A.I. Understands the "Beauty Standard"

A.I. now has the ability to understand a human’s perception of attractiveness, a difficult task even for humans themselves, as beauty is incredibly subjective.

Reading Time: 3 minutes

What does “attractive” mean to you? You may take a look at people you find attractive and list similar features: perhaps a sharp nose and soft jaw or apple cheeks and almond eyes. But even with these descriptions, it is difficult to imagine what you consider “ideal.” With the use of artificial intelligence (A.I.), however, this may no longer be the case.

A.I., or the usage of software to create a system to perform tasks such as speech recognition, visual and odor perception, and decision-making, has become more prevalent over the years. From Siri’s interaction with the average person to robots assisting doctors in surgery, A.I. makes an appearance on a daily basis.

A.I. now has the ability to understand a human’s perception of attractiveness, a difficult task even for humans themselves, as beauty is incredibly subjective. Since many cultural and psychological factors unconsciously influence one’s preferences, it is difficult to identify or evaluate what one considers “beautiful.” Yet it is relatively simple for a machine to discern when an individual finds someone attractive, as their body emits different signals through dopamine.

Using this knowledge, as well as face-generating software, researchers at the University of Helsinki and the University of Copenhagen conducted an experiment to provide a machine with knowledge regarding human perception and test its accuracy. They first gave a Generative Adversarial Neural Network (GAN), a type of artificial neural network, the task of generating hundreds of portraits of people. These images were presented to 30 volunteers, whose responses to the photos were recorded through electroencephalography. The volunteers looked at the images while electrodes on their scalp monitored their brain activity. When an individual saw an “attractive” person, the left ventral tegmental area of the brain became active and released dopamine. The data was then provided to GAN, which evaluated it and produced a new face built from the most prevalent characteristics in the “attractive” faces. These new images, which were tailored to the individual’s preferences, were matched against the control group to check the accuracy of the program. The results were satisfactory: the generated face aligned with the volunteers’ preferences over 80 percent of the time.

By understanding subjective preferences, much can be achieved. Many apps currently use manually entered data or a user’s pattern of interactions with the app to personalize content. With advances in this research, the software can better align with the user’s interests, especially since hashtags or likes are not the only determining factors in whether a user will find the content interesting. Tinder, for example, puts users on a hierarchy by assigning points for every left and right swipe and then pairs people with similar points with each other, though this algorithm is quite faulty, as taste is subjective. With A.I., however, the app could provide users with profiles that are consistent with the user’s preferences rather than their “Tinder status.” The technology can also find its use in the beauty industry, as versions of the program could sort through applications without humans having to manually flip through each one. It could generate a “base image” reflecting target features or looks of a certain model agency as a means of creating a pool of applicants that aligns with the photoshoot’s concept.

Despite such potential benefits, the software presents several ethical questions. The program ultimately feeds into harmful beauty standards. While its impact can be sugar-coated as reflective of a single person’s preferences, there is no denying that Eurocentric features have become the beauty standard. With the use of A.I., this “beauty standard” becomes even more apparent, as “ideal” features are used to generate an “ideal” face. Though our society is slowly departing from beauty represented through Eurocentric facial features, a majority of models, actors, makeup gurus, celebrities, and representatives of the beauty industry still fall under that category. It is no surprise that creating the “ideal” attractive face will pursue this standard. With the addition of this research, an idea that a particular face with particular features is better than any other is cemented as the portrait is artificially created to combine “superior” traits. This is a step backward from the idea that beauty comes in all shapes, sizes, and forms.

There are many concerns with the implications and implementations of this new research and there is no doubt that it will exacerbate the biases that society already carries. Perhaps a subtle and less direct use of the software could make technology more personalized for the user, though it may never entirely escape ethical criticism. We should be taking a step away from consolidating beauty under a strict standard. While this research may be successful in the STEM field, it seems destructive on a humanitarian level.