Summary: A pioneering international study in developmental psychology and artificial intelligence has revealed that while children as young as 3 years old can instinctively read intentions and personal preferences in a human’s eyes, they fail to recognize this same nonverbal communication in the gaze of a humanoid robot.
By demonstrating that a robot simply imitating a human signal like gaze is insufficient to establish a true communicative bond, the findings redefine the engineering standards required for embodied AI and child-robot interaction.
Key Facts
- The Developmental Eye Test: Coordinated by Professor Antonella Marchetti, researchers evaluated Italian children aged 3 to 5 years old to explore how human and robotic gazes evoke different structural impressions in a child’s mind. The test required children to observe either a person or a humanoid robot looking at a specific object to assess if they could deduce which item was “preferred” by the agent.
- Decoding Intent vs. Mechanical Staring: The results unmasked a distinct cognitive split. Children consistently interpreted the human gaze as a meaningful, intentional signal, assuming that if a person looks at an object, they must like it. However, when a humanoid robot stared at the exact same object, the gaze was not enough for the children to attribute a true psychological preference or desire to the machine.
- The Sovereignty of Child Preference: The study highlighted that while a gaze, whether human or robotic—helps a child decode what another agent likes, it does not exert a transformative effect on the child’s own personal preferences or choices.
- The Failure of Isolated Mimicry: Professor Marchetti noted that simply programming a robotic artifact to imitate an isolated human signal like eye movement is not enough to make it communicative to a child. To create a functional connection, intelligent technologies designed for children must incorporate richer, developmentally appropriate interactions consisting of words, physical gestures, reciprocity, context, and shared presence.
- The Mandate for Embodied AI: In the broader debate on artificial intelligence, the study emphasizes that communication is not merely about verbal output or text-based responses. To help children attribute true mental states, such as intentions and beliefs, to technology, AI must be integrated into physical, interactive systems, a dimension known as embodied AI.
- The ROBIN Project Launch: These findings carry massive clinical implications for supporting children on the autism spectrum, where shared attention and gaze interpretation are vulnerable dimensions of development. To apply this data, the Don Carlo Gnocchi Foundation and Università Cattolica will launch the ROBIN project (ROBot-based Neuropsychomotor INtervention) in June 2026, using humanoid robots to promote imitation skills and socio-communicative rehabilitation.
Source: Universita Cattolica del Sacro Cuore
Very young children (even as young as 3 years old) can read intention and preferences in the eyes of a person, but they do not recognize this type of nonverbal communication in the gaze of a humanoid robot.
This is the finding of a study published in the International Journal of Child-Computer Interaction, coordinated by Antonella Marchetti, Director of the Department of Psychology of Università Cattolica and CERITOM (Research Center on Theory of Mind and Social Competences Across the Lifespan), in collaboration with scholars from Tokyo and Osaka, and colleagues Davide Massaro, Cinzia Di Dio, and Federico Manzi of the Università Cattolica of Milan.
THE STUDY
The research involved Italian children aged 3 to 5 years old to explore how people and robots gaze can evoke different impressions in children’s minds.
The test consisted of showing to children a person and a humanoid robot while looking at an object, assessing whether they could understand which object was “preferred” by the agent looking at it.
The results show that children interpret the human gaze as a meaningful signal: if an individual looks at an object, children tend to assume that the person likes that object. The same does not happen, however, when a humanoid robot is looking at the object. In that case,
the gaze is not enough for children to attribute a true preference to the robot.
In short, children use the human gaze to “read” desires and intentions, while they struggle to do the same with the robot. Furthermore, gaze—human or robotic—does not seem to change children’s personal preferences: it helps them understand what the other person likes, but it does not necessarily change their own preferences.
Professor Marchetti explains, “This does not mean robots cannot play an educational or social role. However, it suggests that simply imitating a single human signal, such as gaze, in a robotic artifact is not enough to make it truly communicative in a child’s eyes.
“Designing robots and intelligent technologies for children requires richer, more natural, and developmentally appropriate interactions: made up of words, gestures, reciprocity, context, and shared presence. This is reinforced by the fact that even human interactions alone are not sufficient to exert clear transformative effects on children’s preferences. These data are particularly relevant in the debate on artificial intelligence,” she continues.
“Many AIs today speak, respond, and make suggestions, but our results highlight that, especially for children, communication is not just about words: presence and shared context also matter. From this perspective, an AI integrated into physical systems — so-called embodied AI, one of the most complete expressions of which is humanoid social robotics — represents a crucial dimension for understanding how children attribute mental states (e.g., intentions, beliefs, preferences) to technologies as well,” she adds.
These findings also have significant implications for applications, particularly in the field of autistic spectrum disorder, where gaze and shared attention represent crucial psychological dimensions of socio-communicative development and can be particularly vulnerable.
In this context, humanoid robots are increasingly being studied as support tools for rehabilitation interventions focused on these skills. Understanding how a child interprets a robot’s gaze as an intentional signal can therefore help design more targeted, natural, and developmentally sensitive interventions.
The ROBIN (ROBot-based Neuropsychomotor INtervention to promote imitation skills in young children with autism spectrum disorder) project, funded by the Ministry of Health as part of the Finalized Research program, which will begin in June 2026, is also part of this research direction, the professor anticipates.
The project is led by the Don Carlo Gnocchi Foundation and the CeRiToM of the Università Cattolica del Sacro Cuore, which is involved as a research group on the role of gaze and psychological processes in these forms of intervention. The project involves interventions with a humanoid robot to promote imitation skills, which also involve understanding the robot’s gaze and its communicative meaning.
Key Questions Answered:
A: Because the child’s brain naturally looks for a mind behind the eyes, not just a physical signal. Children recognize the human gaze as an intentional, living projection of thoughts and preferences. When they see a humanoid robot stare at an object, they treat it as an uncommunicative, mechanical movement, realizing that an isolated action like a robotic glance isn’t enough to prove the machine actually “wants” or prefers something.
A: Not at all; it just means tech developers have been designing them far too simply. Robots can play a massive educational and supportive role, but engineers cannot just copy a single human signal like eye-tracking and expect a child to connect with it. For an AI or robot to be truly communicative, it must engage in a rich, natural interaction that includes spoken words, physical gestures, shared context, and mutual reciprocity.
A: By building smarter, more natural robot-assisted therapies. Children on the autism spectrum often navigate vulnerabilities when developing shared attention and reading gaze cues. Because the ROBIN project aims to use humanoid robots to improve imitation and communication skills, understanding exactly how a child interprets, or struggles to interpret, a robot’s eyes allows clinicians to design targeted, highly sensitive rehabilitation exercises that feel safe and natural.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- Journal paper reviewed in full.
- Additional context added by our staff.
About this robotics and neurodevelopment research news
Author: Nicola Cerbino
Source: Universita Cattolica del Sacro Cuore
Contact: Nicola Cerbino – Universita Cattolica del Sacro Cuore
Image: The image is credited to Neuroscience News
Original Research: Closed access.
“Preschoolers attribute preferences in response to human but not robot gaze” by Federico Manzi, Mitsuhiko Ishikawa, Cinzia Di Dio, Shoji Itakura, Takayuki Kanda, Hiroshi Ishiguro, Davide Massaro, and Antonella Marchetti. International Journal of Child-Computer Interaction
DOI:10.1016/j.ijcci.2026.100822
Abstract
Preschoolers attribute preferences in response to human but not robot gaze
With technological advancements, children increasingly interact with robots designed to mimic human-like behaviors for communication, among which gaze is particularly pivotal from early childhood. This study thus explores how children attribute and form preferences when exposed to human versus robotic gazes.
The research involved 58 Italian children aged 3 to 5 years. They watched videos featuring a human and a robot each gazing at one of two objects. Subsequently, children were asked which object the gazer preferred (preference attribution) and to indicate their own preference (preference formation). Attribution of object preference was evaluated also as a function of children’s Theory of Mind (i.e., false belief) and mental state attributions to human and robot agents.
Results showed that children consistently attributed preferences based on human gaze, but not robot gaze, suggesting that they interpret human gaze as a meaningful communicative signal, likely associated with intentionality. Gaze had no significant effect on children’s own preferences for either agent. Importantly, attribution of mental states to the human, but not to the robot, significantly predicted accurate preference attribution.
No associations were found between performance on the false-belief task and gaze-based responses, indicating that explicit preference attribution may rely on socio-cognitive processes distinct from belief-based reasoning.
These findings provide design-relevant insights for child–robot interaction, suggesting that gaze alone may not function as an effective communicative cue for young children and highlighting the importance of developmentally informed interaction strategies in robotic systems designed for early childhood.
