Scaffold or Crutch? Examining College Students' Use and Views of Generative AI Tools for STEM Education
Open accessThe DOI matches its Crossref record; 48 of the 49 references it cites were found in a registry or an index, and 1 are not checked yet; 8 of the 25 sources its claims cite were checked for support.
Abstract
Developing problem-solving competency is central to Science, Technology, Engineering, and Mathematics (STEM) education, yet translating this priority into effective approaches to problem-solving instruction and assessment has been a significant challenge. The recent proliferation of generative artificial intelligence (genAI) tools like ChatGPT in higher education introduces new considerations: how to define problem-solving competency in a genAI era, and how these tools can help or hinder students' development of STEM problem-solving competency. Our research takes steps in examining these considerations by studying how and why college students are currently using genAI tools in their STEM coursework, with a specific focus on how they employ these tools to support their problem-solving. We conducted an online survey of 40 STEM college students from diverse institutions across the US. In addition, we surveyed 28 STEM faculty to understand instructor views on effective and ineffective genAI tool use in STEM courses and their guidance for students. Our findings reveal high adoption rates and diverse applications of genAI tools among STEM students. The most common use cases of genAI tools in STEM coursework include finding explanations, exploring related topics, summarizing readings, and helping with problem-set questions. The primary motivation for using genAI tools in STEM coursework was to save time. Moreover, we found that over half of the student participants reported simply inputting a problem for AI to generate solutions, potentially bypassing their own problem-solving processes. These findings indicate that despite high adoption rates, students' current approaches to utilizing genAI tools often fall short in enhancing their own STEM problem-solving competencies. The study also explored students' and STEM instructors' perceptions of the benefits and risks associated with using genAI tools in STEM education. Our findings provide insights into how to guide students on appropriate genAI use in STEM courses and how to design genAI-based tools to foster students' problem-solving competency.
Verified 2026-11-19 · 1 reference not checked yet
Overview
CoChat Summary
TL;DR
This survey of 40 STEM college students and 28 STEM faculty found high student adoption of generative AI tools, especially free ChatGPT, for STEM coursework. Although students used genAI for explanations and other tasks, over half reported prompting it to directly solve problems, raising concerns that current use may bypass students’ own problem-solving.
Key findings
- GenAI adoption was high: 37 of 40 students reported using genAI tools, with free ChatGPT most common (n = 37), followed by free Gemini (n = 11) and Copilot (n = 10); only 3 students (7.5%) had not used any genAI tools, and 34 students (85%) used them in STEM courses.
- The most common STEM uses were finding explanations (n = 30, 75%), exploring related topics (n = 25, 62.5%), summarizing papers/readings (n = 24, 60%), supporting in-class activities (n = 23, 57.5%), and helping with problem-set questions (n = 23, 57.5%); the main reason for use was saving time (n = 20, 50%).
- In a physics problem prompt task, 54% preferred direct solution generation (38% copy/paste or paraphrase and ask AI to solve; 16% add instructions for AI to solve), while 46% asked specific questions that did not directly lead to a solution.
- Students rated genAI most helpful for explaining relevant physics knowledge (n = 33, 82.5% helpful/highly helpful) and calculations (n = 26, 65%), but were skeptical about collecting missing information (n = 24, 60% unsure/unhelpful/very unhelpful) and verifying solution correctness (n = 28, 70% unsure/unhelpful/very unhelpful); faculty recommendations were lower, with less than one-third recommending explaining concepts (n = 8, 29%) or gathering missing information (n = 9, 32%), and only 2 (7%) for calculations and 3 (11%) for verifying correctness.
Methods
- The study used an online survey developed iteratively with STEM instructor and expert review plus think-aloud pilot testing with college students; the student survey covered usage, prompt engineering skills, perceived helpfulness in problem-solving, perceived benefits and risks, and demographics/academic background.
- Data collection occurred in two phases: 28 faculty were recruited in November 2023 from American Physical Society listserv and Slack channels, and 40 STEM college students were recruited in May 2024 via Prolific, prescreened as current US undergraduates in specified STEM subjects excluding computer science.
- Analyses combined quantitative aggregation in RStudio with qualitative open coding and a coding framework; two researchers independently coded at least 20% of open-ended responses, with mean inter-rater agreement of 83.4% and disagreements resolved through discussion.
Limitations
- The paper states that the small sample of 40 students and 28 instructors limits generalizability, voluntary online participation may introduce selection bias toward students more familiar with or interested in genAI, and the STEM problem-solving focus limits applicability to non-STEM disciplines.
- The text also makes evident that the faculty sample was physics instructors only, while students came from multiple STEM fields, and the study measured self-reported use and perceptions rather than actual learning outcomes.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
8 Valid · 0 Not valid · 12 Not checked · 67 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“These disciplines have traditionally relied on well-defined problems in homework assignments and exams to develop and assess student conceptual understanding and problem-solving competency (Jonassen and Hung, 2015; Montgomery et al, 2024).”
What the sources say
- Not checked — no full text of the sourcecites Jonassen and Hung, 2015 source
- Partly supportedcites Montgomery et al, 2024 sourceExperimental — accuracy not yet measured
The Introduction describes standard physics-course problems as simplified and well structured, and Problems analyzed confirms that the authors examined homework assignments and exams. The Conclusion finds that traditional physics coursework often provides a single well-defined solution path. The paper supports this characterization for physics, but does not establish it across STEM disciplines or specifically say these problems are used to develop and assess conceptual understanding and problem-solving competency. (LLM, conf=0.86)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“These findings provide preliminary evidence that genAI tools, if not used effectively, can short-circuit students' learning process.”
What the sources say
- Not checked — no full text of the sourcecites Tassoti (2024) source
- Supportedcites Ding et al (2023) source
The Discussion reports that almost half the students accepted ChatGPT’s answers whether correct or incorrect, and warns that incorrect information can lead to learning misconceptions. The Implications section also reports that many students blindly trusted its answers. These passages support the qualified claim that ineffective use of GenAI can undermine learning; our copy provides no page numbers. (LLM, conf=0.88)
- Not checked — no full text of the sourcecites Bastani et al (2024) source
hedgedstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Every source the paper cites for this claim was checked and supports it.
In the paper
“Furthermore, even when uses in STEM courses were addressed, the guidance was often superficial, primarily focusing on genAI's roles in supporting coding and writing tasks rather than providing discipline-specific guidance.”
What the sources say
- Supportedcites McDonald et al, 2024 source
Section 4.6 (p. 17) says that even when institutions mention STEM courses, their discussion is frequently superficial and often focuses on coding alongside writing. The Overview (p. 9) likewise describes code and STEM-related guidance as vague, supporting the contrast with discipline-specific guidance. (LLM, conf=0.94)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Students' tendency to use generative AI as a shortcut for direct solutions, rather than as a scaffold for independent problem-solving, mirrors their preference for passive lectures over active learning experiences (Deslauriers et al, 2019).”
What the sources say
- Supportedcites Deslauriers et al, 2019 source
The paper reports that students preferred passive lectures even though they learned more in active classes. Its Materials and Methods section distinguishes being told how to solve problems from trying to solve them independently. The generative-AI comparison is the citing authors’ analogy, not a finding attributed to this source. (LLM, conf=0.89)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: unclear — It is unclear why the paper says it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“However, LLMs have demonstrated high proficiency in solving these types of problems (OpenAI et al, 2024; West, 2023)”
What the sources say
- Not matchedcites OpenAI et al, 2024
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Supportedcites West, 2023 source
The Introduction reports that GPT-4 performed well on homework- and exam-type physics problems, including an AP Physics 2 score in the 66th–84th percentile. In “A. Answer Choices,” the paper also reports 28 correct answers out of 30 on the Force Concept Inventory. These are examples of high proficiency on well-defined STEM assessment problems. (LLM, conf=0.89)
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Not matched
46 more claims
Create a free account to see all 87 claims and every source verdict.
- It may state two separate claims as one. 24
- Probably not a substantial claim. 20
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Wang et al. (2025) discovered that STEM students utilize AI more critically, employing it for data analysis and problem-solving.”
From Beyond Access: Socioeconomic, Academic and Disciplinary Determinants of Generative AI Use in Higher Education (2026) · cites this paper as Wang et al. (2025)
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 6 Limitations · p. 16
“participants came from diverse institutions across the US”
§ 4.1 Participant Demographics · p. 8
“Among faculty participants, 24 were from North America, while 2 each were from Asia and Europe.”
Citations
Cites
Language Models are Few-Shot Learners
Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al. (2020) · arXiv · doi:10.48550/arxiv.2005.14165 · Cited by 23,372
ChatGPT for Good? On Opportunities and Challenges of Large Language Models for Education
Kasneci, E., Seßler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G., Günnemann, S., Hüllermeier, E., et al. (2023) · doi:10.35542/osf.io/5er8f · Cited by 7,552
ChatGPT for good? On opportunities and challenges of large language models for education
Kasneci, E., Sessler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G., Günnemann, S., Hüllermeier, E., et al. (2023) · Learning and Individual Differences · doi:10.1016/j.lindif.2023.102274 · Cited by 7,326 · cited 1×
“Generative AI technologies such as LLMs have the potential to transform teaching and learning practices through automating the creation of instructional materials, enhancing assessment methods, and providing readily available personalized support to students [25].”
Students' voices on generative ai: perceptions, benefits, and challenges in higher education
Chan, C. K. Y., & Hu, W. (2023) · arXiv · doi:10.48550/arxiv.2305.00290 · Cited by 2,504
Students’ voices on generative AI: perceptions, benefits, and challenges in higher education
Chan, C. K. Y., & Hu, W. (2023) · International Journal of Educational Technology in Higher Education · doi:10.1186/s41239-023-00411-8 · Cited by 2,301
Few-Shot Parameter-Efficient Fine-Tuning is Better and Cheaper than In-Context LearningPreprint
Liu, H., Tam, D., Muqeeth, M., Mohta, J., Huang, T., Bansal, M., & Raffel, C. A. (2022) · Neural Information Processing Systems · doi:10.48550/arxiv.2205.05638 · Cited by 1,587
Deslauriers, L., McCarty, L. S., Miller, K., Callaghan, K., & Kestin, G. (2019) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1821936116 · Cited by 1,371
PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization
Zhang, J., Zhao, Y., Saleh, M., & Liu, P. J. (2019) · arXiv · doi:10.48550/arxiv.1912.08777 · Cited by 1,206 · cited 1×
“…and prompting methods [21], [22], these models have evolved beyond next-token prediction to unlocking a wide range of sophisticated capabilities, including engaging in real-time conversations as chatbots [23], summarizing complex texts [24], and generating writings in diverse genres [20].”
Finetuned Language Models Are Zero-Shot Learners
Wei, J., Bosma, M., Zhao, V. Y., Guu, K., Yu, A. W., Lester, B., Du, N., Dai, A. M., & Le, Q. V. (2021) · arXiv · doi:10.48550/arxiv.2109.01652 · Cited by 1,101
The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking
Mueller, P. A., & Oppenheimer, D. M. (2014) · Psychological Science · doi:10.1177/0956797614524581 · Cited by 1,051
Research works printed but not linked: 7
- 2024, Report of the NEA task force on artificial intelligence in education
- Shiohira, 2021, Understanding the impact of artificial intelligence on skills development. education 2030, UNESCO-UNEVOC International Centre for Technical and Vocational Education and Training
- Honey, Stem education for the future - 2020 visioning report
- 2021, Criteria for accrediting engineering programs, 2022–2023
- Levy, 2013, Dancing with robots: Human skills for computerized work
- Salehi, 2018, Improving problem-solving through reflection
- Goldberg, Generative ai in higher education: Insights from a campus-wide student survey at a large public university, 757
Cited by
Generative AI in Pre-Service Science Teacher Education: A Systematic Review
Spasopoulos, T., Sotiropoulos, D. J., & Kalogiannakis, M. (2025) · Advances in Mobile Learning Educational Research · doi:10.25082/amler.2025.02.007 · Cited by 52
Tsakeni, M., Nwafor, S. C., Mosia, M. S., & Egara, F. O. (2025) · Journal of Intelligence · doi:10.3390/jintelligence13110148 · Cited by 83
Fiedler, A., & Döpke, J. (2025) · International Review of Economics Education · doi:10.1016/j.iree.2025.100321 · Cited by 59
Ncube, P. D. N., Dzvapatsva, G. P., Matobobo, C., & Ranga, M. M. (2026) · AI and Ethics · doi:10.1007/s43681-025-00871-w · Cited by 31
“Furthermore, some scholars have called for the expansion of AI literacy training as a means of addressing the challenges posed by inappropriate or uninformed AI use [ 73, 75, 86, 95, 101].”
Luo, Y. T., Liu, T., Pang, P. C., Wang, Z., & Chan, K. I. (2025) · Applied Sciences · doi:10.3390/app15137515 · Cited by 18
Metadata
- Authors
- Wang, K. D., Wu, Z., Tufts, L., Wieman, C., Salehi, S., & Haber, N.
- Affiliations
- Graduate School of Education
- Department of Physics
- Stanford University (US)
- Venue
- 2025 IEEE Global Engineering Education Conference (EDUCON)
- Publisher
- IEEE
- Published
- 2025-04-22
- Pages
- 1-10
- OpenAlex
- W4410986723
- Cited by
- 28
- References
- 33
- Citations by year
- 2026: 19 · 2025: 9
- FWCI
- 32.02
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- IEEE Global Engineering Education Conference (EDUCON) (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Social Sciences › Education › Artificial Intelligence in Education
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Social Sciences › Education › Problem and Project Based Learning
- Keywords
- generative AI, ChatGPT, STEM education, problem-solving competencies, university students, student use of AI tools, teacher perceptions, AI adoption in higher education, student motivation, AI-based learning tools
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-26, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-26 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Scaffold or Crutch? Examining College Students' Use and Views of Generative AI Tools for STEM Education. https://app.cochat.ai/papers/10.1109/educon62633.2025.11016406
- The paper
- Wang, K. D., Wu, Z., Tufts, L., Wieman, C., Salehi, S., & Haber, N. (2025). Scaffold or Crutch? Examining College Students' Use and Views of Generative AI Tools for STEM Education. 2025 IEEE Global Engineering Education Conference (EDUCON). https://doi.org/10.1109/educon62633.2025.11016406
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.