Usage of AI Methods

Summary

To provide transparency, the usage of AI tools and automation processes in the development process should be documented.

Purpose

The usage of and opportunities provided by AI methods, especially LLMs, are fastly increasing. The development of reference ontologies can be supported by such tools, however, as being a community based consensus creation process¹ , it can hardly be fully automated. Reference ontologies are intended to represent general true statements about the world, which contradicts the susceptibility of LLMs to hallucinations. To provide transparency, the usage of AI tools should be documented.

Implementation

The usage of AI tools in the development process MUST be documented, see also the documentation principle.

• If it refers to generic tasks this SHOULD be documented externally.
• If it refers to specific entities, this SHOULD be documented in situ per notation.

Examples

• ChEBI provides each entity with a star rating. Entities that are created externally, including automation tools, receive two stars. Only entities that were human-curated by a ChEBI editor receive three stars, see also https://www.ebi.ac.uk/training/online/courses/chebi-the-online-chemical-dictionary-for-small-molecules/what-is-chebi/where-does-the-data-come-from/

---

1. F. Neuhaus, J. Hasitings: Ontology development is consensus creation, not (merely) representation, Applied Ontology. 2022;17(4):495-513. doi:10.3233/AO-220273 ↩