Do you look like a Mark? Have you been told you seem ‘more like a Jane?’ You may think you look suited to a certain name and now science has shown others may agree with you. A study conducted by the American Psychological Association has found people are more accurate at correctly matching names to faces than can be attributed to pure chance. The downside? It may be due to cultural stereotypes.
In perhaps the most scientific application of the Guess Who? game ever recorded, hundreds of participants from Israel and France joined an experiment to match photographs of people with their corresponding names, from a list of four or five choices. The aim of the study was to explore the impact of social identifiers upon physical appearance.
The students were given a mix of French and Israeli faces and names. The French students were found to be better than random chance at matching only French names and faces, whereas the Israeli students were better at matching only Hebrew names and Israeli faces. On average, participants in the experiment achieved 25 to 40 per cent accuracy, whereas random chance would have predicted an accuracy of 20 to 25 per cent.
Yonat Zwebner, a PhD candidate at The Hebrew University of Jerusalem and lead author of the study, believes the results of the experiment may suggest people subconsciously alter their appearances to conform to cultural norms and cues associated with their names.
"Prior research has shown there are cultural stereotypes attached to names, including how someone should look. For instance, people are more likely to imagine a person named Bob to have a rounder face than a person named Tim. We believe these stereotypes can, over time, affect people's facial appearance," he says.
Factors such as hairstyles or tattoos - aspects of appearance that are controlled by the individual - may, therefore, be part of subconscious cultural conditioning.
“A social tag may influence one's facial appearance," added co-author Ruth Mayo, PhD, also from The Hebrew University of Jerusalem. "We are subject to social structuring from the minute we are born, not only by gender, ethnicity and socioeconomic status, but by the simple choice others make in giving us our name."
In order to determine the true bias of human subjects, a computer was trained with a learning algorithm to similarly match faces with names. This was done using Deep Convolutional Neural Networks, which are currently state-of-the-art for image-classification tasks. The trained learning algorithm was able to accurately match the true name significantly above chance level - without the knowledge of semantic, phonetic, or cultural information that a name holds.
This facet of the study included more than 94,000 facial images and the computer proved significantly more likely to match the correct names and faces with an accuracy of 54 to 64 per cent.
Nir Rosenfeld, the co-author of the study told WIRED that, unlike how people perceive information, a deep learning system lacks any cultural, historical, or contextual information: "Name A is name A only because it is not name B." The computer algorithm allowed for a separate tool of measurement in deducing what the team had hypothesised generated the phenomena in humans.
This raises the question of whether there is a correlation between someone's face and their name, beyond stereotypes, if a computer can determine them with such accuracy. When WIRED posed this question to the researchers, they said it was simply a coincidence. They added the cross-cultural human-subject experiment shows that for humans, cultural context is important. For an algorithm, it is simply inaccessible.
This article was originally published by WIRED UK
