Education is one of the first places people look when a new technology emerges. New tools often bring excitement that they will “revolutionize” learning. Just as often, these claims end up being overblown and the promised breakthroughs fall short.
One area where new technology may actually live up to its most ardent proponents’ expectations is in individualizing learning through customized learning paths. Advances in artificial intelligence and generative AI can potentially power tools that adjust to each learners’ pace, background knowledge, and interests. This informed optimism about AI is grounded in decades of research into the science of learning. Specifically, research into intelligent tutoring systems has provided insights into the design of educational technologies that can drive significant improvements in learning outcomes.
Innerloop and outerloop adaptivity
This research has identified different forms of adaptivity that are worth considering separately.
- Inner-loop adaptivity refers to the adjustments that happen within the course of a single learning task or experience. For example, if a learner makes a mistake while solving a problem, the system might offer feedback or hints that are specific to the error made.
- Outer-loop adaptivity refers to the sequencing of learning experiences. By tracking characteristics of the student (such as behaviors, prior knowledge, performance, and motivational elements like goals or interests), an adaptive educational technology can create a model to select an appropriate next learning experience. If a student is struggling with one topic because they lack a particular skill, a system that is adaptive can provide training on that specific skill.
Both forms of adaptivity are critical to creating effective customized learning paths, and both are areas where AI can have a significant impact. While current systems can handle some aspects of inner- and outer-loop adaptivity fairly well, AI brings the potential for more precise and responsive personalization.
The future of inner loop adaptivity: Real-time customization
In some AI-powered systems, such as Khan Academy’s Khanmigo, we are already seeing examples of how AI can enhance inner-loop adaptivity. Earlier learning technologies required preprogrammed responses to student errors, which limited their ability to address more nuanced mistakes or provide detailed explanations. Now, with generative AI systems like ChatGPT, customized feedback and support is generated on the fly. Instead of offering a generic “try again” message or simply giving the correct answer immediately, the AI-powered tool can guide the learner, asking follow-up questions or breaking down the problem in a way that helps the student reach the solution on their own. This aligns with best practices from tutoring research, which emphasizes the importance of supporting students in solving problems through scaffolding rather than providing answers immediately. While not all generative AI systems will produce these kinds of responses by default, careful prompting can help them behave like effective virtual tutors.
The future of outer-loop adaptivity: Optimizing and customizing learning sequences
The role of AI in outer-loop adaptivity—the sequencing of learning experiences—might be even more transformative. Advances in machine learning allow AI systems to create more detailed models of each learner, incorporating not only their mastery of different topics but also factors like motivation, interest, and even learning goals. This richer understanding enables AI systems to choose the next task that keeps students in their Zone of Proximal Development—the sweet spot where they are challenged enough to grow beyond their current capabilities.
For example, an AI-driven system could recommend the ideal next problem based on a student’s performance in previous lessons, ensuring that the material is appropriately challenging. At the same time, AI could factor in motivational elements, such as aligning tasks with the student’s interests, creating a more engaging and meaningful learning experience. Research by Candace Walkington and colleagues in math education, for instance, has found that students are more likely to understand a new concept when it is presented in a context that aligns with their existing interests. As an example, consider how a standard problem involving calculating the area of a fence might be reframed as a task related to building a soccer field for a student who loves sports. This small adjustment can make a big difference in how the student connects with the material, leading to longer-term learning benefits. In the world of professional learning, it is critical that what learners spend their time on will be useful and relevant to them, so this kind of customization may prove highly effective.
Making the adaptive choice
Educational technologies that build learning paths tailored to an individual’s goals, interests, and proficiency feel closer now, in this emerging era of AI, than ever before. However, it’s important to approach these innovations with a critical eye. While the potential is enormous, not every solution will be built with thoughtful design and integration with existing research on adaptivity. Careful examination of the specific tool or product is necessary; take the time to evaluate the claims, approaches, pedagogical perspectives, and evidence of impact that the solution’s creators provide. It is important to be on the lookout for what some leading computer scientists call “AI snake oil” and to be mindful of claims that sound too good to be true. And don’t lose sight of the value of human input and expertise; learning materials designed by subject matter experts and experienced instructional designers are still likely to be superior to anything that an AI system generates on the fly. Solutions that combine that kind of excellent content with technologies that provide a more customized experience will likely emerge as the most effective.
