Skip to main content
ChatGPT may not be able to communicate authoritatively regarding scientific interpretation because it does not have a vocabulary specific to scientific knowledge. Natural language processing is not equivalent to scientific expression.
We will need to carefully separate the scientific knowledge (conventionally represented in Methods and Results sections) from the natural language "persuasive" arguments about the scientific knowledge (conventionally represented in Introduction and Discussion sections).
To invite ChatGPT (or other entities of its species) to participate in scientific communication, we need to establish a shared language with a machine-interpretable representation of scientific knowledge.
Other issues, like inventing citations and evidence that do not exist[1], will need to be addressed as well, but a computable expression of science is a step that could enable AI participation in scientific discourse.
A machine-interpretable representation of scientific knowledge is being made possible by the extension of Fast Healthcare Interoperability Resources (FHIR, the standard for health data exchange) to evidence-based medicine (EBM) knowledge assets with the EBMonFHIR project.[2,3] Participation is open to all through the Health Evidence Knowledge Accelerator (HEvKA, formerly COVID-19 Knowledge Acclerator [COKA])[4,5] with details at https://fevir.net/HEvKA
1. Kim SG. Using ChatGPT for language editing in scientific articles. Maxillofac Plast Reconstr Surg. 2023 Mar 8;45(1):13. doi: 10.1186/s40902-023-00381-x. PMID: 36882591; PMCID: PMC9992464.
2. Alper BS. EBMonFHIR-based tools and initiatives to support clinical research. J Am Med Inform Assoc. 2022 Dec 13;30(1):206-207. doi: 10.1093/jamia/ocac193. PMID: 36228125; PMCID: PMC9748541.
3. Lichtner G, Alper BS, Jurth C, Spies C, Boeker M, Meerpohl JJ, von Dincklage F. Representation of evidence-based clinical practice guideline recommendations on FHIR. J Biomed Inform. 2023 Mar;139:104305. doi: 10.1016/j.jbi.2023.104305. Epub 2023 Feb 3. PMID: 36738871.
4. Alper BS, Richardson JE, Lehmann HP, Subbian V. It is time for computable evidence synthesis: The COVID-19 Knowledge Accelerator initiative. J Am Med Inform Assoc. 2020 Aug 1;27(8):1338-1339. doi: 10.1093/jamia/ocaa114. PMID: 32442263; PMCID: PMC7313978.
5. Alper BS, Dehnbostel J, Afzal M, Subbian V, Soares A, Kunnamo I, Shahin K, McClure RC; COVID-19 Knowledge Accelerator (COKA) Initiative. Making science computable: Developing code systems for statistics, study design, and risk of bias. J Biomed Inform. 2021 Mar;115:103685. doi: 10.1016/j.jbi.2021.103685. Epub 2021 Jan 21. PMID: 33486066; PMCID: PMC9387176.