The company asked to take high-resolution photographs of the face in order to consult with plastic surgeons and create a mockup of an improved version of the face. Stone was advised to reshape the chin and fill in the hollows under the eyes. She was impressed then, but two years is a long time for the technology industry, and today such recommendations are not given by people, but by algorithms.
Many plastic surgeons donate photographs of clients for processing by AI, which creates a three-dimensional image of the face, taking into account facial expressions. Further, the algorithm tells what operations a person should do and by what percentage they will increase his attractiveness.
Plastic surgery is big business in the US and around the world. Last year, about 17.7 million such operations were performed in the United States. The American Society of Plastic Surgeons estimates that clients spent about $ 16.5 billion on them (excluding reconstruction procedures). Many surgeons are interested in any tool that can increase the profitability of their already successful business, and increasingly, these tools include artificial intelligence in one way or another.
Stone studied the programs of about 20 plastic surgery conferences that took place this year, and found that most of them involved discussion of AI-based programs that do everything from measuring facial attractiveness to making recommendations for working with specific clients.
However, some experts say that passing a beauty score to an algorithm might not be a good idea. "The use of AI in aesthetic judgments has the potential to destroy the cultural diversity of beauty," wrote plastic surgeon Jungen Koimizu in The Plastic and Reconstructive Surgery Journal in March 2019.
Many companies are using AI to analyze and predict customer behavior, and plastic surgery has not stayed away from this trend. Specifically, Heather Levits, a plastic surgeon at Duke University School of Medicine, is using a mood analysis tool from startup Cognovi Labs. With his help, she studies messages on social networks that mention plastic surgery.
For example, the program searches for tweets with words and phrases such as liposuction and breast augmentation, analyzes them and determines what users are interested in and what they have mixed feelings about. The classification is carried out by recognizing six emotions: surprise, anger, joy, disgust, fear and sadness. The algorithm then sorts the data by three dimensions: awareness, engagement, and motivation. The higher the motivation of the tweet author, the more likely he will decide on the procedure.
The results surprised Levits. Nose reshaping surgery is often discussed on social media, but Cognovi Labs' tool has found strong negative reactions to the procedure. “During it, we have to break the bones of the nose, and this causes anger and frustration,” explains Levits. People were less familiar with liposuction, but this surgery ranked first for emotional response. The analysis helped Levits understand how patients feel about different procedures, and she is now refining the parameters for another study. Ultimately, she hopes to develop a tool for surgeons in different regions that will allow them to adapt to the wishes of patients.
Other popular applications of AI among plastic surgeons are services such as BioMedX and Crisalix, which show patients 3D models of their bodies after surgery. But such programs use 3D scanning and are poor at taking into account changes in lighting, age, or skin tone.
Zurich-based developer Endri Dibra, who specializes in creating realistic 3D images, says his AI-based software for modeling the appearance of breasts after surgery is not suitable for African Americans. The fact is that the technology is based on data provided by surgeons from Switzerland, where only 0.6% of people with dark skin live.
AI bias is a well-documented problem. In particular, companies like Amazon and IBM have been found to incorporate gender and racial bias into their algorithms. The Amazon recruiting app biased against female candidates, and the IBM and MIT portrait generator changed the skin color of Asians and African Americans to white. These kinds of prejudices can be especially detrimental when judging beauty.
Some surgeons use AI tools that determine the beauty of a patient before surgery (often based on the principles of the golden ratio). And re-scanning the face after the procedure can provide quantitative data on how much more attractive a person has become. This has the potential to protect surgeons from lawsuits from patients who remain dissatisfied with the operation.
In a 2014 study, it was noted that the same tool will predict how a person will look after surgery and by how many percent they will become more beautiful. “Quantifying aesthetic improvements will help not only set expectations, but also dissuade patients from procedures that will give only marginal results,” said surgeon Jonathan Kanevsky. If a person becomes only 2% more beautiful, he will be able to think again whether the operation is worth the time and effort. In short, these programs have many applications, but who will decide what beauty is?
Koimizu, who wrote an article with concerns about evaluating the beauty of AI, worries that surgeons may tackle face fitting to the Western ideal. Result? “Marginalizing the value of beauty in other cultures,” he warned.
Attractiveness isn't the only AI-measured metric that raises questions. One study published in The Plastic and Reconstructive Surgery Journal in October 2019 assessed whether algorithms could determine the success of facial feminization surgery. Using four publicly available neural networks, doctors tested how AI types trans women. Before the operation, he made mistakes in 47% of cases, but after the operation he gave the correct answer in 98% of cases. For trans people, objectively identifying their gender will help them feel more confident, but defining what is "woman" or "man" can be as fraught with consequences as defining the degree of beauty.
There are also clearly positive examples of the use of AI in plastic surgery. For example, surgeons at Harvard Medical School, the Massachusetts Eye and Ear Center, the Royal Australian College of Surgeons, and other research institutions have commissioned AI to evaluate the results of skull surgeries in patients with facial paralysis. Specifically, they wanted to know if postoperative smiles convey genuine emotion. This is a helpful assessment. In Italy, surgeons are using AI to heal wounds. Their algorithm detects damaged skin with 94% accuracy, allowing for a more effective treatment program
Some of the uses of AI in plastic surgery are obviously beneficial. But listening to his assessments of beauty and recommendations for changing your appearance is creepy. It's good that surgeons now use AI as a guide, not as a guide to action. As long as this is so, everything is in order.