Research Press Release

Technology: Enhancing empathic support with artificial intelligence

Nature Machine Intelligence

January 24, 2023

An artificial intelligence (AI)-based chat interface, named HAILEY, that offers assistance to mental health peer supporters who interact online with individuals seeking support, is described in a paper published in Nature Machine Intelligence. Peer supporters who took part in the study reported an overall increase in conversational empathy when using the interface and also felt more confident about providing support.

Mental health has a serious impact on global health, with over 400 million individuals suffering from mental health disorders worldwide. Although access to therapy and counselling is limited, peer-to-peer platforms in non-clinical settings can provide some support, which provides health benefits and is strongly correlated with mental health symptom improvement.

Tim Althoff and colleagues designed HAILEY, a chat interface, which uses a previously developed language model that is specifically trained for empathic writing. They recruited 300 mental health supporters from the peer-to-peer platform TalkLife. In a controlled trial, participants were split into two groups, with one group receiving feedback on responses they wrote via HAILEY to real-world posts that were filtered to avoid harm-related content. HAILEY provided suggestions for phrases to either replace or insert, which participants could choose to ignore or adopt; for example, suggesting replacing the phrase “Don’t worry” with “It must be a real struggle.” The authors found that the human–AI collaboration approach led to an almost 20% increase in conversational empathy, evaluated by a previously validated AI model, and an almost 40% increase in conversational empathy for peer supporters who had reported experiencing difficulties in providing support.

The authors suggest that their findings show the potential for human–AI collaboration with tasks such as empathetic conversation, but they note that further research is needed to ensure the safety of these applications.

doi:10.1038/s42256-022-00593-2

Return to research highlights

PrivacyMark System