the word “but”. And finally Vague statementsWithout many details. For example, when you are talking about “this” or “this situation” without indicating what.
What does science say?
In the analysis of scientific literature, we find that the latter is quite common. For example, in 2018, a team of American scientists developed a study in which Machine training To determine the terms most used by people than They lie in working interviewsFor example, this uncertainty of lies, but also other features, such as Excessive use of justification and negative phrases.
Earlier, in 2003, a team from Texas University He also showed that the liar language is usually Loaded with negative phrases. In addition, there is little self -dumping. It is unusual for them to use direct pronouns in the first person for the only number. In fact, this is what was noticed in other studies of the subject conducted in the last two decades. In addition, there are studies, as in 2015, an international group of scientists, which also indicate links to sensory perception or cognitive processes.
In short, liar language Loaded with excusesNegative links and vague messages, Without indicating or deepening feelings or reasoning. They are trying to leave the road. This is that it seems that this was noticed with a computational analysis of recent times. But can artificial intelligence at the user level detect a liar?
Let’s see what artificial intelligence is capable of
I presented several real messages received in my WhatsApp from friends who did not initially inform me if they lie or told the truth. Of course, I have all the permission to share these messages.
First example:
“But yesterday I got on Instagram, and I liked it. They make Paulu Echevarrrian, Rosie de Palma and many more well -known. This is called Lausett. They are very charged, but cool. A bad thing that also encounters. ”
According to ChatGPT, there are signals such as the word “but”, the use of well -known characters or the expression “truth” that can be used to enforce reliability to the message. This may indicate that it is a lie, but they are also the natural signs of a random language. Completes the following:
“There are no clear signs of fraud in this message. The language used is spontane, with personal data and opinions. Although this includes a small initial excuse (“but”), it seems more natural to start a phrase than an attempt to cover. “
Asking how likely it is that it will be a lie, he notes that it is impossible to find out the available information; But even in this case, based on “linguistic analysis and models of deception”, he risks calculating From 10 % to 20 %, the probability of deception.
It should be said that The message is real Extracted from a conversation, and a person who said that this did not lie at any time. This is not a bad calculation.
The second example:
“Well, last Tuesday I got to the top of the ham and shrimp, and I began to be fatal until I had to tear it, I got out of control. I am sorry that I did not leave at the end of your birthday or gave you something, that I felt a lot, but when you could not. “
GPT chat talks about suspicious, but not convincing elementsThe field among suspicious elements is indicated by the dramatic and very sensory narrative (despite the fact that studies show that sensory links usually do not make), extensive justification and emotional language to soften the impact. Read Lies the probability between 40% and 60% and completes the following:
“The message can be Sincere excuse or emotionally well -built liesThere is no strong evidence there. If this person has a scheme for justification of their absence with the help of complex stories, this will increase suspicion. “
In this case, The message is falseSo, in a sense, we can say that it succeeded.
The third example:
“Well, I have reached the work, I cleaned several pots.
In this case, the GPT chat indicates that “message He does not show clear signs of lies And, in fact, it contains several indicators that are usually associated with Truthfulness “The field of some examples are specific and everyday details, a natural chronological order, a neutral tone and the absence of justification.
The probability that it is false is very low, 5 % to 10 %The field and this is right, since this is a message in which True says.
Fourth example:
“A few years ago, 3 or 4 years, I went to eat in a vegan restaurant, and it was more expensive than expected, but even in this case I ate the like. When I went to pay, they were erroneous, which would be accused, what was in the case of Bizum or effective, and I paid less than half. ”
According to the GPT chat, There are no obvious signs of liesAlthough some suspicious. This indicates that temporary ambiguity means sincerity, not knowing whether it is 3 or 4 years old, the natural language and amazing, but plausible details of the payment method. In addition, he sees sincerely, that there is no emotional excuse or attempt to look good. He sees only a little suspicious who speaks of a small traceable payment method, which is too favorable for those who say this. These are only two minimum details, so the probability of lies is From 10 % to 20 %The field again, it is appropriate that the probability was low, since The message is true.
Fifth example:
“Damn it, baby, I am very angry because I found out that Peter talked about the discussion that we spent the other day. And, of course, he said his version, because he has a good “bad say”.
In this case, the GPT chat believes that the message “has many characteristics Genuine emotional and narrative sinceritySo this is unlikely to lie. ”
These indicators of sincerity are emotional load, reasonable internal contradictions, colloquial language and subjective prospect. The probability that the liar wrote, according to artificial intelligence, is 5 %-10 %. Unfortunately, in this case Yes, this was not possible, since the message is incorrect.
It should be said that these messages were written by different people. Some knew that this would be used for this article, others knew AposteriorThe field of those who lie did it intentionally, although they could choose, lie or tell the truth. The GPT chat did quite well, but in one case he also failed loudly. Although this is not a study, but a daily experiment, this indicates that the AI is not as infallible as we thought, and that a liar hunt is more complicated than it seems. We cannot trust the algorithms for everything.
Source: Hiper Textual
