THEMIS 5.0 Approaches and Conceptual Modeling (for Trustworthy AI)

This report outlines the conceptual modelling of the THEMIS 5.0 trustworthiness optimization framework to assess and enhance the trustworthiness of an AI System. The modelling approach is designed to align with human-centered and risk management principles to support the decision-making process of the system’s AI actors.  

To achieve this objective, the framework integrates the Trustworthiness Optimization Process (TOP) and risk management approach. Each of these components is briefly introduced, followed by a demonstration of how they work in conjunction. A key advantage of this combination is the ability to adapt based on the specific context and requirements of the System Under Test (SUT). Realizing such adaptability requires the collection of necessary information about the system through the use of documentation cards. A total of six different documentation cards are considered, each with a unique purpose and tailored to capture unique aspects of the AI System’s socio-technical environment. 

A more detailed description of the TOP methodology and its stages is provided. Each stage is divided into sub-processes to clarify the steps involved, along with an overview of how AI actors, documentation cards, and related assets are employed through these stages.  

Furthermore, the document analyses the conceptual architecture of the THEMIS 5.0 framework, which builds upon the defined framework and the respective notions described. The architecture serves as a blueprint to illustrate how the underlying decision processes work collaboratively to assess and enhance the trustworthiness of an AI system. Moreover, the decision processes and the assessment of legal compliance of the THEMIS platform are described in more detail. 

Finally, the deliverable concludes by examining its relevance to other WPs and workflows of the project. It also offers some key insights about the socio-technical environment and continued lifecycle development of the AI System, the aspects of human-centered and risk assessment as well as the ethical and legal considerations.   

______________________________________________________________________________

THEMIS 5.0 has received funding from the EU Horizon Europe Research and Innovation programme under grant agreement No. 101121042, and from UK Research and innovation under UK government's Horizon funding guarantee.