Generativism: Toward a Learning Theory for the Age of Generative Artificial Intelligence

Technical Summary: Generativism: Toward a Learning Theory for the Age of Generative Artificial Intelligence

Problem Statement
The proliferation of generative artificial intelligence (AI) in educational settings exposes significant conceptual limitations in the four dominant learning theories: behaviorism, cognitivism, constructivism, and connectivism. These frameworks were formulated prior to the emergence of AI systems capable of generating, synthesizing, and reasoning about knowledge. Consequently, they struggle to address critical questions regarding learning in the AI age: How is learning defined when cognitive processes are distributed between humans and machines? What occurs when learners produce sophisticated outputs before fully understanding underlying concepts? And when does the offloading of cognitive tasks to AI support versus undermine learning? The authors argue that existing theories fail to account for the qualitative shift from technology as a repository of information to technology as a generative partner in cognition.

Methodology
This article employs a theoretical synthesis and critical analysis rather than empirical data collection. The methodology involves:

1. Critical Examination: A systematic review of the core assumptions of behaviorism, cognitivism, constructivism, and connectivism to identify specific points of friction with generative AI capabilities.
2. Interdisciplinary Synthesis: Drawing upon existing scholarship in distributed cognition, extended mind theory, human-AI collaboration (specifically "centaur" models), AI literacy, cognitive offloading, and metacognitive research.
3. Framework Construction: Integrating these theoretical strands to propose a new learning theory, "Generativism," organized around four distinct principles.

Key Contributions
The primary contribution of the paper is the proposal of Generativism, a learning theory designed specifically for the age of generative AI. The authors distinguish this term from Chomsky's generative grammar, defining it instead as a theory of how humans learn in partnership with generative AI systems. The framework is anchored by four core principles:

1. Epistemic Partnership: Learning is redefined as the iterative co-construction of knowledge between human learners and AI systems. In this partnership, the AI provides speed and generative capacity (candidate explanations, examples), while the learner provides goals, context, values, and evaluative judgment. This shifts the view of AI from a passive tool to an active cognitive partner.
2. Distributed Agency: Agency is not transferred to the AI but is enacted through negotiation. The framework emphasizes the explicit, intentional, and pedagogically purposeful allocation of cognitive labor. It highlights the importance of "Decision-level Agency," where learners maintain control over intentions and commitments, ensuring they do not passively adopt AI-steered options.
3. Generative Literacy: Moving beyond functional or ethical competence, this principle defines literacy as the capacity to prompt, evaluate, refine, and integrate AI-generated knowledge artifacts. It comprises four capacities: prompt competence, evaluative judgment, epistemic vigilance (critical scrutiny of fluent outputs), and integrative synthesis (combining AI output with prior knowledge).
4. Adaptive Metacognition: This principle addresses the "cognitive offloading" paradox. It posits that effective learning requires the learner to continuously monitor and regulate the division of cognitive work between self and AI. Learners must detect over-reliance, reallocate effort, and reflect on how AI shapes their thinking to ensure genuine understanding rather than mere performance.

Results and Theoretical Implications
The paper does not present empirical results but offers a set of theoretical implications for educational practice:

• Instructional Design: The authors suggest moving away from banning AI or using it uncritically. Instead, they propose designing for "productive friction," where tasks require learners to evaluate, revise, and justify AI-generated outputs. Strategies include "production-comprehension inversion" (using AI drafts as starting points for critical analysis) and making the distribution of cognitive labor visible.
• Assessment: Assessment must differentiate between contexts requiring independent performance and those where AI collaboration is the target competence. The paper proposes new assessment indicators aligned with the four principles, such as evaluating a learner's ability to question AI outputs, explain their delegation of tasks, and demonstrate epistemic vigilance.
• Expertise Development: Expertise is reconceptualized not merely as the accumulation of domain knowledge (which AI can access) but as the capacity for effective epistemic partnership. This involves knowing when to rely on AI, when to override it, and how to structure collaboration to achieve outcomes surpassing either partner alone.

Significance and Limitations
The authors position Generativism as a preliminary response to the transformative challenges of generative AI. Its significance lies in providing a conceptual lens for educators, designers, and policymakers to rethink instructional design, learning, and assessment in AI-rich environments. The paper argues that without such a framework, stakeholders lack the tools to navigate the fluid boundaries between human and artificial cognition.

The authors explicitly acknowledge the limitations of this work:

• Conceptual Nature: Generativism is a conceptual framework that has not yet been empirically tested; its principles require systematic validation.
• Equity: The framework does not fully address issues of access and equity regarding differential access to generative AI tools across socioeconomic contexts.
• Technological Volatility: The rapid pace of AI development means the specific technologies referenced may evolve, potentially altering human-AI learning dynamics in ways the current framework cannot anticipate.

In conclusion, the paper asserts that while existing learning theories remain valuable, they are insufficient to address the unique dynamics of generative AI. Generativism offers a necessary theoretical foundation for investigating and refining the understanding of learning in an era where cognition is increasingly shared with generative systems.