Millions of people worldwide suffer from depression or anxiety disorders. But exactly how these conditions develop in the mind remains elusive to researchers. A team of medical professionals, psychologists, and computer scientists at the Technical University of Dresden has now investigated whether AI language models can replicate typical patterns of human emotions and thought processes.
How Researchers Teach a Computer to Feel Fear
At the Else Kröner Fresenius Center for Digital Health at TU Dresden, researchers are working with so-called large language models. These are AI programs like ChatGPT that understand and generate human language. The team examined six such models and specifically triggered seven emotional states: anxiety, fear, anger, disgust, sadness, worry, and stress.
To do this, they used standardized text inputs and measured the reactions using the same evaluation scales that are also used in psychological research. The models subsequently exhibited patterns resembling human emotional reactions.
Another finding was even more surprising. When the researchers previously triggered sadness, the programs tended to complete sentences in a negative way. This pattern, known from depression research, is called cognitive bias. It is familiar from human behavior. The language models showed similar patterns.
