Sasin’s Research Seminar series continued with a look at the relationship between an individual’s face and their name. The talk explores whether someone’s name could be predicted based on their features and is based on a paper in progress. Dr. Pinnaree Tea-makorn began the lecture by establishing that people’s faces constantly change. Factors that alter appearance can be physical, such as age, diet, exposure to the sun, or emotional states and personality types. This is especially true now that plastic surgery can help people choose their appearance. There has been considerable research into the links between faces and psychological traits. For example, studies have shown that attractive people tend to be more extroverted, and conversely – some extroverts become more attractive over time. Prior research into these issues has faced several problems. Data collection can be complicated, costly, and limited due to smaller population samples. Additionally, methodology issues result from reliance on humans, making data harder to quantify and standardize. However, with internet access, collecting data and getting participants from broader demographics is a lot easier. When it comes to methodology, computers, Artificial Intelligence, and algorithms can help with a more objective analysis. Dr. Pinnaree then discussed a paper written by a team from Israel and France that claimed they had trained a computer algorithm that could identify names from faces. This seemed unlikely and was the inspiration behind her deciding to replicate the study using a US sample. The topic of baby-naming was then examined. Parents can name children based on family names, places, or even influential people, and names can have consequences. You can tell quite a lot from a name, including gender, ethnicity, religion, age, and even socioeconomic status. However, the French and Israeli study claimed that an algorithm could be trained to look beyond these factors and identify names from faces. After conducting eight studies across Israel and France, they concluded people and computer algorithms could match faces to names significantly more accurately than chance. Certain factors lend credence to this, such as names or nicknames based on a physical feature. There’s also the ‘Dorian Gray effect’, where personality can change a face. Then there’s the idea that people might form a stereotype of what someone with a particular name might look like and treat them in a certain way. This treatment might affect the personality, and consequently, alter the face. To test this, Dr. Pinnaree got photos of 48 white Americans in the San Francisco Bay Area, with the top 300 most popular names for their age group (20-29). The photos came from two sources – social media and lab conditions. It was thought that social media would have a higher name correlation, as the pictures would have been chosen to reflect personalities. The lab setting allowed for hair to be tied back, removing hairstyle influence. The preliminary results surprisingly showed the lab photos produced a higher correlation. Empirical popularity was measured using a variety of checks and equations. Next, to account for human bias, machine learning algorithms were also employed, using a large dataset of over 300,000 white Americans and 182 names. A facial recognition algorithm gave a 256-dimensional face vector which was used in relation to a pair of names. The algorithm was trained and then gave a probability for which face was most likely to match with which name. The highest probability was then chosen as the answer. The results showed that humans were not able to match a name to a face. However, facial recognition algorithms were considerably more accurate than chance. The paper is still being written, and Dr. Pinnaree pointed out there is still a lot more to be done. The talk was followed by a Q&A session that discussed a range of topics, including how this could be applied to Thai nicknames, and how it might apply in different cultures.