Even though the maturity of the Artificial Intelligence (AI) technologies is rather advanced nowadays, according to McKinsey1, its adoption, deployment and application is not as wide as it could be expected. This could be attributed to many barriers including cultural ones, but above all, the lack of trust of potential users in such AI systems. The diferent factors that afect the users' trustworthiness on AI systems were studied in [ according to [2] refers to the “techniques that enable human users to understand, appropriately trust, and efectively manage the emerging generation of artificially intelligent partners”. However, the explainability of AI systems is necessary but far from suficient for understanding them and holding them accountable [3]. Therefore, in order to develop trustworthy AI systems, not only should they be explainable, but also accountable.
CEUR Workshop Proceedings
The first phase is related to the development of the predictive model and its deployment in production where it will be executed. In the second phase both the procedure followed to develop the deployed predictive model and the results produced by the predictive model are annotated with the adequate ontology terms. As for the third phase, it is responsible for managing the annotations of the previous phase and facilitating their exploitation by users via SPARQL queries.
The topics that can be identified in Machine Learning system and that may be of interest to annotate with ontologies are: the forecast made by the predictive model, and the procedure followed for making such a forecast. Likewise, the latter procedure-related information can be divided in the information that addresses the training data and the information concerning the predictive model itself. After considering and evaluating the suitability of diferent ontologies, finally, three Ontology Design Patterns (the AfectedBy ODP 2, the Execution-Executor-Procedure (EEP) ODP3 and the Result-Context (RC) ODP4) and the ML-Schema5 have been chosen for representing this knowledge.
The full potential of Semantic Technologies to fill existing gaps and unsolved challenges towards trustworthy AI systems is yet to be unlocked. This article is aimed at paving the way for future research in this direction.
This work is partly supported by the project 3KIA (KK-2020/00049), funded by the SPRI-Basque Government through the ELKARTEK program and the AI-PROFICIENT project which has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement no. 957391 2https://w3id.org/affectedBy 3https://w3id.org/eep 4https://w3id.org/rc 5http://www.w3.org/ns/mls