Reconceptualizing learning sciences in the era of artificial intelligence: Toward hybrid intelligence and adaptive learning systems

Synthesis: Malik and Bashir (2026) ran a PRISMA 2020-guided systematic review of 32 studies (38 records retrieved across seven databases, Jan–Mar 2026; Braun and Clarke six-phase thematic analysis; 53 initial codes → four themes) asking why Learning Sciences and AI in education have developed as separate literatures. Their answer has four parts: (1) Learning Sciences is epistemologically fragmented — constructivist, sociocultural, cognitive, and design-based perspectives are complementary but no unified theoretical framework exists (Fischer et al. 2018); (2) theory–technology contradictions — computational systems rely on prediction, optimization, and measurable learner behaviors, while educational theories emphasize context, social interaction, and meaning-making; (3) the datafication of learning — behavioral traces, assessment records, and predictive analytics increasingly stand in for learning itself; and (4) Hybrid Intelligence as an emerging paradigm in which learners, educators, AI systems, and learning analytics function as complementary components of adaptive learning ecosystems. The contribution is a Hybrid Intelligence Framework integrating all four domains — but it is explicitly conceptual: the authors state the synthesis "does not empirically validate the proposed framework," the intersection-based search retrieved only 38 records (a deliberate specificity-over-sensitivity tradeoff), and future work must test the framework with longitudinal, mixed-methods, and experimental studies. Tool vintage: literature 2004–March 2026.

Summary

Learning Sciences (LS) emerged as an interdisciplinary field integrating psychology, cognitive science, education, neuroscience, sociology, and computer science (Kolodner 2004; Nathan & Alibali 2010; Sawyer 2022), conceptualizing learning as active, situated, socially mediated, and developmental rather than knowledge transmission. It draws on constructivist theories (active knowledge construction), sociocultural theories (learning through participation in communities), and design-based research (iterative development in authentic environments), with cognitive tutoring research demonstrating how learning-science principles can inform technology-enhanced instructional design (Koedinger & Corbett 2006). Yet the field "has not yet achieved a fully unified theoretical framework capable of integrating its diverse paradigms into a single explanatory model." Meanwhile AI in education has expanded through intelligent tutoring systems, adaptive platforms, learning analytics, recommender systems, machine learning, and large language models (Almasri 2024; Gligorea et al. 2023; Luckin & Cukurova 2019; Zhai et al. 2021) — technologies enabling adaptive content delivery, automated feedback, predictive modeling, and intelligent decision support. Learning analytics and educational data mining interpret learner data to improve decision-making (Minn 2022; Salas-Pilco et al. 2022), and recent work extends into predictive analytics, multimodal analytics, affective computing, and educational planning (Järvelä et al. 2020; McNichols et al. 2026; Van Audenhove et al. 2024). The review's premise is that these streams "have largely progressed across partially separate research streams, resulting in fragmented theoretical perspectives and limited conceptual integration" — a theory–technology gap in which AI development outpaces coherent pedagogical foundations.

The synthesis positions Hybrid Intelligence as the integrative concept. Within the Learning Sciences, this perspective extends theories of self-regulated learning and socially shared regulation by considering how AI systems may function as co-regulatory partners supporting reflection, adaptive Scaffolding, collaboration, metacognitive processes, and distributed cognition (Järvelä et al. 2020, 2023). Hybrid intelligence "emphasises the complementary integration of human expertise and computational capabilities to support learning, problem-solving, decision-making, and adaptive educational support" — informed by collective hybrid intelligence, which considers the architectural, workflow, and interactional dimensions of human–AI integration (Moradi et al. 2019). The framework's arc: Learning Sciences establish the theoretical foundations guiding AI for education; AI technologies generate learner data interpreted through Learning Analytics to support Adaptive Learning Systems; these converge in Hybrid Intelligence, where human expertise and AI collaboratively support ethical, explainable, and evidence-informed decision-making.

Key Findings

• Theme 1 — Epistemological fragmentation of Learning Sciences. Across the 32 studies, Learning Sciences was consistently described as an interdisciplinary field informed by psychology, cognitive science, education, sociology, neuroscience, and computer science, with constructivist, sociocultural, cognitive, and design-based approaches presented as "complementary but conceptually diverse, with limited evidence of a single unified theoretical framework." The review's implication for AI design: "translating diverse pedagogical theories into computational models remains theoretically complex."
• Theme 2 — Theory–technology contradictions. Research on intelligent tutoring systems, adaptive learning, and AI-based applications demonstrated that computational systems primarily relied on prediction, optimization, and measurable learner behaviors, while the Learning Sciences literature described learning as socially situated, context-dependent, and interaction-based. The synthesis indicates these differences "may reflect contrasting assumptions about how learning is conceptualized, represented, and evaluated," highlighting the need for stronger theoretical integration between educational psychology, Learning Sciences, and AI system design. (In the coding framework this theme was supported most strongly by codes including theory–technology alignment, machine learning, deep learning, and intelligent tutoring systems.)
• Theme 3 — Datafication of learning. Learning analytics, educational data mining, and adaptive systems increasingly represent learning through digital data — learner interactions, behavioral traces, assessment records, and predictive analytics — frequently reported as supporting personalized learning, adaptive feedback, and educational decision-making. But the synthesis warns that "reliance on measurable behavioural indicators may not fully capture the cognitive, social, and contextual dimensions of learning," and calls for combining data-driven practice with broader Learning Sciences perspectives.
• Theme 4 — Hybrid Intelligence as an emerging educational paradigm. Hybrid intelligence recurred across the reviewed studies (Cukurova 2025; Gibson et al. 2023; Järvelä et al. 2023) as collaboration between human learners, educators, and AI systems. The present review extends prior conceptualizations (which treated it primarily as a technological capability) by "conceptualising Hybrid Intelligence as an educational framework in which learners, educators, AI systems, and learning analytics function as complementary components of adaptive learning ecosystems." Recurring implementation considerations: learner autonomy, human agency, ethical responsibility, transparency, and algorithmic authority.
• Methodology: a deliberately narrow intersection search. The four-block Boolean strategy (Learning Sciences AND AI-in-education AND adaptive/intelligent technologies AND learning-regulation constructs including self-regulated learning, metacognition, and hybrid intelligence) was designed for "conceptual relevance and specificity rather than maximum retrieval sensitivity." Retrieval: Scopus 10, Web of Science 7, ScienceDirect 6, SpringerLink 4, Wiley 5, ACM 2, Google Scholar 4 = 38 records → 32 included studies. The authors repeatedly caution that the small retrieval volume reflects search specificity, not the literature's size — "a methodological limitation because broader search terms or less restrictive combinations may have identified additional relevant studies."
• The framework is conceptual, and its authors say so. Quality assessment was qualitative (peer-review status, relevance to LS and AIEd, methodological transparency, theoretical contribution, educational relevance) rather than a numerical score; all 32 studies met criteria. The conclusion is explicit about the boundary: "the synthesis is conceptual in nature and does not empirically validate the proposed Hybrid Intelligence Framework."

What this means for practice

• Ground AI-supported systems in learning theory before adding capability. The review's core practical message: "the development of AI-supported education requires continued integration of technological capabilities with cognitive, sociocultural, and pedagogical perspectives," and "future AI-supported educational systems should be grounded in established principles of Learning Sciences while ensuring that technological innovation remains aligned with pedagogical objectives, human agency, transparency, and ethical responsibility."
• Audit what your system treats as "learning." Theme 3 is directly actionable for anyone building or procuring adaptive systems: if the measurable signals are interactions, behavioral traces, and assessment records, ask what dimensions of learning the system cannot see — the cognitive, social, and contextual ones the review flags — and pair analytics with professional interpretation rather than automated decision-making alone.
• Design for complementarity, not substitution. Hybrid intelligence's design principle is that human expertise and AI are deliberately combined: AI supports instructional guidance, learning regulation, feedback generation, planning, and monitoring, while humans retain the agency, ethical responsibility, and override capacity the review repeatedly names. The co-regulatory framing (AI as a partner in self-regulation rather than a replacement for it) is the concrete design pattern.
• Explainability and ethics are framework requirements, not add-ons. The proposed framework explicitly connects epistemological foundations, AI technologies, learning analytics, and hybrid intelligence "to support evidence-informed, explainable, and human-centred educational decision-making" — future research directions named by the authors include ethical governance, explainable AI, and human–AI collaboration.

Limitations

• Conceptual synthesis only; no empirical validation. The authors state plainly that the framework awaits empirical examination "across diverse educational contexts and learner populations" via longitudinal, mixed-methods, and experimental studies.
• Small, search-limited corpus. 32 studies from a deliberately intersection-based search; the authors warn the retrieval volume reflects the strategy's specificity and that broader searches "may identify further relevant studies and perspectives."
• No primary data, no effect estimates. The review synthesizes published literature; thematic frequencies (e.g., Learning Sciences code frequency 18, Human–AI Collaboration 15, Constructivism 14) describe the reviewed corpus, not the field.
• Two-author review with manual screening and consensus coding; while Braun and Clarke's reflexive approach and an evidence matrix (Table 6) provide an audit trail, single-team coding carries the usual interpretive risks.
• The included studies skew conceptual. The evidence matrix is dominated by conceptual papers, reviews, and foundational works (Kolodner, Sawyer, Luckin & Cukurova, Cukurova, Gligorea, Zhai) rather than classroom experiments — appropriate for a framework paper, but it means the framework's claims about learning outcomes are untested.

Citation

Malik, A., & Bashir, I. (2026). Reconceptualizing learning sciences in the era of artificial intelligence: Toward hybrid intelligence and adaptive learning systems. Acta Psychologica, 270, 107977.