The term "Data-Centric AI" (DCAI) suggests that the next generation of robust, fair, and responsible AI systems with sustainable performance requires an iterative, human-in-the-loop process involving complex patterns of data selection, engineering, and curation, model training, model testing, deployment and monitoring.
In this talk, we explore this perspective, and suggest that providing infrastructure to systematically record the end-to-end, data-to-AI patterns in detail is key to achieving all the -ilities we can dream of: explainability, reproducibility, accountability, sustainability...
This is part of AIM RSF Open Invitation seminar series and took place in May.
The slides can be accessed: https://zenodo.org/records/11197109