"If You Can't Dance Your Program, You Can't Write It": Challenges and Implications for AI in Education

Synthesis: Videla, Penny, and Ross (2026) offer a critical theoretical analysis of the body's role in AI within education, challenging the dominance of symbolic and disembodied AI models based on abstract information processing. Arguing from post-cognitivist (4E/SEEED) frameworks, they advocate a paradigm shift toward embodied AI grounded in situationality, emergence, and sensorimotor coupling, and diagnose GenAI's cognitive and experiential limits — loss of proprioception and Multimodal AI agency, and the devaluation of embodied making practices. The paper proposes redefining Human–Computer Interaction as a perceptual–affective choreography and closes with seven embodied design principles aimed at a more inclusive, emancipatory, critical, and situated learning. It connects to Embodied Learning, Critical Thinking, Human AI Collaboration, and Learning Theories.

The Ontological Divide Between Embodied and Representational Cognition

The paper's central claim is that human pedagogy is rooted in shared, embodied experience — a "performative idiom" — that is fundamentally at odds with the "representational idiom" of digital computing and AI, which presumes a separation between materiality and information. The authors argue there is an ontological divide between embodied, enactive, situated, materially engaged cognition and cognitivist/computationalist conceptions that treat knowledge as symbolic manipulation. This divide has direct bearing on how AI should be deployed pedagogically. Drawing on Learning Theories and post-cognitivist frameworks (embodied, enactive, embedded, extended "4E" and "SEEED" cognition), they contend that knowledge arises from body–environment interaction rather than from an objective model separated from the agent — a position that directly challenges conventional computing education still rooted in symbolic manipulation.

Why GenAI Is Fundamentally Disembodied

AI lacks a body and material form, accessing only pre-processed symbolic content. Because the most common form is language-based, it privileges semantically expressible knowledge and erases the tacit, embodied knowing an engaged classroom promotes. The authors answer "provisionally in the negative" whether AI can demonstrate experiential knowledge, likening LLMs to "stochastic parrots" that correlate data without understanding meaning. They trace this to AI's symbolic origins at the 1956 Dartmouth Conference and the Von Neumann model, contrasting it with Brooks's behavior-based robotics ("the world is its own best model"). GenAI's passive statistical training produces hallucinations because it lacks real-world validation and the opportunities for action that active agents possess — a limitation of Generative AI with direct consequences for AI in Education.

Sensorimotor Debilities and the Redefinition of HCI

Drawing on Penny's concept of "sensorimotor debilities in digital cultures," the authors argue that GenAI reflects a profound disconnection from embodied experience: loss of multimodal agency, weakened proprioception, kinesthetic awareness, and tactile engagement, and the devaluation of craft and do-it-yourself making. They propose redefining Human–Computer Interaction so the body is not a peripheral interface but the constitutive core of thought. Engaging Seberger's critique of the "unmarked human" and "infrastructure with fangs," they argue for a perceptual–affective choreography in which interaction is lived, embodied, and negotiated in real time — a move toward emancipation rather than normalization and exclusion, with strong implications for Critical Thinking and Higher Education pedagogy.

Creativity, Cognitive Debt, and the Risks of Banking Pedagogy

The paper critiques GenAI's flattening of creativity: AI samples cases within a genre to generate variations ("combinatorial creativity"), whereas substantive creativity challenges the domain's definition, and heavy Large Language Models (LLMs) use tends to produce "regression to the mean" and "accumulation of cognitive debt" that weakens critical thinking and epistemic ownership — echoing Cognitive Offloading and Creativity research. It warns that transmissive, Freirean "banking" pedagogy, reinforced by algorithmic models acting as omniscient oracles, produces control, homogenization, and dependency, and can culminate in "epistemicide" that devalues embodied and ancestral knowledge. These risks are particularly salient for Human AI Collaboration design in education.

Embodied Design Principles

The paper proposes two foundational dimensions — rethinking pedagogical design toward open, dynamic, embodied approaches where AI is a co-creation agent, and building complex digital ecosystems integrating sensors, robotics, and extended reality — operationalized through seven design principles: (i) decenter LLMs as epistemic centers; (ii) use situationality to generate meaningful connections; (iii) support distributed creativity and co-agency; (iv) embrace uncertainty, failure, and improvisation; (v) cultivate critical awareness of bias and algorithmic performativity; (vi) integrate embodied sensorimotor multimodality; and (vii) pursue ecological design for open pedagogical innovation. These are grounded in Design-Based Research and aim to make AI a means for agency, creativity, and situated understanding rather than an end in itself.

What this means for practice

• Instructors. Decenter LLMs as the epistemic center: design activities in which meaning emerges through body–environment interaction, materials, and making rather than through prompt-and-response text production.
• Instructors. Build in room for uncertainty, failure, and improvisation, treating exploration and epistemic risk-taking as legitimate inquiry instead of forcing rigid, deterministic task designs.
• Designers. Integrate sensorimotor multimodality — gesture, voice, movement, touch — and favor open ecosystems over closed tools, since even multimodal platforms such as NAO's Choregraphe still lack flexibility in unpredictable scenarios.
• Researchers. Interrogate algorithmic performativity, training conditions, and embedded bias as part of designing with AI, particularly where outputs risk epistemic exclusion or cultural homogenization.

Limitations

• The article is a critical theoretical analysis, not an empirical study: it reports no participants, intervention, or measured outcomes, and its ontological-divide claim rests on conceptual argument plus cited examples.
• Its seven embodied design principles are proposals; the authors state that implementation requires material, institutional, and formative conditions often lacking in educational settings, and warn that the gap could deepen existing inequities in access and participation.
• The embodiment technologies invoked are immature — the authors note that NAO's programming environment still lacks flexibility in unpredictable scenarios — so the sensors, robotics, and extended-reality ecosystems the framework needs are aspirational rather than available.

Citation

Videla, R., Penny, S., & Ross, W. (2026). "If You Can't Dance Your Program, You Can't Write It": Challenges and Implications for AI in Education. ACM Transactions on Computing Education, 26(3), Article 43.