📄 social_science

When Confidence Shifts Toward the Tool: AI Appropriation, Reliance Calibration, and School Guidance in High School Academic Work

A survey of 1,444 US high school students reveals that AI appropriation involves complex, co-occurring learning and shortcut behaviors while frequent use correlates with reduced self-confidence and a widening reliance gap, suggesting that schools must move beyond general guidelines to provide targeted teacher-guided practice for effective AI reliance calibration.

Original authors: Larry Zhiming Xu, Gabriel Velez, Aubrie Robinson

Published 2026-07-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Larry Zhiming Xu, Gabriel Velez, Aubrie Robinson

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

The Big Picture: It's Not About If They Use It, But How

Imagine a school where almost every student has a super-smart, invisible tutor in their pocket (AI). By late 2025, about two-thirds of U.S. teenagers were using these tools for homework.

The researchers asked: Does it matter if a student uses AI?
Their answer: Not really. Just knowing a student uses AI is like knowing a driver has a car. It doesn't tell you if they are driving safely, speeding, or using the GPS to find a shortcut they shouldn't take.

Instead of counting how often students use AI, this study looked at how they use it (what they ask it to do) and how they feel about it (do they trust the AI more than their own brain?).

Key Concept 1: The "Appropriation" (How They Use the Tool)

The researchers found that students don't fit into neat boxes like "Good Student" vs. "Cheater."

The Analogy: Think of AI like a Swiss Army Knife.
The Finding: Most students use the knife for many different things. The same student who asks the AI to "explain this math concept" (like using the knife to open a letter) is often the same student who asks the AI to "write this whole essay for me" (like using the knife to cut a rope they should have tied themselves).
The Takeaway: You can't judge a student just by how much they use AI. A student who uses it a lot might be using it to learn, to cheat, or to do both at the same time. The "good" and "bad" uses are often mixed together in the same person's workflow.

Key Concept 2: The "Reliance Gap" (The Confidence Shift)

This is the most important discovery in the paper. The researchers measured two things:

How confident a student feels doing work without AI.
How confident they feel that AI can help them do the work.

They calculated the difference between these two feelings, calling it the "Reliance Gap."

The Analogy: Imagine a tightrope walker.
- Low Gap: The walker trusts their own legs more than the safety net. They know they can walk the line, but the net is there just in case.
- High Gap: The walker trusts the safety net more than their own legs. They feel shaky on their own but feel safe only when the net is there.
The Finding:
- Students who don't use AI feel very confident in their own abilities and don't trust the AI much.
- As students use AI more often, their confidence in their own ability drops slightly, but their confidence in the AI shoots up.
- The "Crossover": Among students who use AI every day, something interesting happened: their confidence in the AI's help became slightly higher than their confidence in their own ability. They haven't lost their skills, but they now trust the tool more than themselves.

Key Concept 3: The "Missing Middle" (What Schools Are Doing Wrong)

The researchers asked: "Do schools help students figure out how to use AI correctly?"

The Analogy: Imagine a gym. The school has put up a sign saying, "You are allowed to use the weights." (This is a Policy). But the school hasn't hired a trainer to show you how to lift them without hurting your back.
The Finding:
- Schools are mostly providing Rules (e.g., "You can use AI if you cite it") and Resources (links to websites).
- Students are screaming for Practice (e.g., "Show me how to use this tool to learn, not just to get an answer").
- The study found that having a school policy did not help reduce the "Reliance Gap." Students who felt they had good guidance didn't necessarily trust themselves more than the AI.
- The Gap: The biggest unmet need was for teachers to guide students through actual practice—showing them when to use the tool and when to put it down.

The Four Types of Students (Profiles)

The researchers grouped students into four "personalities" based on their habits:

The Skeptical Independent (18%): Rarely use AI. They trust themselves 100% and the AI 0%.
The Unsupported Routine User (36%): Use AI often, but feel the school isn't helping them. They want more guidance but aren't getting it.
The Guided High-Efficacy User (31%): Use AI often, but they have good teacher support. They still trust themselves the most.
The Shortcut-Reliant Power User (15%): The "Heavy Users." They use AI constantly, mostly to get answers or write drafts. They trust the AI more than themselves. This group has the biggest "Reliance Gap."

The Bottom Line

The paper argues that schools need to stop just asking, "Are you using AI?" and start asking, "Who are you trusting more: your own brain or the tool?"

The goal isn't to ban AI or just say "it's okay." The goal is Calibration. Schools need to teach students how to use the tool as a helpful assistant without letting the tool become the master of their thinking. Right now, schools are good at making rules, but they are bad at teaching the practice of using the tool wisely.

In short: Students are using AI to do their homework, and as they use it more, they are starting to trust the robot more than their own brains. Schools need to step in and teach them how to keep their own confidence strong while using the tool.

Technical Summary: When Confidence Shifts Toward the Tool

Problem Statement
As artificial intelligence (AI) permeates high school education, traditional metrics of "adoption" (i.e., whether students use AI and how often) have become insufficient for understanding the technology's evolving role. Schools currently monitor usage frequency and the existence of formal policies, but these indicators fail to capture how students integrate AI into their cognitive processes. Specifically, existing frameworks cannot distinguish between uses that support learning (e.g., explaining concepts) and those that bypass it (e.g., generating full drafts), nor can they assess whether a student's confidence in their own academic ability is eroding as reliance on the tool increases. The central problem is the lack of a diagnostic framework to measure the dynamic relationship between technological capabilities and students' self-perceptions of competence in an environment where AI is readily available.

Methodology
The study employed a cross-sectional survey design with a sample of 1,444 students across a network of Catholic, college-preparatory high schools in the United States. Data were collected in Spring 2025.

Measures: The instrument assessed AI use frequency, tool breadth, and specific "appropriation" purposes (categorized as learning-oriented vs. shortcut-oriented). Crucially, it measured two distinct confidence levels: confidence in handling schoolwork without AI and confidence in AI's ability to help with schoolwork.
Key Construct: The study operationalized "reliance calibration" through the reliance gap, defined as the difference between confidence in AI's help and confidence in independent ability. A positive gap indicates confidence has shifted toward the tool.
Analysis: The researchers utilized hierarchical ordinary least squares (OLS) regressions with school-clustered standard errors to predict the reliance gap. Variables included demographics, use frequency, purpose breadth, AI-related beliefs (self-efficacy and optimism), and perceived school guidance. Additionally, k-means clustering was used to generate descriptive student profiles based on appropriation patterns.

Key Contributions
The paper makes three primary contributions:

Conceptual: It shifts the analytical lens from "adoption" to appropriation (how users assign purposes to technology) and introduces reliance calibration to the educational context. This framework posits that appropriate reliance depends not only on the tool's reliability but on the student's developmental need to build independent competence.
Empirical: It captures a large, multi-school sample at a critical juncture—after AI access became routine but before institutional guidance matured. It provides evidence that learning-oriented and shortcut-oriented uses often co-occur within the same students, challenging the binary view of "good" vs. "bad" AI users.
Practical: It identifies a "missing middle" in school guidance, where policy-level rules exist but pedagogical practice (teacher-guided instruction) is lacking. It argues that the reliance gap is a more diagnostic metric than independent confidence alone for understanding student-AI dynamics.

Results

Appropriation Patterns: 94% of respondents used at least one AI tool. Learning-oriented uses (e.g., tutoring, outlining) were common (65% for tutoring), but shortcut-oriented uses (e.g., finding answers, drafting) were also substantial (32% and 18%, respectively). Crucially, these purposes frequently co-occurred; students who used AI for explanations were often the same ones using it for answers, suggesting a complex, intertwined usage pattern rather than distinct user groups.
The Reliance Gap: Use frequency was negatively associated with independent confidence ( $r = -0.18$ ) but strongly positively associated with the reliance gap ( $r = 0.45$ ). Among daily users, confidence in AI's help ( $M=4.17$ ) slightly exceeded confidence in independent ability ( $M=4.02$ ), indicating a "crossover" where the tool becomes the primary anchor of confidence.
Predictors of the Gap: Hierarchical regression revealed that use frequency, the breadth of learning and shortcut purposes, and AI optimism were significant predictors of a larger reliance gap. Notably, perceived school guidance did not significantly predict the gap once usage and beliefs were accounted for.
Guidance Mismatch: While general guidelines were the most commonly reported form of school support (48%), the largest unmet demand was for teacher-guided practice (42% wanted it vs. 27% received it) and dedicated AI classes.
Student Profiles: Four distinct profiles emerged:
1. AI-skeptical independent learners (18%): Rare users, high independent confidence, negative gap.
2. Unsupported routine users (36%): Frequent users, low guidance, high mismatch.
3. Guided high-efficacy users (31%): Frequent users, high guidance, retained high independent confidence.
4. Shortcut-reliant power users (15%): Highest usage, universal shortcut use, lowest independent confidence, and the only group with a positive reliance gap.

Significance and Claims
The paper argues that as AI becomes part of the ordinary infrastructure of schoolwork, the focus must shift from tracking if students use AI to understanding how they calibrate their reliance. The study claims that the reliance gap captures a critical dynamic: the point at which a student's sense of academic capability shifts from the self to the tool.

The authors assert that the co-occurrence of supportive and substitutive uses complicates simple "cheating vs. learning" dichotomies, suggesting that the boundary between augmentation and offloading is procedural rather than categorical. Furthermore, the finding that perceived school guidance (specifically policy-level rules) does not correlate with better calibration implies that current guidance models are misaligned with student needs. The paper concludes that schools must move beyond acceptable-use policies to provide "guided practice" that helps students distinguish between AI assistance that supports learning and assistance that displaces the cognitive work necessary for skill development. The reliance gap is proposed as a practical, scalable metric for schools to monitor this shift in confidence and competence.