Computers

AI can tell women from men based on their smiles

AI can tell women from men based on their smiles
Say cheese: A new AI algorithm uses smiles to identify gender
Say cheese: A new AI algorithm uses smiles to identify gender
View 2 Images
The algorithm analyses the movement of 49 points on the face
1/2
The algorithm analyses the movement of 49 points on the face
Say cheese: A new AI algorithm uses smiles to identify gender
2/2
Say cheese: A new AI algorithm uses smiles to identify gender

Artificial intelligence can tell women from men based on their smiles, new research suggests. This is because women's smiles are broader than men's. The finding has allowed researchers to create an AI algorithm that can identify gender 86 percent of the time. The research raises interesting questions about AI's ability to deal with complex issues of sex and gender.

"Anecdotally, women are thought to be more expressive in how they smile, and our research has borne this out," says Professor Hassan Ugail, who led the research. "Women definitely have broader smiles, expanding their mouth and lip area far more than men."

The method looks at the movement of the face when smiling, rather than using a still image. It's thought to be the first work in the field to do so. It works by tracking the movement of 49 points on the face. These "landmarks" are mainly close to the eyes, mouth and nose. The team analyzed 109 people: 69 females and 40 males.

"Because this system measures the underlying muscle movement of the face during a smile, we believe these dynamics will remain the same even if external physical features change, following surgery for example," Professor Ugail explains. "This kind of facial recognition could become a next-generation biometric, as it's not dependent on one feature, but on a dynamic that's unique to an individual and would be very difficult to mimic or alter."

The algorithm analyses the movement of 49 points on the face
The algorithm analyses the movement of 49 points on the face

The team thinks there's room to improve on the success rate. "We used a fairly simple machine classification for this research as we were just testing the concept, but more sophisticated AI would improve the recognition rates," Professor Ugail adds.

However, the research raises questions about gender identity and transgender people, which the team hopes to address with more research. The team points out that gender recognition plays a role in "face perception, age, ethnicity, identity analysis, video surveillance and smart human computer interaction," among other things.

The use of the word gender rather than sex here is interesting. Sex is based on anatomy, whereas gender accounts for how a person feels and identifies. We've asked the researchers about the significance of this term in relation to the research.

Separate research has shown that existing facial recognition technology is often biased towards white males. Research such as this, working alongside similar technology, could help to make facial recognition software more sophisticated. The team thinks other facial expressions might help with gender recognition, including "surprise, fear, anger and disgust."

This research is being undertaken at the University of Bradford in the UK. It has been published in Visual Computer: International Journal of Computer Graphics. We'll update this article should we hear more from the researchers.

Source: University of Bradford

UPDATE (Mar. 20/18): Professor Hassan Ugail got back to us on the question of gender and sex. "Sex has a strong association with the anatomical features," Ugail tells New Atlas. "The smile one bears relates to the characteristics of facial muscles which must be different in the two sexes. We have used the term gender because this is commonly used in the literature, especially in computer science."

6 comments
6 comments
KirkAugustin
Wonderful, but the problem of this is like the problem of all AI, autonomous vehicles, etc., in that 85% is worthless. If you can not be 100% reliable, as humans are, then the calculation is not useful. People in particular object strongly to be sexually misidentified 15% of the time, or have fatal accidents 15% of the time. Computers are centuries away from the inherent abilities of humans. We must not forget that.
Douglas Bennett Rogers
I think I can identify a female smile with 99.5% confidence.
Derek Howe
Kirk - Humans are not 100% reliable, that's a joke. Hence the phrase "I'm only human".
AI is behind the human brain, no doubt, but it won't take centuries, it will take less then 20 years, and they will out smart us in every way imaginable.
Daishi
There used to be (is?) an annual image recognition competition where teams to have to classify images from the imagenet database. Over the years the teams first approached and then passed the human error rate on the data set. As of 2015 I think Microsoft had only a 3.5% error rate. Humans are good and facial (and emotional) recognition specifically is a skill humans have evolved to be very good at but humans are not 100% accurate. An 86% success rate in identifying gender for a first a first attempt is pretty impressive.
bwana4swahili
Could really confuse the transgender issue! I feel like a woman in a male's body (or visa versa) but AI thinks I'm actually a man (or woman); wow, where do I go from here. So CONFUSED!!
dib
@KirkAugustin Really. You think humans are 100% at anything. The issue is that how is a human or AI supposed to identify the 1.2 million different genders? Identifying the "sex" of a person is about 98% accurate for both AI and human. It's not the AI or humans fault someone doesn't like the sex they were born as. It is dealing with facts.