Is health technology assessment ready for generative pretrained transformer large language models? Report of a fishbowl inquiry
Bibliographic record
Abstract
OBJECTIVES: The Health Technology Assessment International (HTAi) 2023 Annual Meeting included a novel "fishbowl" session intended to 1) probe the role of HTA in the emergence of generative pretrained transformer (GPT) large language models (LLMs) into health care and 2) demonstrate the semistructured, interactive fishbowl process applied to an emerging "hot topic" by diverse international participants. METHODS: The fishbowl process is a format for conducting medium-to-large group discussions. Participants are separated into an inner group and an outer group on the periphery. The inner group responds to a set of questions, whereas the outer group listens actively. During the session, participants voluntarily enter and leave the inner group. The questions for this fishbowl were: What are current and potential future applications of GPT LLMs in health care? How can HTA assess intended and unintended impacts of GPT LLM applications in health care? How might GPT be used to improve HTA methodology? RESULTS: Participants offered approximately sixty responses across the three questions. Among the prominent themes were: improving operational efficiency, terminology and language, training and education, evidence synthesis, detecting and minimizing biases, stakeholder engagement, and recognizing and accounting for ethical, legal, and social implications. CONCLUSIONS: The interactive fishbowl format enabled the sharing of real-time input on how GPT LLMs and related disruptive technologies will influence what technologies will be assessed, how they will be assessed, and how they might be used to improve HTA. It offers novel perspectives from the HTA community and aligns with certain aspects of ongoing HTA and evidence framework development.
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 | 0.208 | 0.242 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.002 |
| Science and technology studies | 0.009 | 0.023 |
| Scholarly communication | 0.011 | 0.017 |
| Open science | 0.004 | 0.021 |
| Research integrity | 0.005 | 0.011 |
| Insufficient payload (model declined to judge) | 0.005 | 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; the direct Gemma label and the distilled Codex classifier agree on what is shown here.
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".