Program Recognizes Frustration Better Than People Do
The difference between delight and frustration seems pretty clear, yet a computer program can tell the difference between delighted and frustrated faces better than people can, according to a new study.
Part of the confusion among humans seems to stem from a fact the study discovered: that most people actually smile when frustrated, even though they are not aware of doing so. It's a quirk of human nature that may be difficult to suss out but is important for engineers to acknowledge in building robots that read human emotions.
The research team published their work in April in the journal IEEE Transactions on Affective Computing, along with results from a program they wrote that identifies frustrated smiles much better than people can.
This study is one of several done around the world to get computer programs to recognize human emotions. The research team, from MIT's Media Lab, wrote in its paper that it will be important for robots in the future to see people's happiness, anger, fear, surprise and other emotions. For example, engineers may build robots to serve people at stores ? and such salesbots shouldn't mistake frustration for delight. [Top 7 Useful Robots You Can Buy Right Now]
The research also may be helpful for some people, such as those with autism, who have trouble understanding others' emotions just by actively looking for cues such as smiles. "The goal is to help people with face-to-face communication," Ehsan Hoque, a graduate student who led the study, said in a statement.
To capture people's frustrated faces, Hoque and his colleagues recruited 27 people to fill out a form on a computer. Unknown to the volunteers, the form was programmed to erase the people's input and give them error messages. Afterward, the study volunteers told researchers they were frustrated during the exercise. Nevertheless, videos of the volunteers showed 90 percent of them smiling.
To get the same people to feel delighted, the researchers showed them a popular YouTube video of a laughing baby and recorded their faces, all containing smiles. The volunteers later acknowledged their delight.
Then the researchers wrote several computer programs and used their frustrated-smile and delighted-smile images to train the programs to find differences between the two expressions. When they tested their programs with new photos, they found their best one was 92 percent accurate overall and identified all frustrated smiles correctly.
Meanwhile, new study volunteers could correctly identify frustrated smiles only 53 percent of the time.
People and computers could identify genuine, delighted smiles equally well, the researchers found.
How do people tell, then, the difference between two very different feelings that produce the same reaction? For one thing, most people don't just look at pictures of faces when they're interacting with each other. They can probably tell the difference from people's tone of voice, what they're saying and other body language, researchers wrote. Another cue is the duration of the smile, they wrote. In the study, frustrated smiles appeared and disappeared in 7.5 seconds, while genuine smiles lasted an average of 13.8 seconds.
Here are the answers to the smiles above: