Big Graphs, AI Black-boxes, And Human-out-of-the-Loop: Regaining Human FAIRness Through Human-Centric Curation And Preservation Of Intelligent Artefacts

The acknowledgement of the role played by machines in production and discovery of knowledge has underpinned the FAIR principles as a key-tenet since their inception. At the core of the principles lie the concepts of machine-readability and machine-actionability, and the pledge to satisfy, in a digital research environment, the specific information needs and information behaviours of machines, considered as e-Science ‘stakeholders’ and primary data users. Given the extensive upstream and downstream deployment of artificial intelligence systems in the current e-research landscape, however, a FAIR machine-centric approach risks to contribute to the ongoing decentring of human-agents in knowledge discovery processes. The need to mitigate such decentring has kindled the recent plea for a FAIREr approach to actionability of data and information in automated knowledge environments (Vogt 2023; Vogt et al. 2024), to deliver systems pivoting on the idea of “cognitive interoperability” (Naudet et al. 2023) between machine- and human-users. This lightning talk will discuss the properties of cognitively interoperable artificial intelligence artefacts (openness, transparency, explainability) and the challenging factors that impair them. An overview of computational stewardship approaches and tools that promise to support cognitive interoperability will be presented and assessed against standardised data curation and preservation practices.