Bodily Movement Lessons Sharpen Preschoolers’ Inhibitory Control in AI Ethics Task

A two-week program of embodied learning activities—physical, movement-based lessons built around artificial intelligence ethics—significantly improved five-year-old children’s performance on a Stroop task measuring inhibitory control, according to a randomized controlled study published on 1 September 2026 in the open-access journal AI, Brain and Child. The intervention, tested on 60 senior kindergarteners in China, reduced the children’s Stroop interference effect by more than 120 milliseconds, while a control group receiving conventional storytelling showed no statistically significant change. The researchers argue that this cognitive gain offers indirect evidence that embodied learning can support young children’s capacity for ethical decision-making in interactions with AI systems.

While AI education for children has largely focused on instrumental skills—teaching how to use technology—the researchers note that empirical work specifically addressing AI ethical literacy in young children remains scarce. Most existing research on technological ethics has concentrated on primary, secondary, and higher education cohorts, and traditional methods such as questionnaires, text-based reading, and moral dilemma interviews are unsuitable for children under six.

The rationale for the study rests on a specific cognitive argument. Inhibitory control, a core component of executive function, is the capacity to suppress automatic or prepotent responses in the face of internal impulses or external distractors. When young children’s intuitive impulses conflict with societal moral norms, the authors write, they rely heavily on this mechanism as a kind of braking function. Developmental research has shown that children’s capacity for inhibitory control influences their early internalization of rules. The team therefore reasoned that Stroop task performance—though it directly measures inhibitory control rather than ethical decision-making behavior itself—can serve as a cognitive indicator of young children’s AI ethical decision-making capability, drawing on theoretical, neuroimaging, and behavioral evidence linking conflict resolution in the Stroop task to moral decision-making processes.

To test this, the researchers recruited 60 senior kindergarteners from the Kindergarten Affiliated to Anhui Normal University, all typically developing with no reported history of psychiatric disorders. An a priori power analysis conducted before data collection indicated a minimum required sample of 59 participants. The children were randomly assigned using a random number generator to either an experimental group of 30 children with a mean age of 5.70 years or a control group of 30 children with a mean age of 5.66 years. The groups were well matched at baseline: analyses showed no significant differences in gender composition, age, or scores on the Raven’s Colored Progressive Matrices, an intelligence test administered to control for baseline cognitive ability. All 60 children completed both the pre-test and post-test Stroop tasks.

The experimental group received a two-week embodied learning intervention focused on AI ethics, consisting of 10 sessions of 15 to 20 minutes per day, designed in alignment with China’s Guidelines for the Learning and Development of Children Aged 3–6. The program translated abstract ethical value judgments—fairness, privacy, safety, and responsibility—into bodily-spatial actions. Children used opening and closing postures of their limbs to represent their ethical attitudes toward different AI behaviors, engaged in role-playing to simulate AI interaction contexts, and interacted with AI dolls. Teachers described simple social situations in short sentences, creating meaningful connections between AI ethical knowledge from real-life contexts and children’s movements. The design drew on the two-dimensional classification framework of embodied learning proposed by Skulmowski and Rey, which holds that the effectiveness of embodied learning depends on whether there is a meaningful functional connection between bodily activity and the learning task, rather than on bodily movement per se. The control group, by contrast, engaged in conventional storytelling that included verbal instruction on AI ethical concepts.

Measurement relied on a modified Stroop task adapted from previous research paradigms and administered with E-Prime software in a familiar activity room within the kindergarten. The task comprised three blocks: a neutral block with solid green and red circles, an affective block with smiling and crying faces, and an AI ethical block with pictures of friendly and unfriendly AI. In congruent conditions, the semantic content matched the background color; in incongruent conditions, it conflicted, requiring children to ignore the meaning of the image and respond rapidly to the background color by pressing a keyboard key. Each block contained three runs of 20 trials. Before the formal task, participants received instructions and practice examples to ensure comprehension of the rules.

The baseline data confirmed that the adapted paradigm worked as intended for this age group. Across all 60 children at pre-test, mean reaction times in the incongruent condition reached 1245.65 milliseconds, significantly longer than the 985.42 milliseconds observed in the congruent condition, while mean accuracy fell to 76.85 percent in the incongruent condition compared with 92.45 percent in the congruent condition. Even under the most demanding condition, accuracy remained significantly above the 50 percent chance level, ruling out a floor effect and indicating that the task difficulty was appropriate for senior kindergarteners. Together, these results confirmed the classic Stroop interference effect and provided the baseline against which the intervention’s effects were assessed.

The intervention results were striking. The Stroop interference effect for AI ethical stimuli in the experimental group fell from 265.40 milliseconds at pre-test to 142.15 milliseconds at post-test, a reduction of 123.25 milliseconds that was statistically significant. The control group showed only a decrease from 258.90 to 235.60 milliseconds, a reduction of 23.30 milliseconds that did not reach statistical significance. A mixed-design analysis of variance yielded a highly significant time-by-group interaction, along with significant main effects of both time and group. Simple main effects analysis confirmed that the two groups did not differ significantly at pre-test, but the experimental group showed significantly lower Stroop interference than the control group at post-test. The authors interpret this pattern as evidence of intervention specificity: the improvement in the experimental group is more likely attributable to the embodied learning activities than to natural maturation or practice effects from repeated testing, since a maturation or practice account would predict comparable improvement in the control group.

The researchers offer a theoretical interpretation grounded in embodied cognition theory, which posits that higher-order cognitive processes are grounded in the body’s sensorimotor system and its interactions with the physical environment. During the intervention, children transformed abstract ethical principles into perceivable bodily-spatial actions, and the authors suggest that this training enabled them to draw upon sensorimotor representations when facing cognitive conflicts in the post-test Stroop task, rather than relying solely on abstract logical reasoning. They also note that the age of five to six represents a period when children transition from intuitive action thinking toward concrete figurative thinking, making purely verbal instruction about abstract concepts such as AI ethics particularly ineffective at this developmental stage.

The authors are careful to acknowledge the study’s limitations. The sample of 60 children was drawn from a single region and limited to senior kindergarten classes, constraining both statistical power and external validity. The study relied on reaction time and accuracy measures rather than directly observing neural mechanisms, and a gap remains between Stroop task performance and actual AI ethical decision-making behavior. The intervention lasted only two weeks with an immediate post-test and no delayed follow-up, so whether the improvement persists over time remains unknown. The authors suggest that future research incorporate neurophysiological techniques such as functional near-infrared spectroscopy or electroencephalography, develop more ecologically valid tasks that approximate real-world AI interaction contexts, and adopt longitudinal follow-up designs to evaluate the sustainability of the intervention effects. Despite these constraints, the team argues that the findings provide preliminary empirical evidence for introducing an embodied cognition perspective into early childhood AI literacy education, and they suggest that kindergartens consider incorporating embodied AI ethics activities into regular teaching schedules at a basic frequency of three to four sessions per week, each lasting 15 to 20 minutes.

Subject of Research: A randomized intervention study testing whether embodied learning activities improve inhibitory control related to AI ethical decision-making in kindergarten children.

Article Title: The impact of embodied learning activities on young children’s AI ethical decision-making: a study based on the stroop task

Article References: Sun, T., Xia, J., & Ouyang, Y. (2026). The impact of embodied learning activities on young children’s AI ethical decision-making: a study based on the stroop task. AI, Brain and Child, 2(1), Article 18. https://doi.org/10.1007/s44436-026-00042-6

Image Credits: AI Generated

DOI: 10.1007/s44436-026-00042-6

Keywords: embodied learning, inhibitory control, Stroop task, AI ethics, early childhood, kindergarten, executive function, AI literacy, randomized controlled trial, embodied cognition, moral development, intervention study