Human-AI collaboration will encourage empathic conversations as around 20% adults in the US face mental illness and are actively seeking treatment. This number has increased with the demand for psychologists, who report having to cancel appointments due to patient overload.
The lack of providers is being addressed through a peer-to-peer model, connecting millions of people online. Peer supporters can be effective with tools to help them, such as empathy.
The UBC mentors were tasked with creating a more empathetic digital experience. The team from the University of Washington (UW), led by Assistant Professor Tim Althoff, assistant professor at the Paul G. Allen School of Computer Science of Engineering and director of the Behavioral Data Science Group, collected data on peer-to-peer conversations about mental health on various platforms. They found empathy does not increase over time, however it can decrease if you don’t have a mental health professional.
This study examined how AI could collaborate with humans to create empathy in online peer-to-peer support conversations. One use for this would be providing feedback and training to human supporters, which could make them more effective. There are also potential uses in machine-in-the loop writing systems that could help express higher levels of empathy.
Computational methods of training empathy could scale a mental health platform to meet the needs and improve outcomes for people who are seeking support. Online peer-supporters would be able to respond more empathetically through computational methods.
In this project, the team at UW made use of natural language processing models to identify and improve empathy in peer support conversations. They designed interactive tools for providing model-based feedback to peer supporters, including a learning agent that can make edits to text in real-time to increase empathy.
This research has led to a better understanding of the role of empathy. Consequently, a TalkLife survey showed that by providing peer support groups to its clients, they had an increase in expressed empathy. This is because those who used the app had what is called “cognitive empathy” increase by 20% and “affective empathy” increase by 39%, which reflects an improvement in their ability to relate to other people.
People who received AI feedback were able to go from feeling incapable to confident and empowered. They were also able to adopt the feedback, while still maintaining a healthy autonomy.
Empathy is critical to being successful in conversations. It improves health and leads to greater alliance and rapport. This can be seen online, as empathic conversations receive significantly more “hearts” and “likes” than non-empathetic ones, are 80% more likely to result in the formation of relationships, and can help improve mental health platforms.
The research is designed to have a large impact on the community. The authors will receive a WebConf Best Paper Award in 2021 for their work because of the algorithmic approach, experimental approach, and amount of value it will provide to the community.