Attachment to robots and therapeutic efficiency in mental health

Social robots are robots that are “designed to interact with people in human-centric terms and to operate in human environments alongside people” (Breazeal et al., 2016). In their interactions with humans, these systems follow the behavioral norms and expectations that are defining features of social interactions, such as emotional expressiveness, verbal communication, user engagement, and an appealing physical appearance (Scassellati et al., 2012). The roles social robots can play in delivering mental health care interventions for children and older adults (especially those with ASD and dementia, respectively) has been widely studied (for recent reviews see Cifuentes et al., 2020, Marchetti et al., 2022). These investigations have shown that social robots can be effective in engaging users, improving their mental health, mood and cognitive capacities, and enhancing their quality of life. According to a recent review, in mental health settings, social robots are typically used in three major contexts: acting as therapists/coaches, mediators, or assistants (David et al., 2014). In this Perspectives article, our goal is to refine this distinction about the possible roles of social robots and to point out that different roles require different levels of attachment. We will argue that the attachment between the client and the robot is a fundamental ingredient of any helping relationship and that the full potential of using social robots in mental health settings can only be realized if the strength of attachment is appropriately correlated with the type of relationship established.

The paper draws on Bowlby’s attachment theory to frame the therapeutic relationship as a vehicle for moving clients from insecure to secure attachment, with therapists serving as attachment figures who provide a secure base and corrective emotional experiences (Bowlby, 1982; Sherman et al., 2015; Degnan et al., 2016; Fraley and Roisman, 2019). Mallinckrodt (2010) identifies five characteristics of attachment relationships—proximity seeking, safe haven, secure base, separation anxiety, and the “stronger and wiser” attribute—arguing that most of these are evident in therapeutic relationships, supporting the potential for therapy to modify insecure internal working models. An alternative perspective views therapists as “alternative support figures” whose relationships can enhance clients’ worthiness, self-perception, and reflective functioning (Saunders et al., 2011; Fonagy et al., 1996). The article then reviews evidence that humans form attachments to nonhuman agents. Anthropomorphism can transform interactions with objects and robots into human-like social exchanges, fostering object attachment and fulfilling needs for comfort, identity, and efficacy (Epley et al., 2007; Waytz et al., 2010; Wan and Chen, 2021; Norberg and Rucker, 2021). Studies report empathy toward robot pain and reluctance to harm robots, along with neural overlaps between empathy for humans and robots (Rosenthal-von der Pütten et al., 2013; Suzuki et al., 2015; Chin et al., 2023; Darling et al., 2015), and that inducing empathy increases prosocial behavior toward robots (Kühnlenz et al., 2013; Weiss et al., 2010; Seo et al., 2015). Collectively, these findings suggest that emotional bonds with robots can mirror human attachment features, implying that robot-based therapy should enable development of proximity seeking, safe haven, secure base, separation responses, and perceptions of competence in the robot therapist.

• The authors refine existing role taxonomies for social robots in mental health and identify six distinct roles: (1) diagnostic tools, (2) robot-mediated interviewers, (3) promoters of social connections (social mediators), (4) coaches, (5) social companions, and (6) therapists.
• They order these roles by the required level of client attachment, from lowest to highest: diagnostic tools → robot-mediated interviews → promoting social connections → coaches → social companions → therapists.
• Evidence across domains indicates social robots can produce clinically relevant engagement and outcomes: robots can elicit diagnostically valuable social responses in ASD (Diehl et al., 2012) and help discriminate social anxiety (Rasouli et al., 2022); robot-mediated interviews yield similar engagement and content as human interviews, and can increase disclosure (e.g., more reports of bullying; Bethel et al., 2016; Wood et al., 2013a,b); as mediators, robots facilitate conversation, engagement, collaboration, and participation (Adikari et al., 2023; Dautenhahn, 2003; Feil-Seifer and Matarić, 2011; Costa et al., 2010); as coaches, robots can sustain engagement and foster working alliance (e.g., Autom; Kidd and Breazeal, 2008) and produce statistically significant improvements in well-being and readiness to change (e.g., Jibo positive psychology coach; Jeong et al., 2023).
• Therapeutic applications (robot as therapist) are argued to require high attachment, enabling safe haven and secure base functions, and promoting reflective functioning and mentalization—surpassing the attachment needs of roles focused on behavior change alone (e.g., coaching) or companionship.
• The paper posits that to realize therapeutic potential, robot design should foster appropriate levels of attachment aligned to role, and anticipates that neural responses to robots will vary with attachment level across roles.

The article addresses the central question of how attachment should be calibrated in human–robot helping relationships by proposing a structured mapping between robot roles and the requisite strength of client attachment. This mapping clarifies why some robot roles (e.g., diagnostics, interviews) can function effectively with minimal attachment, while others (e.g., therapy) demand robust attachment to enable secure base/safe haven functions and corrective emotional experiences analogous to those in human-led therapy. By integrating attachment theory with empirical findings on anthropomorphism and empathy toward robots, the paper argues that therapeutic efficacy hinges on designing interactions that elicit and maintain the appropriate attachment features (proximity seeking, safe haven, secure base, etc.). This framework has practical significance for robot design and deployment: calibrating attachment affordances to the intended role can maximize engagement, improve outcomes (e.g., behavior change in coaching, social participation in mediation), and potentially narrow the mental health treatment gap. The authors also highlight future tests of this attachment-role alignment, including hypothesized differences in clients’ neural responses across low- vs. high-attachment robot roles and the question of whether secure attachment developed with a robot therapist generalizes to human relationships as in successful human-led psychotherapy.

Social robots could be used in the full spectrum of mental health care, they could decrease the so-called “treatment gap” (the burden implied by the lack of human mental health professionals), and they could also increase the quality of treatments. However, for their effectiveness, their capacity to trigger attachment feelings and behaviors in patients will need to be improved and carefully fine-tuned in line with the requirements of the specific roles the robots would play.

This Perspective presents a conceptual framework rather than new empirical data. It acknowledges key open questions: whether the attachment features elicited by social robots suffice to realize the full potential of psychotherapy; whether secure attachment formed with a robot therapist generalizes to other human relationships as observed in successful human-led therapy; and how clients’ neurological responses to robots differ across roles requiring low vs. high attachment. The paper calls for integrative research across social neuroscience, computer science, and robotics to test these claims, and for careful design to foster appropriate attachment without overreliance on unvalidated assumptions about anthropomorphism and bond formation.