Paying the Cognitive Debt: An Experiential Learning Framework for Integrating AI in Social Work Education
Open accessThe DOI matches its Crossref record; 27 of the 27 references it cites were found in a registry or an index; 12 of the 62 sources its claims cite were checked for support.
Abstract
The rapid integration of Generative Artificial Intelligence in higher education challenges social work as student adoption outpaces pedagogical guidance. This paper argues that the unguided use of AI fosters cognitive debt: a cumulative deficit in critical thinking, ethical reasoning, and professional judgment that arises from offloading cognitive tasks. To counter this risk, a pedagogical model is proposed, synthesizing experiential learning, andragogy, and critical pedagogies. The framework reframes AI from a passive information tool into an active object of critical inquiry. Through structured assignments across micro, mezzo, and macro practice, the model guides students through cycles of concrete experience with AI, reflective observation of its biases, abstract conceptualization of ethical principles, and active experimentation with responsible professional use. Aligned with professional ethical standards, the model aims to prepare future social workers to scrutinize and shape AI as a tool for social justice. The paper concludes with implications for faculty development, institutional policy, accreditation, and a forward-looking research agenda.
Overview
CoChat Summary
TL;DR
This conceptual paper argues that unguided student use of generative AI in social work education produces "cognitive debt"—a cumulative deficit in critical thinking, ethical reasoning, and professional judgment caused by cognitive offloading and automation bias. It proposes a pedagogical framework synthesizing Kolb's experiential learning cycle, Knowles' andragogy, and the critical pedagogies of Freire and hooks, operationalized through micro, mezzo, and macro practice assignments that treat AI as an object of critical inquiry rather than an answer machine.
Key findings
As a conceptual paper, it reports no original empirical results; its claims are theoretical and literature-based.
- Student AI adoption far outpaces faculty and institutional readiness: a 2024 global survey indicates 86% of students use AI tools in academic work, while 80% of faculty report a lack of clarity on how to apply AI in teaching and only 4% feel their institution's AI guidelines are fully comprehensive (Digital Education Council, 2025b, as cited).
- The paper redefines "cognitive debt," borrowed from Marchant and Howard's (2015) Alzheimer's research, as an educational opportunity cost accrued through the absence of effortful thinking, deep processing, and schema-building—driven by cognitive offloading and automation bias (errors of commission and omission).
- It proposes a four-stage Kolb cycle (Concrete Experience → Reflective Observation → Abstract Conceptualization → Active Experimentation) applied to AI assignments across micro (simulated intake interview), mezzo (agency data privacy policy), and macro (community needs assessment) practice, with TPACK as strategic filter and SAMR as tactical toolkit.
- It argues the framework aligns with CSWE EPAS competencies, the NASW Code of Ethics, and UNESCO's AI ethics principles, and calls for a research agenda including quasi-experimental pre-test/post-test designs and longitudinal tracking of graduates.
Methods
Conceptual/theoretical paper: it synthesizes foundational education and social work theory (Kolb, Knowles, Freire, hooks, TPACK, SAMR) with a review of emerging empirical and ethical literature on AI to propose a novel pedagogical model; the author states the goal "is not to present new empirical data."
- No sample, dataset, or statistical analysis; the paper states "No new data were created or analyzed in this study."
- Illustrative assignment designs and two tables (Table 1 mapping CSWE competencies to Kolb stages; Table 2, a dimensional risk assessment framework for AI use cases) are presented as practical guidance, not tested interventions.
Limitations
Stated by the paper: (1) it is a conceptual model requiring empirical validation, so its effectiveness is "at present, theoretical"; (2) successful implementation hinges on significant institutional investment in faculty development, and barriers including faculty resistance and resource constraints "are substantial and cannot be overlooked"; (3) the framework is intentionally flexible and may require significant adaptation across diverse program contexts; (4) the professional standards and examples are primarily rooted in the United States (e.g., CSWE, NASW), so future work should adapt and test it internationally.
- Evident from the text: the empirical support cited for the framework's efficacy (e.g., Cong-Lem et al., 2025; Flaherty et al., 2025) comes from other contexts and populations, not from social work education, and the paper's own tables are illustrative rather than validated instruments.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
11 Valid · 1 Not valid · 39 Not checked · 36 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“a 2024 global survey indicates that 86% are now using AI tools in their academic work (Sharma et al., 2025).”
What the sources say
- Supportedcites Sharma et al., 2025 source
The Abstract reports that a recent global survey found 86% of students already use AI tools for tasks such as information search and summarization. The References identify a 2024 report of a global student survey with the same 86% finding. The paper reports the survey; it does not claim to have conducted it. (LLM, conf=0.91)
hedged- 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
“This bias manifests in two distinct error types: errors of commission, where a student accepts and incorporates incorrect AI-generated information into their work, and errors of omission, where a student fails to identify a critical issue or take necessary action because the AI did not prompt them to do so (Kahn et al., 2024).”
What the sources say
- Supportedcites Kahn et al., 2024 source
The “What Is Automation Bias?” section explicitly says automation bias most often presents as errors of omission, when a person fails to act because automation did not alert them, or errors of commission, when a person follows incorrect directions from automation. The citing sentence applies those definitions to a student using AI. (LLM, conf=0.91)
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
“Recent empirical research has confirmed that even a short-term intervention explicitly based on Kolb's cycle can produce statistically significant and large improvements in students' critical thinking skills when using AI (Cong-Lem et al., 2025).”
What the sources say
- Supportedcites Cong-Lem et al., 2025 source
The Conclusion describes a 90-minute intervention that operationalized Kolb’s experiential learning cycle and significantly enhanced EFL learners’ critical thinking when using generative AI. The Abstract reports large effects across critical-thinking subdomains (Cohen’s d = 0.94–1.23) and a significant advantage over the control group. The finding is specific to EFL learners; page numbers were not provided in our copy. (LLM, conf=0.93)
boostedstates 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.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“It is likely that AI will automate many routine administrative and data analysis tasks (Nuwasiima et al., 2024).”
What the sources say
- Supportedcites Nuwasiima et al., 2024 source
Section 1.4 says AI can automate routine data-entry and case-management tasks and streamline analysis of client data and case notes. Figure 1 likewise describes AI and machine learning as automating routine tasks and analysing large datasets in social work. (LLM, conf=0.92)
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.
- Supported
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This disparity ensures that the default mode of AI use is driven by the path of least cognitive resistanceefficiency and task completion-rather than by pedagogical goals.”
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: unclear — It is unclear why the paper says it holds.
30 more claims
Create a free account to see all 87 claims and every source verdict.
- It may state two separate claims as one. 25
- Probably not a substantial claim. 37
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
§ 1. Introduction
“A 2025 global survey found that while a majority of faculty have used AI in teaching, their use is overwhelmingly minimal (Digital Education Council, 2025b).”
§ References
“- Digital Education Council. (2025b, March 3). DEC AI literacy framework . Available online: https://www.digitaleducationcouncil.com/ post/digital-education-council-ai-literacy-framework (accessed on 23 July 2025).”
Citations
Cites
Technological Pedagogical Content Knowledge: A Framework for Teacher Knowledge
Mishra, P., & Koehler, M. J. (2006) · Teachers College Record: The Voice of Scholarship in Education · doi:10.1111/j.1467-9620.2006.00684.x · Cited by 12,604
Cognitive Load During Problem Solving: Effects on Learning
Sweller, J. (1988) · Cognitive Science · doi:10.1016/0364-0213(88)90023-7 · Cited by 11,048
Survey of Hallucination in Natural Language Generation
Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., & Fung, P. (2023) · ACM Computing Surveys · doi:10.1145/3571730 · Cited by 5,025
Risko, E. F., & Gilbert, S. J. (2016) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2016.07.002 · Cited by 1,502
Sisk, V. F., Burgoyne, A. P., Sun, J., Butler, J. L., & Macnamara, B. N. (2018) · Psychological Science · doi:10.1177/0956797617739704 · Cited by 1,159
Walter, Y. (2024) · International Journal of Educational Technology in Higher Education · doi:10.1186/s41239-024-00448-3 · Cited by 1,113
Do Learners Really Know Best? Urban Legends in Education
Kirschner, P. A., & van Merriënboer, J. J. G. (2013) · Educational Psychologist · doi:10.1080/00461520.2013.804395 · Cited by 655
ChatGPT and the rise of generative AI: Threat to academic integrity?
Eke, D. O. (2023) · Journal of Responsible Technology · doi:10.1016/j.jrt.2023.100060 · Cited by 532
Gkintoni, E., Antonopoulou, H., Sortwell, A., & Halkiopoulos, C. (2025) · Brain Sciences · doi:10.3390/brainsci15020203 · Cited by 309
Mollick, E. R., & Mollick, L. (2023) · doi:10.2139/ssrn.4391243 · Cited by 278
Research works printed but not linked: 7
- Easter, 2024, Artificial intelligence in language education: A mixed-methods study of teacher perspectives and challenges, National Social Science Technology Journal, 12, 203
- hooks, b. (1994). Teaching to transgress: Education as the practice of freedom, Taylor & Francis Group.
- Knowles, M. S. (1980). The modern practice of adult education: From pedagogy to andragogy, Cambridge Book Company. [2nd ed.].
- Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X. H., Beresnitzky, A. V., Braunstein, I., and Maes, P. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task. arXiv.
- Ng, B. Y., Li, J., Tong, X., Ye, K., Yenne, G., Chandrasekaran, V., and Li, J. (2025). Analyzing security and privacy challenges in generative AI usage guidelines for higher education. arXiv.
- Parveen, 2024, Cognitive offloading: A review, The International Journal of Indian Psychology, 12, 4668
- Petricini, 2024, Perceptions about artificial intelligence and ChatGPT use by faculty and students, Transformative Dialogues: Teaching and Learning Journal, 17, 63
Cited by
Scheepers, S. J., & Stott, A. E. (2026) · Education Sciences · doi:10.3390/educsci16040573 · Cited by 4
Wang, Y., 楊昆岳, & HAN, D. (2026) · Frontiers in Psychology · doi:10.3389/fpsyg.2026.1831770 · Cited by 4
“On the other hand, technological convenience may induce cognitive offloading and intellectual inertia, undermining deep learning engagement and learner agency (Watts, 2025; Romeo and Conti, 2025).”
InfluentialHuang, D.-H., & Wang, Y.-C. (2026) · Information · doi:10.3390/info17010107 · Cited by 3
Susena, Y. S., Arianto, F., Khotimah, K., Andrianto, R. E., & Syuhada’, M. (2026) · Reading Psychology · doi:10.1080/02702711.2026.2644973 · Cited by 1
“Reviews of AI-enabled learning environments similarly conclude that normalization without pedagogical safeguards risks prioritizing speed and output quality over deep processing, epistemic struggle, and critical scrutiny (Raitskaya & Tikhonova, 2025; Watts, 2025).”
Sabar, F. C., Taneo, R. F. S., Alemoka, C. K., Mai, A. Y., Walalayo, D. Y., Bumen, S., Jazila, S., Peea, E. E., Nanda, N., Cu'la, H., et al. (2026) · F1000Research · doi:10.12688/f1000research.190540.1
Metadata
- Authors
- Watts, K. J.
- Affiliations
- University of Kentucky (US)
- Corresponding author
- Watts, Keith J.
- ORCID
- Watts, Keith J.
- Venue
- Education Sciences
- Publisher
- MDPI AG
- Published
- 2025-10-02
- Volume
- 15
- Issue
- 10
- Pages
- 1304
- ISSN
- 2227-7102 (ISSN-L)
- OpenAlex
- W4414746592
- Cited by
- 18
- References
- 23
- Citations by year
- 2026: 18
- FWCI
- 29.49
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Education Sciences (publishedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Topics
- Social Sciences › Public Administration › Social Work Education and Practice
- Social Sciences › Education › Artificial Intelligence in Education
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Keywords
- generative AI, social work education, cognitive debt, experiential learning, andragogy, critical pedagogy, AI ethics, critical thinking, professional judgment, algorithmic bias, micro, mezzo, and macro practice, faculty development, social justice, accreditation
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 2026-10-05, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-05 (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
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.