Results achieved

During his secondment at Nostrum (Spain) in June/July 2026, Fotis Psomopoulos, Senior Researcher from INAB|CERTH (Greece), collaborated with researchers across the AHEAD consortium to push forward the implementation framework that would connect the collected bias indicators for AI, to the benchmarking infrastructure that would allow for the direct and (semi)automatic assessment. This is a key element of the AHEAD project, as it bridges the systematic and theoretical framework of AI bias assessment to the practical implementation that could be ultimately provided as a direct tool.

Key achievements

• Contributed to the overall work of AHEAD in providing a systematic structure to the indicators, effectively supporting the design of a multi-tiered metadata schema. These metadata will then be consequently used to design the appropriate (or adapt an existing) technological framework to capture them.
• Initiated the work on understanding the OEB benchmark descriptor files, which includes the available formats, structure, as well as the expected way to interact regarding the information needed for AHEAD indicators.
• Participated to the discussions towards a minimal viable product that combines well-defined indicators with the technical infrastructure necessary to assess them.

Why this matters for AHEAD

One of the key challenges in producing a comprehensive assessment of biases, is the lack of common definitions and structure metadata around the respective indicators. Making sure that this is addressed is a key step towards a more systematic effort.

Future work

Future work will focus on developing first prototypes of assessment (from indicator definition and metadata, to automatic assessment through computational tools), and investigating the interoperability between the project’s JSON indicator descriptors and the formats accepted by OpenEBench.