Epistemic Agency in AI-Mediated Education: A PRISMA 2020 Systematic Review and Philosophical Synthesis of Autonomy, Understanding, and Educational Authority
This PRISMA 2020 systematic review synthesizes philosophical scholarship on AI-mediated education to argue that legitimate learning requires preserving human epistemic agency through a "Human Epistemic Stewardship" framework, which prioritizes learners' responsibility for questioning and justifying knowledge over mere performance optimization.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine a classroom where the teacher no longer just hands out facts, but where a machine can write essays, solve complex problems, and explain difficult ideas in seconds. This is the reality brought by generative artificial intelligence, a technology that has moved from being a simple tool to a partner in learning. For centuries, education has relied on a specific kind of work: the struggle a student faces when trying to understand a new concept, the effort of building an argument, and the responsibility of deciding what is true. This process is not just about getting the right answer; it is about the mental journey required to own that answer. When a machine can do the heavy lifting of thinking, a profound question arises for educators and philosophers: if the machine does the work, does the student still learn? The field of educational philosophy is now grappling with this issue, asking whether learning can happen when the cognitive labor is handed over to a system that produces knowledge without actually understanding it.
A recent systematic review by independent researcher Connor Nitchals tackles this question by gathering and analyzing the most serious philosophical discussions on the topic. The author examined twenty peer-reviewed articles published between 2023 and August 2026, focusing on how artificial intelligence changes the nature of learning, authority, and responsibility. The review did not look at test scores or software efficiency; instead, it looked at the deeper structure of education. The central finding is that the danger of artificial intelligence is not that it might give wrong answers, but that it might allow students to bypass the very process of thinking that builds their ability to judge and understand. The study suggests that for education to remain meaningful, students must retain the responsibility for questioning, explaining, and owning the ideas they produce, even when using powerful tools.
The researchers found that the core issue is what they call "epistemic agency," a term that simply means a person's capacity to be an active, responsible participant in the process of knowing. In a healthy learning environment, a student does not just receive information; they select questions, search for evidence, and test their own beliefs. The review argues that generative artificial intelligence threatens this agency by allowing learners to outsource the hard work of thinking. When a student uses a machine to generate an essay or an explanation without doing the mental work of constructing it themselves, they may produce a perfect-looking result while failing to develop the understanding required to defend it. The study distinguishes between "performance," which is the ability to produce a correct output, and "understanding," which is the ability to explain why that output is correct and how it connects to other ideas. The author warns that artificial intelligence makes it easy to achieve performance without understanding, creating a situation where a student can pass a test without ever having truly learned the material.
Another major theme in the review is the changing role of the teacher and the nature of authority. In the past, teachers were the primary source of guidance, but now students can turn to an artificial intelligence that speaks with confidence and fluency. The researchers argue that this creates a new kind of dependency. If a student trusts the machine's answer without question, they are not learning to think; they are learning to obey. The review emphasizes that education is not just about information delivery; it is a human relationship where a teacher guides a student through the difficult process of forming their own judgment. A machine can explain a concept, but it cannot take responsibility for a student's growth or offer the ethical guidance that comes from a human connection. The study suggests that if schools rely too heavily on these systems, they risk creating a system where knowledge is produced by algorithms, but the responsibility for that knowledge is lost.
The review also highlights the problem of "testimonial dependence." Humans naturally rely on others for information, but education teaches us how to decide who to trust and when to ask for proof. Artificial intelligence complicates this because it often presents its answers as facts without showing where they came from. The researchers found that for learning to remain valid, students must be able to challenge the machine, check its sources, and understand its limitations. They propose a framework called "Human Epistemic Stewardship," which sets out five conditions for using artificial intelligence in a way that protects learning. These conditions include the right to challenge the machine's output, the requirement that the student must be able to explain and defend their own work, and the need for "productive friction." This last point is crucial: the study argues that some difficulty in learning is necessary. Just as lifting weights builds muscle, the struggle to understand a concept builds the mind. Artificial intelligence should not remove all the difficulty; it should remove only the irrelevant barriers while leaving the mental work that leads to real understanding.
The author concludes that the debate is not about whether artificial intelligence should be used in schools, but about how it is used. The same tool can be harmful if it replaces the student's thinking, or it can be helpful if it supports the student's own efforts. The review suggests that schools need to redesign their assessments and teaching methods to ensure that students are still doing the work of thinking. This might mean asking students to explain their reasoning in person, to compare different sources, or to show how they used a tool to reach a conclusion. The ultimate goal is not to compete with machines, but to cultivate human beings who can direct, evaluate, and sometimes refuse the help of machines. The study finds that the value of education in an age of artificial intelligence lies not in the production of answers, but in the cultivation of people who know what an answer is worth, why it should be believed, and when it is their responsibility to think for themselves.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.