Therapy chatbots and emotional complexity: do therapy chatbots really empathise?

Abstract

Internet-based cognitive behavioural therapy (iCBT) and, more specific, therapy chatbots are at the heart of this article. We review studies about therapy chatbots in terms of risks, benefits, and 'empathetic design'. Research shows that these chatbots can relieve symptoms of stress, depression, and anxiety, by tracking emotions over time, encouraging self-reflection, and helping to gain insight into cognitive biases, among other things. However, several studies focus on the short term, have limited samples, and include self-report. While there are certainly benefits, risks such as privacy, data security, and a lack of human oversight, cannot be ignored. The increasing use of LLMs like ChatGPT as 'AI therapist' intensifies these concerns. The need for social connection might increase the tendency to anthropomorphise, and users risk becoming emotionally dependent on them. Developers claim that therapy chatbots such as Youper are 'empathetic' and recent studies indeed indicate that informants prefer written AI responses over human therapists' written responses in terms of empathy. Although therapy chatbots and LLMs lack affective and motivational empathy, they can express elements of cognitive empathy. Nonetheless, we contend that we should avoid attributing human characteristics such as 'empathy' to AI, because the risks of manipulation and dependency are too high.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 13 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.