The Evolution of Research on AI and Education Across Four Decades: Insights from the AIxEd Framework

Synthesis: Rismanchian & Doroudi (2025) present the AI×Ed framework, a two-axis typology for categorizing the diverse relationships between AI and education — the role of AI (applied tool vs. analogy to human intelligence) and the end user (researcher to learner) — yielding four quadrants that capture distinct forms of AIED research. They apply the framework to trace the evolution of the Artificial Intelligence in Education field across four decades by locating papers from the AIED proceedings (1985, 1993, 2021, 2024) and the International Journal of Artificial Intelligence in Education (2004, 2014, 2021). Their central finding is that the field has moved from a diverse mix of work — including substantial early research treating AI as an analogy to human intelligence and learning — toward an almost exclusive focus on instrumental, applied uses of AI, a trajectory that the rise of generative AI and large language models may now help partially reverse. The paper contributes a conceptual framework for historicizing the field and a corpus-level research-methods lens for examining how AIED's research agenda has shifted over time.

Key Findings

The AI×Ed framework. AI×Ed uses two axes to position any AIED research project. The horizontal (end-user) axis spans a spectrum from researchers (who use AI to better understand learning) through practitioners, teachers, and parents to learners (who interact with AI directly, including via AI Literacy and AI education). The vertical (role-of-AI) axis distinguishes AI as an applied tool (pragmatic applications such as Intelligent Tutoring systems, Personalized Learning platforms, and teacher dashboards) from AI as an analogy to human intelligence (using computational models and theoretical concepts to illuminate how people think and learn). Their intersection produces four quadrants, extending Kahn's (1977) original "three interactions" and aligning with frameworks by Porayska-Pomsta and Holmes.

A four-decade historical shift. In the AIED 1985 and 1993 proceedings, the authors find a diverse distribution with substantial work near the bottom half of AI×Ed — research using AI as an analogy to human intelligence, including computational models of learning (e.g., Cascade), educational microworlds rooted in Papert's work, agent-based models of knowledge acquisition, and AI curricula. By the 2000s this strand had largely disappeared: no IJAIED 2004 paper occupied the bottom-left quadrant, signaling the field's turn away from the AI-human-intelligence analogy.

The rise of data-driven, researcher-facing work. From 2014 onward the framework's upper-left quadrant (researchers using AI as an applied tool) grew markedly, reflecting the emergence of educational data mining and learning analytics and a broader shift from rule-based, knowledge-engineering approaches to data-driven techniques — with Bayesian knowledge tracing as a recurring example. The authors suggest this data-driven turn partly explains AIED's move away from AI-as-analogy research.

Generative AI rebalances the field. Roughly 50% of full papers in the AIED 2024 proceedings used LLMs in their tools, methodologies, or evaluations, and — notably — three of the four papers in the bottom half of AI×Ed were LLM-based. This includes best-paper work on LLM-based teachable agents (learning by teaching), studies that evaluate LLMs as cognitive agents using behavioral-science methods previously reserved for humans, and work modeling the zone of proximal development through predictive-model uncertainty.

Human-inspired evaluation and learning from LLMs. The paper argues that LLMs' natural-language flexibility requires human-inspired evaluation approaches (e.g., simulated students, AI teacher tests, behavioral analysis) rather than technical metrics alone, and suggests that studying LLMs — cautiously — can generate insight into human learning, a research direction largely absent from AIED 2024.

Future directions beyond generative AI. The authors propose reviving bottom-half research through renewed AI Literacy work that encourages reflection on learning, applications of machine teaching to human instruction, and agent-based models / complex-systems methodology (e.g., their own computational model of the ICAP framework) that bridge contemporary learning theory and computational modeling.

Methodological transparency. The paper locates each paper in AI×Ed based on author judgment of abstracts and full texts, acknowledges this is not a systematic or scalable data-driven categorization, and makes its full dataset publicly available as supplementary material for replication.

Connected Concepts

• History of AI in Education
• AI in Education
• Theory Development in AI in Education
• Research Methods in AIED

Connected Articles

• Control vs. Agency: Exploring the History of AI in Education — a companion historical framing of AI in education's origins and conceptual tensions
• Simulating Students with Large Language Models: A Review of Architecture, Mechanisms, and Role Modelling in Education with Generative AI — LLM-based simulated students, extending the lineage of computational student models central to AI×Ed's analogy quadrant
• Learning-by-Teaching with ChatGPT: The Effect of a Teachable ChatGPT Agent on Programming Education — LLM-based teachable agents in programming education, the renewed learning-by-teaching strand the paper highlights
• Measuring Cognitive Engagement in Collaborative Discourse with an Extended ICAP Framework: Comparing Human Annotation, In-Context Learning, and Reflective LLM Agents — agent-based modeling of the ICAP framework, exemplifying AIED's complex-systems methodology direction

Citation

Rismanchian, S., & Doroudi, S. (2025). The evolution of research on AI and education across four decades: Insights from the AIxEd framework. International Journal of Artificial Intelligence in Education, 35, 2797–2820.