It is possible to prevent "impersonation crime" by AI analyzing mouse movement
"Impersonation crime" is a worldwide problem that sees someone's phone number or credit card number, or improperly acquires login information such as SNS, and other people's pretending to do wrong. However, AI Research has shown that it is possible to detect "spoofing criminality" by analyzing the mouse operation of the other party.
Identity theft can be thwarted by artificial intelligence analysis of a user's mouse movements - Quartz
This research was conducted by a research team at the University of Padua in Italy, and 40 respondents were asked to answer personal questions using a mouse, and we found a technique to spoof an imposter by the difference in reaction . In the experiment, half of the subjects supported to answer the correct contents to the question, while the remaining half of the group memorized information about "lie of the lie" different from yourself and asked questions He said he was letting me answer.
The question asked is "Are you living in Padua in Italy?", "Are you an Italian?", Etc. It was said that a type of which a spoofing offender can memorize relatively easily was chosen about. During reply, the computer tracked the process of tracking the movement of the subject's mouse and choosing the correct answer.
However, the research team made the concealed ball miss in the second lap question and investigated what type of reaction the "spoofing crime" group shows. The content of the question is "What is your constellation?" That was not informed in advance, and if it is about yourself, if you are pretending to be pretty, you will not easily think of content It was said that it was.
As a result, the difference between them clearly appeared in the form that the time required for reply was totally different. When the black line in the graph below is the theoretical best answer pattern, the impersonator group (red) was significantly divergent compared to the group (green) to answer properly.
As for the actual behavior pattern, when choosing correct answer from the pull-down menu displayed by clicking the mouse, the group that answered properly reached a straight line with the correct answer, but the spoofing group once made all the options We saw behavior patterns such as answering after seeing. By giving inquiries asked questions, it became clear that impersonators asked about information that they do not know show a unique behavior pattern.
The research team conducted an experiment to let AI learn this result and judge whether it is actually a spoofing criminal or not. As a result, he succeeded in finding a spoofing criminal correctly with 95% probability. Although it is also stated that this method can be utilized to prevent to spoof online only, it seems to be said that it is worth considering as a new method of crime prevention using AI.
Papers by the research team can be viewed from the following links.
The detection of faked identity using unexpected questions and mouse dynamics