Know better than you what looks can attract your AI

Appearance preferences may change over time (such as having a mustache or wearing a monocle), but in any case, top social status can not only attract a significant other, it may even help you get promoted, or make it easier to meet powerful elites. Despite the multifaceted social impact of "look", the mechanisms behind these preferences remain unfathomable. Rarely do we make a "conscious and logical judgment" about a person's appearance first, and often guess the cause after experiencing the feeling, but artificial intelligence may be able to help solve this problem.

Finnish research team devised a machine learning algorithm that knows better than you what looks attract you

Humans have long wanted to know the answer. For centuries, mathematicians, philosophers and artists have been trying to define what "beauty" is. For example, the golden ratio is used to quantify beauty. Even today, the media still use the golden ratio to judge whether a person is close to "perfect". But in recent decades, the overall symmetry of the face (such as the distance between the facial features) has gradually replaced the golden ratio as the mainstream theory of whether a person looks perfect.

From the proportions of curly ferns, spiraling snail shells, and even the Statue of Liberty, the golden ratio has long been considered an ideal of beauty. But that's not necessarily a good predictor of personal attractiveness.

Research by Tuukka Ruotsalo, associate professor of computer and information science at the University of Helsinki, shows that these theories are clearly not always correct, saying: "Everyone has a different opinion about whether a person is attractive or not, and if the research model only Looking at pictures, it can never really understand what's attractive and what's not. Our research looks at how different individuals respond to various photos, and then feeds the results back to the AI."

We seldom make a "conscious and logical judgment" about a person's appearance first, usually we first experience the feeling before guessing why

In order to truly understand why we find a person to be particularly attractive, while some people are completely unattractive, it is necessary to study the neurons that trigger this biological response, but it is easy to say and difficult to do. Here is an example: "If I want to create a picture that I think is beautiful, it's still difficult to say what I think, because there are so many different characteristics and aspects of appearance." Rusalo and colleagues mentioned in the paper, this study is the first to show Brain responses as interactive feedback to generate neural networks, he said: "What we're doing is controlling a very unknown and complex domain."

Just as the brain operates on the firing of neurons, the "brain" of a machine learning model also operates on something called a "neural network," which is a network of connections that uses algorithms to help artificial intelligence process new inputs and find patterns Learn from it.

The human-machine interface and machine-learning model of this study is divided into three main parts: First, when the participants are exposed to new stimuli (such as human faces) in the experiment, the scientists record the neural data of the participants through the EEG; Second, a machine learning network called a "generative adversarial network" (GAN) learns from a set of training data (usually a gallery of celebrities), then infers patterns to generate new pictures; finally combines the neural data of the brain with the trained The combination of neural networks created a "generative human-computer interface" (GBCI), an interface that uses brain signals to generate new images.

Artificial intelligence may unravel human elusive subjective judgments or thoughts

In short, all the steps were to create a unique and attractive look: The research team showed 240 pictures of celebrities to 30 participants wearing electrode caps, and the participants found it attractive each time they saw it. Scientists watching from the side record the neural activity of their brains when they look at the force. But because the EEG data is too ambiguous to derive any variable (such as whether a person finds a certain lip shape attractive) data, it can only "like or dislike" the picture like a dating app.

Therefore, the neural data of the brain and the trained generative adversarial network are combined into a generative human-machine interface, and the data is used to combine the looks that each participant finds attractive, thereby creating a new and unique look that matches the preferences. When the research team presented the new pictures to the participants, it was found that the majority of the participants rated the new pictures as the most attractive, with an accuracy rate of 80 percent.

When will generative human-machine interfaces affect the future? When will it be applied to dating software? Researcher Rusalo pointed out that the research is still in the early stages, and it is difficult to accurately predict how or in what form the technology will eventually be used. The research team showed that there are many other experiments to be carried out before this, and more participant data and machine learning data are necessary. In fact, the novelty and potential impact of this technology is that it may unlock the elusive subjective judgments or thoughts of humans.