Learning with Generative AI: Strategies for Durable Learning in the AI-Driven Era

Overview

Generative AI can explain difficult concepts, solve problems, and produce answers in seconds, creating opportunities for rapid learning. Yet completing a task quickly does not guarantee that the learner can reproduce the answer later without AI help. The article argues for strategies that promote durable knowledge, including retrieval practice, reflection, and structured support rather than merely supplying solutions.

Key takeaways

• AI can support active recall and exploration, not just final answers.
• Combining apps like Duolingo with large language models creates richer learning ecosystems.
• Reflective journaling helps learners articulate strategies and evolving understanding.
• Learning literacy and AI literacy are essential for responsible use and design.

Overview

Generative AI has the potential to accelerate learning by explaining difficult concepts, solving problems, and providing answers within seconds. This capability creates vast opportunities for education, but it also raises questions about whether learners retain understanding when AI is not available. The article emphasizes that the goal of learning should be durable knowledge, not merely the ability to produce correct answers with AI assistance. It argues for deliberate, strategy-focused use of AI—where learners decide how to deploy AI tools to support retrieval, reflection, feedback, and independent problem solving—so that skills transfer beyond the immediate task.

In discussing language learning as a concrete example, the piece notes how apps such as Duolingo provide structure, repetition, and motivation, while learners can deepen the experience by attempting recall from memory, speaking sentences aloud, and revisiting difficult vocabulary. A key principle highlighted is deliberate retrieval: retrieving information from memory strengthens long-term retention more effectively than simply re-studying material. By contrast, a large language model like ChatGPT can play a different role, offering explanations, comparing sentence structures, generating additional examples, or helping learners probe their own observations.

The article suggests that AI and specialized apps can work together to form a richer learning ecosystem. A small case study of Duolingo users found that reflective journaling—writing down thoughts, goals, strategies, and difficulties—helped learners articulate their evolving approaches, which in turn heightened their awareness of their own learning process. This finding points to the broader importance of metacognitive practices in AI-assisted learning.

Opportunities and Risks of AI in Learning

While AI can provide immediate answers, it can also offer hints, prompt learners to attempt problems first, identify misconceptions, generate new examples, and prompt learners to explain their reasoning. The article emphasizes that the critical question is not simply whether AI is used, but which learning activities it enables learners to stop doing and which it helps them do better.

Evidence from a large field experiment with nearly 1,000 secondary school maths students compared two GPT-4 variants: a standard version that supplied answers and a tutor-designed version intended to encourage tutoring-like guidance. Students given the standard GPT-4 showed performance gains while they had access, but when access was removed, their performance dropped by 17% relative to peers with no AI access. The tutor-designed GPT-4 group largely avoided this decline, highlighting how design choices can influence long-term outcomes and reducing the risk that AI use erodes core learning skills.

The broader implication is not to deny AI or force struggle, but to design AI that supports learners in ways that preserve the mental effort central to learning. The model should offer hints, ask for problem-solving steps, and encourage learners to justify their reasoning rather than simply providing answers.

Self-aware Learning: AI Literacy and Learning Literacy

Beyond practical use, the article argues for explicit attention to AI literacy—understanding what AI systems can and cannot do, how to question their outputs, and when their use is appropriate. It introduces the concept of learning literacy as a public-facing description of the knowledge and judgment people need to regulate their learning in technology-rich environments. A meta-analysis cited in Frontiers in Education reports substantial positive effects of AI-supported interventions on self-regulated and self-directed learning, though the magnitude of effects varied based on how technology and learning experiences were designed. Responsibility for learning outcomes therefore rests not only with learners but also with schools, universities, and developers who can design AI to encourage retrieval before assistance and to prompt explanation of reasoning.

The article closes with a practical reminder: learners can ask themselves what they are actually learning by using AI in a given way. The overarching educational challenge is to become as skilled at recognizing when AI-provided answers support learning as when they merely help finish a task.

Conclusion

AI is transforming how we search for and consume knowledge. The challenge for education is to ensure that AI enhances genuine learning rather than simply enabling quicker task completion. By prioritizing retrieval, reflection, and feedback, and by cultivating AI and learning literacy, schools, developers, and learners can harness AI to deepen understanding and foster independent problem solving rather than dependency.