Artificial Intelligence Has Learned To Look At Magic Tricks In A Human Way - Alternative View

Artificial Intelligence Has Learned To Look At Magic Tricks In A Human Way - Alternative View
Artificial Intelligence Has Learned To Look At Magic Tricks In A Human Way - Alternative View

Video: Artificial Intelligence Has Learned To Look At Magic Tricks In A Human Way - Alternative View

Video: Artificial Intelligence Has Learned To Look At Magic Tricks In A Human Way - Alternative View
Video: 20+ Psychology Tricks to Read Anyone Like a Book 2024, March
Anonim

Spanish researchers taught a computer vision algorithm to perceive the tricks of an illusionist with a coin the same way a person does. To do this, they asked a professional illusionist to show several tricks to the viewer and a recognition algorithm based on DeepLabCut, which is used to track laboratory animals. Two of the seven shown tricks were able to successfully deceive both a person and a computer, and the results of the work can in the future help in studying the perception of such tricks by viewers, the scientists write in a preprint on arXiv.org.

There is no magic in the magic tricks that illusionists show, the whole success of their implementation comes down to sleight of hand. On the other hand, it is also a matter of human perception: the actions of the illusionist are designed to mislead the viewer, playing on his attentiveness and concentration. Therefore, for those who follow the magician's hands extremely carefully, there is no magic, and deception in some tricks can be easily detected if, for example, you record his performance on video and play it slowly.

Of course, the situation with the perception of such tricks by computer vision algorithms is slightly different: in fact, the computer is freed from the possibility of being deceived, and in the case of it, how well it can recognize deception depends on the quality of its work. The researchers led by Alex Gomez-Marin from the Institute of Neurosciences in Alicante (Spain) decided to test whether such an algorithm can be taught to look at the tricks of the illusionists as a person.

To do this, the scientists hired a professional illusionist and asked him to show seven simple visual tricks with coins - without any verbal additions that can distract the viewer and affect the success of the illusion. The tricks were distinguished by the movements of the illusionist's hand necessary for the disappearance of the coin: for example, in one it was important to drag the coin on the table, and in the other, for example, to grab it.

All the tricks were shown to people, as well as an algorithm based on DeepLabCut, which was presented by German scientists last year: it is used to automatically track the movements of laboratory animals and can even analyze the movements of individual parts of their bodies (for example, the paws of mice). The task of the algorithm was to determine the location of the coin at the end of each trick - exactly the same task faced by the research participants.

Scientists compared the results of a person and an algorithm and found that only two cases were possible to deceive both of them. Three tricks that deceived the audience, the algorithm did not deceive - it determined the position of the coin. Also, one trick tricked the algorithm, but not the audience, and one - the other way around. For example, the fourth trick, in which the illusionist lays out coins in a row (you can watch it in the video), turned out to be simple for the algorithm, but was able to deceive the viewer, because the latter's attention during the movements was directed to the hand in which the illusionist initially held the coins. therefore, the fact that the magician was putting in a coin with his other hand went unnoticed. Since an algorithm trained to track a coin has no problem tracking both hands at once, it was not fooled. On the other hand, in the sixth trick - exactly the same as the first,but it was specially made with a mistake - the algorithm, unlike the viewer, failed to recognize the deception, since the coin tossed, apparently, turned out to be an edge in relation to the camera, which caused difficulties in recognition for a computer, and not for a person.

The authors clarify that they were not interested in the algorithm's ability to quickly figure out the illusionist's tricks. Rather, they wanted to see if it was possible to make him look at them the way an ordinary person looks, and not the one who seeks to solve the deception, but the one who actually perceives the trick as some kind of magic. The fact that in some cases DeepLabCut was not really able to recognize deception in the same way as a person, which means, according to scientists, that such algorithms can be used to analyze human perception - just in situations like the tricks of illusionists.

Promotional video:

Elizaveta Ivtushok