by Ioana Tanase
Improving mental health remains a necessity and focus for many people. every year, One in five adults In the United States, mental illness is diagnosed. The demand for mental health professionals has risen in recent years. according to American Psychological AssociationThe number of psychologists who reported receiving an increase in referrals in 2021 doubled from the previous year, and 7 in 10 psychologists who already had a waiting list reported it had increased since the beginning of the pandemic.
To meet the growing need for support, many mental health organizations have taken the path of peer-to-peer (or peer-to-peer) support, connecting millions of people online. This model allows to expand the range of services, but there is scope for more effective and successful conversations between users. Peers usually have good intentions and a desire to help others, but they may not know the best tools and strategies, such as empathy, which is the ability to understand and feel other people’s feelings and experiences.
A team from the University of Washington led by an assistant professor Tim AltoffGive Paul C. Allen College of Computer Science for Engineering and manager Behavioral Data Science Group, to explore ways to build more empathy in peer support conversations on mental health platforms. The team included University of Washington doctoral students Ashish Sharma and Ina Lin, as well as clinical psychologists Adam Miner and Dave Atkins. During previous searchThe group found that empathy can wane over time and is not a self-learning skill.
The goal was to study how AI systems can collaborate with humans to facilitate empathy in online text-based peer support conversations. The idea was to empower peers on an online mental health platform through feedback and coaching. For example, writing systems machine in the loop It can help their peers express higher levels of empathy, making online mental health platforms more effective.
It is difficult to extend traditional methods of empathy training, such as face-to-face training for counselors, to millions of online users. Computational methods can facilitate a more empathetic peer response at scale to meet needs and improve outcomes for a user seeking support.
To exploit this opportunity, the UW team trained natural language processing models to identify and improve empathy in supportive peer conversations. They also designed interactive tools to provide real-time model-based feedback to supporters. This includes partner, a new learning agent able to learn to make text modifications to increase empathy in conversation. By utilizing the Partner Program, researchers have developed and evaluated a collaborative, interpersonal and AI-based approach to help people write more empathetic responses.
A video about the project and a demo can be found at Youtube.
This research helped improve understanding of the role of empathy. A randomized trial of 300 peer-supported TalkLifea mental health app that provides a global network of peer support, showed that this model led to a 20% increase in expression of empathy overall, and a 39% increase for participants who reported having difficulty expressing empathy.
Analysis of human-AI collaboration patterns showed that participants were able to use AI observations directly and indirectly. Users were able to embrace the feedback generated by the AI without being overly dependent on it, and report improvement after receiving the feedback.
Empathy is key to supportive conversations. It has strong links to improving symptoms which is crucial in building links. Specifically, on online support platforms such as TalkLife, researchers found that compared to non-empathic conversations, empathic conversations receive approximately 45% more “hearts” and “likes”. In addition, they have an 80% probability of providing the formation of relationships. “Mental health research rarely has the practical application and potential impact that we think this research will have,” says Jamie Drouet, CEO of TalkLife. “This research directly addresses real challenges and can be implemented to bring about measurable change in mental health platforms.”
Research and project also received WebConf Award for Best Paper in 2021. Kira Radinsky, chair of the committee responsible for the award for best article, shared: “We feel that the work will have a significant impact on society based on the computational approach, as well as the comprehensive experiments carried out in the newspaper itself. All this year, this is the direction in which We feel it will bring the most value to society.”
“Friendly zombie fanatic. Analyst. Coffee buff. Professional music specialist. Communicator.”
