As Large Language Models (LLMs) become increasingly reliable and accessible, middle and high school students will have nearly unlimited access to an unprecedented amount of AI tools in their daily lives. While the ultimate capacity of these tools is still unknown, it’s easy to predict how students will incorporate AI into their education. As with every other tool in human history, students will use AI to accomplish work of value with less effort.
Given this premise, how students use AI tools will depend more on their perceptions of their role in school than the tools themselves. As students reach for AI tools, do they see themselves as active agents in their own learning or as passive recipients of assigned tasks that need to be completed in exchange for points and grades?
Passive Compliance with AI . . .
For some students, school is an exercise in passive compliance. They believe their role is to receive new content, repackage or replicate it, and send it back to their teacher to assess their work and assign a grade. For these students, AI is the ultimate tool for efficiently “doing” school. They can achieve their goals by simply prompting AI tools to complete assigned tasks:
You want them to read a 12-page article?
Their prompt: Summarize this article into a paragraph.
Do math problems 1–21?
Their prompt: Complete the math problems in the photo attached.
Write a five-paragraph essay on To Kill a Mockingbird?
Their prompt: Write a five-paragraph essay from the perspective of a high school student who admires Atticus Finch because of his integrity, honesty, and wisdom.
However, the efficiency of this surface-level use of AI comes at a steep cost. Consider these conclusions drawn from recent research:
- Unguided, superficial prompts harm learning. In one study, students completing assignments using a standard version of ChatGPT typed or pasted the assigned question directly into the AI tool more than 60 percent of the time. In the moment, these students got the correct answers, but after the tool was removed, they performed worse than students who hadn’t used AI on the assigned tasks (Bastani et al., 2024).
- Cognitive ease comes at a cost. Students reported that using AI tools for synthesis tasks made their work easier than navigating information across multiple resources, but their final products were assessed as less thorough and of lower quality than those of students who didn’t use AI tools for the same task (Stadler, Bannert, & Sailer, 2024).
- Overreliance on AI negatively affects students’ independence. By relying on AI tools to find answers or complete tasks, students avoided the productive struggle required to develop a deeper understanding of content and diminished their capacity to navigate complex tasks independently (Zhai, Wibowo, & Li, 2024).
Given these findings, the cost of superficial use of AI as a tool for compliance may be that it short-circuits students’ conceptualization of how learning occurs. If students see summarizing text, solving math problems, and writing effectively as something that magically emerges from AI tools, they may come to attribute success or failure on academic tasks to external mechanisms that need not be either observed or explained. Detaching strategy and effort from one’s results eventually leads to learned helplessness.
. . . vs. Active Engagement with AI
On the other hand, students who value learning focus on a different set of goals and adopt a different role. For these students, their goal isn’t merely to complete their work, but to engage in the inquiry, productive struggle, and reflection that results in deeper understanding.
These students have a strong sense of agency; they take intentional action to exert control over their learning environment (Bandura, 2006). They believe their role is to actively plan, monitor, and adapt their strategies to pursue goals for learning. Not surprisingly, student agency is associated with higher levels of perseverance, engagement, learning, and achievement.
What do highly agentic students do when they interact with AI tools? Consider the following anecdote. Secondary teacher Jack Dougall (2023) gave his students a performance task and told them they could complete it using AI any way they wanted. Most students compliantly asked the AI tool for the answers, and regardless of whether they understood the output, they dutifully replicated those answers and turned them in.
But there was a small group of students who took a more active, agentic approach. According to Dougall, these students were
the rogue learners, the questioners, the bickerers, the challengers. . . . They “chatted” with the AI, questioned it, and they argued with it. They fine-tuned their prompts and engaged in a back-and-forth with the AI. (2023)
This group of students became so engaged in the task that they clamored for more time to write down all they’d learned. On an assessment given after the task was complete, the agentic, “rogue” learners outperformed their compliant classmates.
When Students Lose Agency
Unfortunately, by the time students reach middle school, the agentic, rogue learners are in the minority. Levels of agency and engagement decline as students progress through school (Anderson et al., 2019).
Some of this decline is related to developmental factors. However, some of it is related to how students cope with the realization that they don’t believe they have much control over their learning environment. Consider the following quote from Reeve and Tseng about how learners might express their needs for agency during instruction:
Students might offer input, express a preference, offer a suggestion or contribution, ask a question . . . seek ways to add personal relevance, ask for a say in how problems are to be solved, seek clarification . . . or request assistance such as modeling, feedback, background knowledge, or a concrete example of an abstract concept. (2011, p. 258)
As any kindergarten teacher who reads this quote can tell you, students arrive to school bursting with agency. But as any kindergarten teacher who reads this quote can also tell you, an entire classroom of students with this level of agency at all times would be impossible to teach. Out of logistical necessity, an emphasis on compliance and grades, and curriculum and teaching that often doesn’t encourage exploration or inquiry, students are taught that some passive compliance is necessary to make schooling possible.
As much as we lament students whose only questions seem to be, “How many points is this worth?” and “Can you just tell me the answer?”, we have to acknowledge that they didn’t arrive at school that way. If students conclude that schooling isn’t really about their learning, they’ll make the logical but debilitating choice to trade agency for learning with efficiency.
How AI Can Recharge Student Agency
Fortunately, teachers can use strategies that empower students to act with greater agency (Reeve & Tseng, 2011). Among a variety of factors associated with student agency, two of the most important are autonomy and competence (Ryan & Deci, 2017).
Autonomy is the need to exercise control over one’s own actions. It is fostered when students can make intentional decisions about how, what, where, and when they learn, but undermined when they feel controlled, dependent, or helpless.
Competence is the belief that one can apply strategies to accomplish important, challenging tasks. It is supported when students engage in standards-aligned tasks that balance their needs for productive struggle, incremental progress, and success, but undermined when tasks are too easy or too complex or when students cannot access developmental feedback to adjust and refine their strategy and effort to learn.
Effective teachers strive to balance these competing needs. Autonomy without attention to students’ needs for competence leaves students overwhelmed. Didactic strategies, where telling is accepted as a proxy for teaching, prevent students from exercising their autonomy to apply strategies that build competence.
When prompted intentionally, AI tools can be a nonjudgmental, endlessly patient resource to balance each student’s needs for autonomy and competence.
What I find remarkable about the earlier quote from Reeve & Tseng about the attributes of agency is that responding to these types of observations and inquiries is exactly what generative AI tools excel at. When prompted intentionally, AI tools can be a nonjudgmental, endlessly patient resource to balance each student’s needs for autonomy and competence. Recent research supports this claim. For example:
- AI tools can support students’ needs for competence when prompted in specific ways. In studies that show increased student learning as a result of using AI, the AI tool had been prompted to respond in ways that align with students’ needs for competence: to check for understanding before providing information, to not tell students the answers to any assigned questions until students attempted them first, to ask clarifying questions before moving on, and to limit the length of responses to ensure they meet students’ information needs (Bastani et al., 2024; Kestin et al., 2024; Kumar et al., 2023).
- AI tools can support students’ needs for autonomy when prompted in specific ways. Researchers found that undergraduate physics students learned twice as much—in less time—using an AI tutor than in a whole-group, interactive lesson (Kestin et al., 2024). In this case, the AI tool was fine-tuned to adhere to the competency-supportive constraints such as checking for understanding before providing information and not just telling the students answers. Not only did the AI-tutored students learn more, but they also reported higher levels of engagement and motivation. Why? The authors noted that students could spend as much—or as little—time as necessary in the module, ask and receive immediate answers to clarifying questions, and receive and respond to personalized feedback in real-time.
The conclusion here is not that AI tools can, or should, replace teachers, but that AI tools can support engagement and learning when they are used in agentic rather than superficial ways.
How to Support Students’ Agentic Use of AI Tools
Here are just a few ways teachers can help students use AI tools that balance their need for autonomy and competence and allow them to actively engage in the learning process.
1. Help students build competence by focusing on learning goals rather than task completion.
If students think the purpose of an assignment is to answer the teacher’s questions correctly, they’ll use superficial prompts to gain compliance. To avoid this, teachers can articulate how the task connects to a goal for learning as follows.
- Do: Problem set 3.1 . . . so you can: Use precise mathematics vocabulary to explain the benefits and limitations of applying linear models to real-world scenarios.
- Do: The reading on pp. 134–139 and complete the discussion-prep graphic organizer . . . so you can: Identify and describe factors that impact patterns in immigration across geographic regions.
Clarifying what students need to do and why gives students language to prompt AI tools to help them understand the goal, rather than simply complete the task.
2. Teach students how to prompt AI tools to support their current level of competence.
To engage with AI tools in a manner that supports their current level of competence, students should:
- Share context with the AI tool about the course and one’s current level of understanding: For example, is it a middle or high school course? An introductory course or advanced? Is the material new to the student, or is this review?
- Share the learning goal: By sharing the learning goal, rather than just the assigned task, the AI tool can be used to help support the student’s progress toward the goal rather than simply providing answers to assigned tasks.
- Provide constraints: Give the AI tool specific guidance about what it should not do. For example:
Don’t just give me answers. Ask for my response or my solution first. Check my understanding before moving on. Answer my questions with limited information and then check my understanding. Go slowly, one step at a time.
3. Empower students to monitor their competence by engaging in self-assessment and asking for feedback.
Self-assessment helps students set realistic goals and focus their use of strategies and effort to improve. Drawing on action steps 1 and 2, consider the following sample prompt:
I am a high school student taking a U.S. History course. I am supposed to be able to “compare and contrast the Articles of Confederation with the United States Constitution as related to the separation of powers and state’s rights.” I’ve read the introductory article my teacher provided (attached) and I think it makes sense to me. Can you ask me four open-ended questions about important points discussed in the article? Then, after I’ve replied, review my responses for accuracy and clarity, point out any misconceptions, and ask any clarifying questions that would be beneficial.
4. Teach students how AI tools can support their autonomy to navigate productive struggle and ask for help.
Unfortunately, many students think that productive struggle and the need to ask for help are evidence of their inability to grasp a concept. In reality, these are opportunities to be embraced to accelerate learning. Drawing on the previous action steps, consider the following prompt and sentence frames for a student seeking help with a learning goal that requires the student to compare and contrast scientific phenomena. Remember to add the constraints from action step 2.
I am a 9th-grade student learning about similarities and differences between mitosis and meiosis. I know that [write in as much detail as you can], but I don’t understand how _____ and _____ are different. Specifically, I don’t understand why _____. Can you provide feedback on the accuracy and logic of my current thinking? Then, help clarify where I’m getting stuck.
A similar approach can be used with slightly different sentence frames that align goals to relevant questions for explaining a process, describing part-whole relationships, discerning important similarities and differences, and so on.
5. Encourage students to balance their needs for autonomy and competence by advocating for their learning needs.
Students with a strong sense of agency know that learning isn’t passive. Students can exercise agency with AI tools through persistent, directive questions and prompts that support their learning. Example prompts include:
• I’ve answered some questions from my teacher, but I’d prefer to arrange the information using a graphic organizer. Can you help me set that up?• I’m overwhelmed with a reading assignment. I’ve read it through three times and this section doesn’t make any sense. [Insert section.] Can you ask me a couple of questions to check my understanding?• Without a context for the variables in these problems, I am having a hard time visualizing what this math assignment means. Can you give me a real-world context where this equation could be used?• You are going too fast. I told you to go one step at a time and not to overwhelm me with information. Slow down and provide shorter responses.• The way you explained that concept is a bit different than how it is described in my textbook. I’ve attached a photo of the explanation from my text. Summarize the text I’ve provided and use that explanation as we continue our discussion.
AI for Agency
With the right approach, students can use AI tools in ways that allow them to take control of their learning. If we want students to move beyond superficial uses of AI tools to gain compliance, we can begin by teaching them how to interact with AI tools in ways that value and support their needs as learners.
Reflect & Discuss
➛ What assumptions about grading, learning, or technology prevent your students from seeing AI as more than a tool for compliance? How can you help your students move beyond those assumptions?
➛ Of the five strategies described in the article, which one do you think has the greatest potential to support your students’ learning needs? In what ways?
