Assessing the Accuracy of Generative Conversational Artificial Intelligence in Debunking Sleep Health Myths: Mixed Methods Comparative Study With Expert Analysis
Bibliographic record
Abstract
BACKGROUND: Adequate sleep is essential for maintaining individual and public health, positively affecting cognition and well-being, and reducing chronic disease risks. It plays a significant role in driving the economy, public safety, and managing health care costs. Digital tools, including websites, sleep trackers, and apps, are key in promoting sleep health education. Conversational artificial intelligence (AI) such as ChatGPT (OpenAI, Microsoft Corp) offers accessible, personalized advice on sleep health but raises concerns about potential misinformation. This underscores the importance of ensuring that AI-driven sleep health information is accurate, given its significant impact on individual and public health, and the spread of sleep-related myths. OBJECTIVE: This study aims to examine ChatGPT's capability to debunk sleep-related disbeliefs. METHODS: A mixed methods design was leveraged. ChatGPT categorized 20 sleep-related myths identified by 10 sleep experts and rated them in terms of falseness and public health significance, on a 5-point Likert scale. Sensitivity, positive predictive value, and interrater agreement were also calculated. A qualitative comparative analysis was also conducted. RESULTS: ChatGPT labeled a significant portion (n=17, 85%) of the statements as "false" (n=9, 45%) or "generally false" (n=8, 40%), with varying accuracy across different domains. For instance, it correctly identified most myths about "sleep timing," "sleep duration," and "behaviors during sleep," while it had varying degrees of success with other categories such as "pre-sleep behaviors" and "brain function and sleep." ChatGPT's assessment of the degree of falseness and public health significance, on the 5-point Likert scale, revealed an average score of 3.45 (SD 0.87) and 3.15 (SD 0.99), respectively, indicating a good level of accuracy in identifying the falseness of statements and a good understanding of their impact on public health. The AI-based tool showed a sensitivity of 85% and a positive predictive value of 100%. Overall, this indicates that when ChatGPT labels a statement as false, it is highly reliable, but it may miss identifying some false statements. When comparing with expert ratings, high intraclass correlation coefficients (ICCs) between ChatGPT's appraisals and expert opinions could be found, suggesting that the AI's ratings were generally aligned with expert views on falseness (ICC=.83, P<.001) and public health significance (ICC=.79, P=.001) of sleep-related myths. Qualitatively, both ChatGPT and sleep experts refuted sleep-related misconceptions. However, ChatGPT adopted a more accessible style and provided a more generalized view, focusing on broad concepts, while experts sometimes used technical jargon, providing evidence-based explanations. CONCLUSIONS: ChatGPT-4 can accurately address sleep-related queries and debunk sleep-related myths, with a performance comparable to sleep experts, even if, given its limitations, the AI cannot completely replace expert opinions, especially in nuanced and complex fields such as sleep health, but can be a valuable complement in the dissemination of updated information and promotion of healthy behaviors.
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 | 0.152 | 0.274 |
| Meta-epidemiology (narrow) | 0.001 | 0.002 |
| Meta-epidemiology (broad) | 0.002 | 0.003 |
| Bibliometrics | 0.005 | 0.002 |
| Science and technology studies | 0.003 | 0.003 |
| Scholarly communication | 0.005 | 0.004 |
| Open science | 0.003 | 0.005 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.003 | 0.001 |
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".