You’ve probably seen it happen too. Students open an AI tool, get a polished essay outline in minutes, submit their assignments, and leave feeling productive. They did well in the exam. The grade is real. But when you ask someone to explain the same concept to you three months later, the room goes silent.
This is not a new problem. Shortcuts in learning have been around forever. But AI has greatly amplified the problem. The problem is not that AI competes with textbooks, but that it competes with friction. Textbooks are slow by design. They force you to sit with a concept for hours, tracing its intellectual history, making your own connections, taking notes, and struggling to understand it. That conflict is not a flaw in the learning process. that teeth process.
Think of your brain as a muscle. To build lasting strength, you need to overcome resistance. The same goes for learning. Neuroscientists call the mechanism synaptic plasticity: The brain’s ability to form and strengthen connections between neurons when actively engaged with new substances. When you grapple with a difficult concept and arrive at an understanding, your brain does more than simply store the information. It actually rewires itself. This rewiring makes knowledge permanent, transferable, and truly yours.
Learning science research distinguishes between surface learning and deep learning. Surface learning relies on memorization and routine recall of information absorbed in isolation, with little or no personal investment or connection to broader understanding. Deep learning creates meaningful connections between new and existing knowledge, and there are clear differences between the learning outcomes of these two levels. Passive and superficial engagement is associated with forgetting 50 to 70 percent of new material within the first 24 hours. A randomized controlled trial at Georgetown University found that medical students who used ChatGPT during study sessions received higher scores on an immediate assessment, but this advantage disappeared when assessed one week later.
The dangers of relying on chatbots for thinking include: cognitive outsourcing. This manifests itself when students ask an AI to brainstorm their ideas before posting them, paste a research paper into a chat window and ask for an interpretation without looking at the numbers, or accept a summary of literature without checking it with the original source. In both cases, the brain is bypassed and the student is left with only the illusion of mastery.
The problem boils down to: how Rather than using AI Whether or not It uses AI, right? Using AI tools during learning is extremely powerful. You can speed up the process and improve its quality by acting like a tutor or coach, providing instant personalized feedback, finding gaps in learning, and presenting concepts in different ways.
But users beware. In a research context, cognitive outsourcing carries additional risks. AI tools tend to generate fabricated content that is presented as established fact, such as fabricated citations to non-existent peer-reviewed papers. Beyond academic integrity, there is a more subtle cost: the loss of intellectual individuality. Original research relies on unique perspectives and novel frameworks. When AI replaces that process, the creative core that makes your work your own is lost.
What does good practice actually look like?
In my (Jacob’s) undergraduate physics course, we don’t prohibit AI tools, nor do we treat them as shortcuts to avoid them. Instead, I strive to incorporate AI into the learning process so that the strengths and limitations of AI are visible to students. What I discovered is that while AI will not replace learning, it will fundamentally reshape how students approach learning, and whether that reshaping will help or hurt will depend almost entirely on how we teach our students.
One of the places where this is most evident is in how students read scientific literature. In a series of research and writing courses that I teach, first-year undergraduates use AI tools early on to identify papers and generate summaries about their research questions. For most of these students, this is their first time reading scientific literature. When they see a clean, professional-sounding summary, they initially tend to accept it at face value. Because we don’t yet have a framework for questioning it.
This is where course structure becomes important. After creating a summary, students should go back and analyze the original paper directly and compare the content of the AI summary with the content of the paper. Students begin to notice what is missing, such as caveats in experimental design, data limitations, or the overall flow of the discussion that is not included at all in the abstract. Often, they discover their research ideas from precisely that “missing piece.”
What this creates is a kind of transitional moment. AI moves from being an apparent authority to a starting point. Students begin to realize that the output is useful but incomplete and, more importantly, something they are responsible for verifying. The goal is not just to help you find articles more efficiently. This is because reading a paper is not just about extracting relevant information, it is about fully interpreting that information and data to ask better research questions.
Several specific expectations can be derived from this framework. First, transparency is non-negotiable. When students use AI, they must document exactly how they use it, including by providing a transcript of the conversation. Second, students should avoid treating AI as a black box. For example, you need to not only prompt the AI tool to do the analysis, but also generate a custom program to use the AI tool to perform data analysis. In this way, you can maintain control over the tool because the code is the tool, not the AI itself. Third, students should never accept their first reaction. Whether it’s a literature summary or a block of code, it needs to be checked, tested, and improved.
With this in mind, here’s my practical three-step framework for getting students to use AI.
- Always start with your own attempts. Draft a rough answer, outline, or plan. This exercise gives you something to compare and criticize. AI feedback is most useful when you already have skin in the game and a perspective to defend.
- Never accept the first response. Whether it’s a literature summary, a block of code, or an explanation of a new concept, check it against primary sources, test it, and refine it. Ask the AI to make inferences and then verify those inferences yourself.
- Document usage. Transparency is not just an academic policy, it is a professional practice. Recording your interactions with AI allows you to recognize where the AI was actually helpful and helps you create an iterative feedback process.
It is important to develop good habits early on. Because what I’m really doing in the undergraduate curriculum is preparing students for the much more dangerous environment of graduate study. What is important here is not only the course content, but also the time, sense of responsibility, and reliability of research. After all, time is the most limited resource graduate students have, and AI can Helps you save time and speed up your learning. But that efficiency improvement must be based on understanding, validation, and a clear sense of intellectual property rights.
The same philosophy carries over to how we frame AI more broadly within the curriculum. I explain the difference by naming students as principal investigators on a project and AI tools as lab assistants, and show students the distinction between using AI to support the work they are doing and to replacing it. If AI is helping you perform something that is not a core intellectual task (such as generating Python code to analyze data collected in an experiment you run), it may be an appropriate use. But if it’s doing the core thinking for them, it will undermine their learning.
I believe there is reason for optimism. This is due to the fact that these habits can be developed early on. Even as students learn to use AI as a tool to direct rather than an authority to follow, they do not lose their ability to think deeply. In fact, it strengthens it.
