The growing use of AI chatbots for emotionally significant purposes raises questions about how dependence on these systems may transform us as affective beings.

This talk addresses one such concern: that some individuals may become too emotionally reliant on chatbots. Drawing on the situated affectivity literature, I adopt a descriptive account of chatbots as affective scaffolds, then develop a normative account of emotional overreliance that distinguishes it from benign or beneficial forms of deep affective dependence.

Emotional overreliance, I argue, is a pattern of affective scaffolding that is self-undermining: the reliance itself erodes the agent’s capacity to revise their scaffolding relations where their situation calls for it, and it can do so even while the reliance continues to feel beneficial. Because human life predictably calls for such revision, an agent whose capacity to enact it is compromised is exposed to foreseeable harm.

I then apply this account to AI chatbots, focusing on psychotherapy chatbots. I argue that core features commonly marketed as advantages of these products are also features that promote the pattern of overreliance, such that current chatbots pose a particularly serious risk of this pattern by design.

Millions of people are using chatbots for mental health support, some informally, and some as a component of formal mental healthcare.

The promise is that chatbots might help reach people who lack adequate access to formal mental healthcare. Chatbots are scalable, affordable, available on demand from anywhere with an internet connection, in multiple languages, without procedural red tape, and with minimal perceived stigma.

There are also significant concerns attached to this technology. One question we are just beginning to understand is what the evidence says about the safety and effectiveness of these tools.

Another under-explored question is how this may differ across communities. Given what we know about algorithmic bias in other applications, can chatbots can be trusted to serve the marginalized communities most in need of access to mental healthcare?

‘Contemporary Debates in Philosophy of Artificial Intelligence’

  • McGill University

  • May 1st, 2026

  • Free, no registration required