Robots as mirrors of the self: Investigating Social Cognitive Neuroscience via Social Robots

Abstract

Social interactions are fundamental to human life, traditionally studied through humanhuman interactions. However, the rise of Artificial Intelligence (AI) and robotics, social robots occupy a curious space that is neither human nor object and therefore present new challenges and opportunities: how do we interact with artificial agents that mimic humans in appearance and behavior? Thus far, social robots have fallen short of being effective social partners with interactions often awkward or superficial. A better understanding of how social robots are perceived and perhaps what we are willing to accept as a social agent will help improve robotic design and subsequent interactions with these artificial agents. Three studies herein examine how robots are perceived from a social cognitive and social vision standpoint offering compelling and growing insight into whether there are limits to socialness in the brain.

Chapter 2: Beyond human-likeness: Socialness is more influential when attributing mental states to robots

We sought to replicate and expand previous work showing that the more human-like a robot appears, the more willing people are to attribute mind-like capabilities and socially engage with it. Forty-two participants played games against a human, a humanoid robot, a mechanoid robot, and a computer algorithm while undergoing functional neuroimaging. We confirmed that the more human-like the agent, the more participants attributed a mind to them. However, exploratory analyses revealed that the perceived socialness of an agent appeared to be as, if not more, important for mind attribution. Our findings suggest top-down knowledge cues may be equally or possibly more influential than bottom-up stimulus cues when exploring mind attribution in non-human agents. While further work is now required to test this hypothesis directly, these preliminary findings hold important implications for robotic design and to understand and test the flexibility of human social cognition when people engage with artificial agents.

Chapter 3: Disjoint in children’s explicit report and implicit BOLD signal in mentalizing to robots

This study is the first to use fMRI to examine children’s mentalizing with social robots, highlighting the importance of both implicit and explicit social perception measures. Using the same experimental design as Chapter two, 18 children (and 18 adults) played RPS against a human, a humanoid robot, a mechanoid robot, and a computer algorithm while undergoing functional neuroimaging. While children might rate robots and humans similarly on behavioral measures, their BOLD signal revealed a more mature pattern akin to adults – increasing engagement with increasing socialness of game partners. Adults, however, showed linear consistency between their ratings and BOLD signal in mentalizing regions, whereas children did not. Children’s behavioral ratings were generally higher than adult’s ratings overall and did not discriminate amongst the 4 game players. This discrepancy between children’s explicit ratings and implicit neural responses indicates the need to use both cognitive measures and neuroimaging to fully understand children’s social perceptions.

Chapter 4: Perception of humans, robots, and objects in social interactions

We sought to understand how social interactions between humans and humanoid robots were perceived in the visual and social areas of the brain and, very broadly, whether these humanoid robots were perceived more like humans, objects, or something different entirely. We further wanted to know whether brain areas which putatively identify human bodies would also be sensitive to robot bodies, as humanoid robots have a human-like body shape. In our main experimental task, we found that both humans and robots engaged the social cognition and vision systems, including body areas. Additionally, during social interactions, the composition of the dyad type mattered to the social cognition system, and in some cases, dyads containing robots engaged these areas more than in purely human dyads. We also found that strong social cues, such as overtly social gestures, can override typically non-social scenes (such as dyads facing away or even when one 'partner’ is a static object). However, when social cues are minimised and stimuli are more tightly controlled (as in our “body localizer”), only humans, not robots, engaged “body areas” in the brain. Overall, these three studies provide growing evidence for the importance of perceived socialness in robots during social interactions with humans. Additionally, these studies provide further support the use of social robots in cognitive neuroscience as not only a clever tool to contribute to, and advance, knowledge in the cognitive neuroscience field but the findings have significant implications for robotic design.

Date of Award	22 Aug 2025
Original language	English
Awarding Institution	Bangor University
Sponsors	ERC
Supervisor	Kami Koldewyn (Supervisor)