Evaluation of Few-Shot AI-Generated Feedback on Case Reports in Physical Therapy Education: Mixed Methods Study
Bibliographic record
Abstract
BACKGROUND: While artificial intelligence (AI)-generated feedback offers significant potential to overcome constraints on faculty time and resources associated with providing personalized feedback, its perceived usefulness can be undermined by algorithm aversion. In-context learning, particularly the few-shot approach, has emerged as a promising paradigm for enhancing AI performance. However, there is limited research investigating its usefulness, especially in health profession education. OBJECTIVE: This study aimed to compare the quality of AI-generated formative feedback from 2 settings, feedback generated in a zero-shot setting (hereafter, "zero-shot feedback") and feedback generated in a few-shot setting (hereafter, "few-shot feedback"), using a mixed methods approach in Japanese physical therapy education. Additionally, we examined the effect of algorithm aversion on these 2 feedback types. METHODS: A mixed methods study was conducted with 35 fourth-year physical therapy students (mean age 21.4, SD 0.7 years). Zero-shot feedback was created using Gemini 2.5 Pro with default settings, whereas few-shot feedback was generated by providing the same model with 9 teacher-created examples. The participants compared the quality of both feedback types using 3 methods: a direct preference question, the Feedback Perceptions Questionnaire (FPQ), and focus group interviews. Quantitative comparisons of FPQ scores were performed using the Wilcoxon signed rank test. To investigate algorithm aversion, the study examined how student perceptions changed before and after disclosure of the feedback's identity. RESULTS: Most students (26/35, 74%) preferred few-shot feedback over zero-shot feedback in terms of overall usefulness, although no significant difference was found between the 2 feedback types for the total FPQ score (P=.22). On the specific FPQ scales, few-shot feedback scored significantly higher than zero-shot feedback on fairness across all 3 items: "satisfied" (P=.02; r=0.407), "fair" (P=.04; r=0.341), and "justified" (P=.02; r=0.392). It also scored significantly higher on 1 item of the usefulness scale ("useful"; P=.02; r=0.401) and 1 item of the willingness scale ("invest a lot of effort"; P=.02; r=0.394). In contrast, zero-shot feedback scored significantly higher on the affect scale across 2 items: "successful" (P=.03; r=0.365) and "angry" (P=.008; r=0.443). Regarding algorithm aversion, evaluations for zero-shot feedback became more negative for 83% (15/18) of the items after identity disclosure, whereas positive perceptions of few-shot feedback were maintained or increased. Qualitative analysis revealed that students valued zero-shot feedback for its encouraging tone, whereas few-shot feedback was appreciated for its contextual understanding and concrete guidance for improvement. CONCLUSIONS: Japanese physical therapy students perceived few-shot feedback more favorably than zero-shot feedback on case reports. This few-shot AI model shows potential to resist algorithm aversion and serves as an effective educational tool to support autonomous writing, facilitate reflection on clinical reasoning, and cultivate advanced thinking skills.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.060 | 0.105 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.003 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.001 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".