Scientists at Osaka University used a neural network trained in facial recognition to identify
For their experiment, scientists formeda sample of 200 student faces. All participants were photographed from the front with a neutral facial expression without glasses or accessories. The researchers also assembled a group of 30 volunteer evaluators who did not know anyone from the original base.
The evaluators were shown randomly for 0.5 sall photos from the set are in order. They had to rate how much they would trust money to that person on a scale of -3 to +3. The researchers used the pre-trained ArcFace AI system to determine the difference vector between each estimator and the model.
The vertical scale is the similarity vector,calculated by AI (the smaller the number, the more similar people are to each other), the horizontal scale is the level of trust. The red line shows the results for the model and the estimator of the same sex, the blue line shows the results for opposite-sex couples. Source: Nakano, Yamamoto, Nature
The study showed a strong correlationbetween the increase in confidence and the similarity of facial features if the rater and model were of the same sex. The authors of the publication note that previous work has established that trust is suppressed when the amygdala is activated. Scientists believe that when we see similar facial features, this part of the brain remains inactive.
The ability to evaluate the level of trust based onFace similarity using AI will be widely used, the researchers believe. For example, this approach can be used to create personalized promotional offers or avatars in P2P lending systems that will appeal to a particular user.
Read more:
A black hole in the galaxy proved Einstein right. The main thing
"James Webb" will show space before the Big Bang: astronomers on the verge of a global discovery
Look at the celestial "Titanic" that will run on nuclear power