Perhaps in the future, the Internet will be cleared of fraudsters, and the users themselves will have to carefully monitor their words. Researchers at Florida State University have developed artificial intelligence that perfectly acts as a polygraph and detects lies by monitoring the words the user sends and the speed of their typing. The test results showed that he is already able to distinguish truth from lies in 85% of cases, while the accuracy of real lie detectors fluctuates only in the region of 60-70%.
To train the system to distinguish truth from falsehood, the developers resorted to machine learning. A group of forty volunteers were asked to play an online game in which they were randomly assigned the roles of "saint" or "sinner." During 80 game sessions, they answered questions, and the artificial intelligence carefully monitored their messages.
It turned out that "saints" or "sinners" use completely different sets of words and type them at different speeds. For example, liars often use the words "always" and "never" to emphasize their confidence and respond very quickly. Truth-telling people often use the word "maybe" and type more thoughtfully and accordingly slowly.
As a result, by paying attention to the frequency of words and the speed of response, the online polygraph has learned to recognize a lie with an accuracy of 85 to 100%. The researchers admit that too few people took part in the test, but even now it's clear that the technology has a lot of potential. They intend to improve it and contact the developers of various messengers to popularize their creation.
Do you want to implement such a system in instant messengers and other methods of communication on the Internet? How often do you come across lies on the web?
Ramis Ganiev