AI app may predict conflict between couples before it occurs, scientists say

Researchers at the University of Southern California (USC), have used an artificial intelligence system to analyze language patterns and physiological signs in order to detect conflict in couples. The work, which the team published in the journal IEEE Computer, demonstrates the first time such monitoring has been shown to work outside of a psychology lab.

Led by USC’s Adela Timmons, the researchers used smartphones and wearables on 34 couples in the comfort of their homes. The devices were equipped with machine learning algorithms that could pick up on patterns in speech and physiology, such as an increased heart rate and skin conductance level, two signs that previous studies have shown to be associated with conflict. The red flag language would be all too familiar to anyone who has ever argued with a significant other, including more use of words like “you,” “always,” and “never.”

During the day-long trial, a sensor on each partner’s chest measured heart rate, a wrist band measured skin conductance level (think, electrical activity), and a smartphone collected audio recordings while tracking the participants by GPS. The smartphone would ask the couples to confirm if a conflict had occurred when one was detected — which surely made things worse but, hey, it’s for science.

All things considered, the system was able to accurately detect conflict 79.3 percent of the time. With language cues alone, the accuracy rate was a less impressive 62.3 percent. However, the system’s ability to detect conflict doesn’t yet amount to predicting — and even preventing — altercations. Timmons plans to investigate that next.

“Our next steps are to predict conflict before it occurs and to develop adaptive, real-time intervention systems,” Timmons told Digital Trends. “This might involve sending warnings or alerts that conflict is likely, prompting relaxation exercises or breaks, or helping couples re-initiate positive contact or reflect about what happened after an argument occurs.”

Correction: This study comes out of the University of Southern California, not the University of California, Los Angeles, as previously stated.