Understanding others’ emotions is crucial for navigating everyday life. But how do we interpret someone else’s feelings? And how accurate are our interpretations? 

Seeking answers, Marianne C. Reddan, M.A., Ph.D., and colleagues used fMRI scans to track brain activity in 100 study participants while they watched videos of others describing key life events. Viewers rated how emotional the speakers appeared to feel, while the speakers rated the intensity of their own emotions. When brain patterns between observers and speakers aligned, observers were deemed to be better at recognizing others’ feelings—a measure called empathic accuracy. Using this data, the team developed two machine-learning models: one to estimate a speaker’s intended emotional intensity, the other to estimate observer’s inferences. They tested the models on video clips not used in the earlier training, finding that the fMRI models of intent and inference can be combined to predict empathic accuracy. 

Perhaps the most compelling finding was that even when observers misinterpreted the speakers’ emotions, their brain activity showed patterns matching the speaker’s actual feelings—suggesting our brains subconsciously register others’ emotions, even when our conscious interpretations go awry. The findings could lead to new treatments for people who have trouble making emotional connections or understanding social cues, such as those with autism, anxiety, or depression. The study was published online on July 8 in Nature Communications.

Dr. Reddan is an assistant professor of psychiatry and behavioral sciences at Einstein.