Use of Artificial Intelligence in Scientific Writing. The Danger of Trying Too Hard to Please
Bibliographic record
Abstract
The author describes his experience using several artificial intelligence programs to assist in the process of editing or preparing manuscripts for publication. While the programs were very useful to increase clarity and optimize the English of previously written text, problems arose when asked to review the literature to describe or expand on concepts, and then to cite references to support those statements. Citations were sometimes fabricated. The artificial intelligence program would even provide PubMed identification (PMID) numbers for references cited, which sometimes pointed to completely different publications. In references that were correctly cited, a request to extract data from their abstracts yielded data that were completely fabricated or incorrect. The newer versions of these artificial intelligence programs appear to be enormously helpful in helping with the process of scientific writing, but one needs to assiduously verify every statement made with regard to their interpretation of the medical literature and double-check any citations by retrieving and reading those references. Because statements made by these agents are proffered with great confidence and using excellent writing skills, the resulting errors can be difficult to anticipate. The motivations behind such errors are unknown, but appear to be related to a desire of AI to please the user at any cost. Programmers who create these tools somehow managed to allow such errors, and this issue must be addressed in future versions of AI. Also, it would help greatly if AI had access to the full text of medical scientific articles, which might be achieved by contractual agreements with the main medical publishers.
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 | 0.078 | 0.163 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.006 | 0.005 |
| Science and technology studies | 0.006 | 0.029 |
| Scholarly communication | 0.028 | 0.020 |
| Open science | 0.003 | 0.014 |
| Research integrity | 0.006 | 0.012 |
| Insufficient payload (model declined to judge) | 0.008 | 0.008 |
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".