Can we use ChatGPT for Mental Health and Substance Use Education? Examining Its Quality and Potential Harms
Bibliographic record
Abstract
BACKGROUND: The use of generative artificial intelligence, more specifically large language models (LLMs), is proliferating, and as such, it is vital to consider both the value and potential harms of its use in medical education. Their efficiency in a variety of writing styles makes LLMs, such as ChatGPT, attractive for tailoring educational materials. However, this technology can feature biases and misinformation, which can be particularly harmful in medical education settings, such as mental health and substance use education. This viewpoint investigates if ChatGPT is sufficient for 2 common health education functions in the field of mental health and substance use: (1) answering users' direct queries and (2) aiding in the development of quality consumer educational health materials. OBJECTIVE: This viewpoint includes a case study to provide insight into the accessibility, biases, and quality of ChatGPT's query responses and educational health materials. We aim to provide guidance for the general public and health educators wishing to utilize LLMs. METHODS: We collected real world queries from 2 large-scale mental health and substance use portals and engineered a variety of prompts to use on GPT-4 Pro with the Bing BETA internet browsing plug-in. The outputs were evaluated with tools from the Sydney Health Literacy Lab to determine the accessibility, the adherence to Mindframe communication guidelines to identify biases, and author assessments on quality, including tailoring to audiences, duty of care disclaimers, and evidence-based internet references. RESULTS: GPT-4's outputs had good face validity, but upon detailed analysis were substandard in comparison to expert-developed materials. Without engineered prompting, the reading level, adherence to communication guidelines, and use of evidence-based websites were poor. Therefore, all outputs still required cautious human editing and oversight. CONCLUSIONS: GPT-4 is currently not reliable enough for direct-consumer queries, but educators and researchers can use it for creating educational materials with caution. Materials created with LLMs should disclose the use of generative artificial intelligence and be evaluated on their efficacy with the target audience.
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 distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 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 teacher head, 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".