Generative AI in Higher Education: Evidence from an Elite College

Germ\'an Reyes; Zara Contractor

arxiv: 2508.00717 · v2 · submitted 2025-08-01 · 💰 econ.GN · q-fin.EC

Generative AI in Higher Education: Evidence from an Elite College

Zara Contractor , Germ\'an Reyes This is my paper

Pith reviewed 2026-05-19 01:29 UTC · model grok-4.3

classification 💰 econ.GN q-fin.EC

keywords generative AIhigher educationstudent adoptionChatGPTlearning augmentationinstitutional policiessurvey dataperceived benefits

0 comments

The pith

Students at a selective U.S. college adopted generative AI at over 80 percent within two years of ChatGPT's release, using it more to enhance learning than to replace their own coursework.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper uses survey responses from students at one selective college to track how generative AI spread after ChatGPT launched. It shows adoption is common but differs by field, background, and grades, with most students turning to AI for explanations and feedback instead of having it produce finished assignments. Because students who see AI as helpful for learning adopt it at higher rates and because rules land unevenly across groups, the work points to the need for policies that support useful uses while limiting shortcuts.

Core claim

Using survey data from a selective U.S. college, we document rapid generative-AI adoption, reaching over 80 percent within two years of ChatGPT's release. Adoption varies sharply across disciplines, demographics, and achievement levels. Students use AI both to augment their learning by obtaining explanations and feedback and to automate coursework by generating final outputs, with augmentation more common than automation. Students generally perceive AI as benefiting their learning, and these beliefs are strongly correlated with adoption. Institutional policies shape usage but have uneven effects, in part because awareness and compliance vary across student groups. These findings suggest that

What carries the argument

Survey data that separates augmentation uses such as seeking explanations and feedback from automation uses that generate final outputs.

If this is right

Effective policies must separate uses that enhance learning from those that substitute for student effort.
Uneven awareness and compliance mean broad rules may miss some student groups and require targeted outreach.
Strong links between perceived learning benefits and adoption suggest that education on helpful uses could increase constructive applications.
Discipline and demographic differences in adoption call for tailored rather than uniform institutional approaches.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

If the drivers of adoption and the preference for augmentation hold at other colleges, similar survey patterns would appear beyond elite institutions.
As AI tools grow more capable, the current distinction between augmentation and automation may require ongoing updates in policy design.
Controlled experiments that embed AI into specific assignments could test whether they raise augmentation without raising automation.

Load-bearing premise

Self-reported survey answers from students at this single selective college accurately reflect real AI usage, learning views, and policy compliance without large bias and can guide policy at other colleges.

What would settle it

A follow-up study that measures actual AI tool logs or collects responses at a non-selective college and finds substantially lower adoption or more automation than augmentation would challenge the main claims.

read the original abstract

Generative AI is transforming higher education, yet systematic evidence on student adoption, usage patterns, and perceived learning impacts remains scarce. Using survey data from a selective U.S. college, we document rapid generative-AI adoption, reaching over 80 percent within two years of ChatGPT's release. Adoption varies sharply across disciplines, demographics, and achievement levels. Students use AI both to augment their learning -- by obtaining explanations and feedback -- and to automate coursework by generating final outputs, with augmentation more common than automation. Students generally perceive AI as benefiting their learning, and these beliefs are strongly correlated with adoption. Institutional policies shape usage but have uneven effects, in part because awareness and compliance vary across student groups. These findings suggest that effective AI policies must distinguish between uses that enhance learning and those that substitute for it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This paper supplies original survey numbers on fast generative AI adoption at one selective college, with students reporting more augmentation than automation, but the self-report design leaves the key distinctions open to bias.

read the letter

The main takeaway is a set of survey results from one selective U.S. college showing generative AI adoption over 80 percent in two years, with sharper differences by discipline, demographics, and grades than I expected. Students report more use for getting explanations and feedback than for generating complete assignments, and those who see learning benefits are more likely to adopt it. Policies influence behavior but awareness and compliance are uneven across groups. These patterns give a concrete early picture of how the tools are landing in practice rather than just speculation about them. The paper does the basic job of fielding a survey that captures these distinctions and links them to perceptions and rules. That is new data in a space where most evidence has been informal. It gives concrete percentages and correlations that can anchor discussions about what institutional responses should look like. The authors appear to have thought through the different ways students might use the tools and built questions around that. The finding that augmentation is more common is interesting and could push policy away from blanket bans. The soft spots center on the data collection itself. Self-reports are the only source, so the split between augmentation and automation rests on how accurately students classify their own actions, which leaves room for bias. The stress-test concern holds up here because there is no cross-validation with logs or outcomes. One college also limits how much the policy findings travel to other settings with different student bodies or enforcement styles. This is for people who follow AI in education and want numbers to cite or test. It is worth a serious referee to examine the methods section, sample details, and whether any robustness steps were taken. I would send it for peer review.

Referee Report

3 major / 0 minor

Summary. The manuscript reports results from a survey of students at a selective U.S. college regarding their adoption and use of generative AI tools like ChatGPT. Key findings include rapid adoption reaching over 80% within two years of ChatGPT's release, significant variations across disciplines, demographics, and achievement levels, a distinction between using AI for learning augmentation (explanations and feedback) versus automation (generating final outputs), with augmentation being more prevalent, positive perceptions of AI benefits strongly correlated with adoption, and institutional policies having uneven effects due to varying awareness and compliance.

Significance. This paper contributes timely descriptive evidence on generative AI usage in higher education, which is currently scarce. The findings on adoption patterns, usage types, and policy impacts could inform institutional strategies if the data quality supports the claims. The single-institution nature and reliance on self-reports are limitations that affect the strength of the conclusions for broader policy recommendations.

major comments (3)

The abstract summarizes key findings but supplies no information on sample size, response rate, survey instrument, statistical methods, or controls, leaving the support for the reported variations and correlations difficult to evaluate.
The central claims rest on self-reported survey data without objective validation such as usage logs, assignment grades, or observed outputs. This leaves claims about augmentation versus automation and policy effects vulnerable to reporting bias.
The discussion of institutional policies shaping usage with uneven effects, in part because awareness and compliance vary across student groups, requires more specific evidence on how policies were measured and compliance assessed to support the uneven effects claim.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We believe the feedback will help improve the clarity and robustness of the manuscript. We respond to each major comment below.

read point-by-point responses

Referee: The abstract summarizes key findings but supplies no information on sample size, response rate, survey instrument, statistical methods, or controls, leaving the support for the reported variations and correlations difficult to evaluate.

Authors: We agree with this observation. The abstract in the current version focuses on the substantive findings but omits key methodological details. In the revised manuscript, we will expand the abstract to include the sample size, response rate, a description of the survey instrument, and the statistical methods employed, including any controls used in the analyses of variations and correlations. revision: yes
Referee: The central claims rest on self-reported survey data without objective validation such as usage logs, assignment grades, or observed outputs. This leaves claims about augmentation versus automation and policy effects vulnerable to reporting bias.

Authors: This is a valid concern regarding the limitations of survey data. While we recognize that self-reports can be subject to biases, such as over- or under-reporting of AI use, surveys are the standard approach for capturing student perceptions and self-described behaviors in this context. The manuscript already discusses potential limitations of self-reported data. We will revise the limitations section to more thoroughly address the possibility of reporting bias and its implications for the claims about augmentation, automation, and policy effects. Unfortunately, objective validation data like usage logs were not available for this study. revision: partial
Referee: The discussion of institutional policies shaping usage with uneven effects, in part because awareness and compliance vary across student groups, requires more specific evidence on how policies were measured and compliance assessed to support the uneven effects claim.

Authors: We appreciate the request for more specificity. The survey included targeted questions measuring students' awareness of institutional AI policies, their self-reported compliance with those policies, and how these factors influenced their usage patterns. To strengthen this section, we will provide additional details on the exact survey questions used and include supplementary analyses or tables showing variations in awareness and compliance across demographic and academic groups, thereby better supporting the claim of uneven effects. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical survey analysis relies on new primary data

full rationale

The paper collects and analyzes original survey responses from students at one selective U.S. college. All central claims—adoption rates above 80%, variation by discipline/demographics/achievement, augmentation versus automation usage, perceived benefits, and policy effects—are direct descriptive statistics and correlations computed from this fresh dataset. No equations, fitted parameters, predictions, or self-citations are invoked as load-bearing steps in any derivation chain. The analysis is therefore self-contained and does not reduce any result to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper is a descriptive empirical survey study; its claims rest on standard assumptions about self-reported data rather than new mathematical constructs, free parameters, or invented entities.

axioms (1)

domain assumption Self-reported survey responses accurately capture students' actual AI usage, perceptions of learning benefits, and awareness of institutional policies
All documented patterns, correlations, and policy effects depend on the validity of student answers to the survey questions.

pith-pipeline@v0.9.0 · 5661 in / 1453 out tokens · 56494 ms · 2026-05-19T01:29:28.250935+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith.Cost.FunctionalEquation washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Institutional policies substantially influence students’ reported likelihood of using generative AI. ... explicit prohibition creates a dramatic shift: only 13.4 percent of students report they would be likely ...

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.