I’ve just posted a new pre-print, A Procedure for Classifying Attachments and Affective Social Bonds in Human-Robot Dyads, which is currently under review.

Human-robot interaction (HRI) talks a lot about “attachment” and “bonding” with robots. The terms are borrowed from developmental psychology, comparative ethology, and neuroendocrinology, but the criteria that make them meaningful in those fields have rarely come along with them. As a result, it is often hard to tell what a given study is actually claiming, or what it has actually shown. Essentially, we are now left with a “wild west” of claims and methods around human—robot relationships, disconnected from the root disciplines and, more importantly, from each other.

The paper proposes a short reporting standard to fix this. It allows researchers to ask four clear questions of a human—robot relationship, each grounded in the disciplines from which these terms were borrowed. We then apply the procedure to a set of published human-robot and human-agent studies, to show where claims of attachment or bonding are well supported and where they are, at best, untested.

Along the way, we argue for the more careful term affective social bonding over bare “social bonding”, and the paper offers a definitional synthesis and a set of contributions for anyone designing or evaluating long-term human-robot relationships.

This builds on my recent workshop paper on bonding, trust, and the evaluation of human-robot relationships, and is part of a larger series of work on how we measure and talk about relationships with artificial systems.

As it is still under review, I’d very much welcome comments and critique, so please get in touch if this is your area.