Insight SFI Research Centre and industry partner Genesys have published a joint paper entitled Towards a Semantic Approach for Linked Model and Data Cards in Dataspaces. The paper explores ways of supporting explainability of AI models in an effort for more trustworthy AI. The project is part of the EMPOWER Data Governance programme.

Director of Insight at the University of Galway Professor Ed Curry said, “We are developing a strong collaboration with Genesys to research tools and methods to enable the creation of trustworthy AI systems. For example, Linked Model and Data Cards can support the explainability and fairness of data-driven AI models.”

In a statement, Genesys said: “At Genesys we’re committed to responsible AI development, with security, privacy and safety key principles. In collaboration with Insight SFI Research Centre for Data Analytics at University of Galway, we’re proposing a flexible and interoperable data model to document AI/machine learning datasets, models, services, and dataspaces. Through this work, our goal is to facilitate the governance of this evolving field and help ensure the ethical use of data and AI applications across our business.”

The paper’s abstract states: “The vast majority of artificial intelligence practitioners overlook the importance of documentation when building and publishing models and datasets. However, due to the recent trend in the explainability and fairness of AI models, several frameworks have been proposed such as Model Cards, and Data Cards, among others, to help in the appropriate re-usage of those models and datasets. In addition, because of the introduction of the dataspace concept for similar datasets in one place, there is potential that similar Model Cards, Data Cards, Service Cards, and Dataspace Cards can be linked to extract helpful information for better decision-making about which model and data can be used for a specific application. This paper reviews the case for considering a Semantic Web approach for exchanging Model/Data Cards as Linked Data or knowledge graphs in a dataspace, making them machine-readable. We discuss the basic concepts and propose a schema for linking Data Cards and Model Cards within a dataspace. In addition, we introduce the concept of a dataspace card which can be a starting point for extracting knowledge about models and datasets in a dataspace. This helps in building trust and reuse of models and data among companies and individuals participating as publishers or consumers of such assets.”

You can read the full article on the ACM Digital Library