2026 EDITION

When Data Pipelines Decide AI Trust and Efficiency: theory and practice

May 5, 2026 - 16:00 - 17:00

Room H3

The majority of AI/ML model training relies on highly curated, high-quality subsets of available data, not the full data universe. Most data is either ephemeral, never stored, or remains ‘dark’ (unused for analytics or AI). In fact, IDC estimates that less than 3% of all data generated globally is actually used to fuel AI models today. If we focus on the enterprise context, IDC data indicate that only 10–15% of data is considered ‘AI-ready’ and actively leveraged for AI/ML workloads. The rest is unstructured, siloed, or lacks sufficient quality/governance for effective AI use.

However, trends show that by 2028, more than 40% of archived data is expected to be reclassified as ‘strategic’ due to AI-driven pattern recognition; this does not mean all of it will be used for model training—just that its potential value is recognized. Synthetic data is also on the rise, with 30–35% of organizations expected to rely on it for AI by 2026, further diversifying the data sources for AI models.

In view of the current situation and such trends, this workshop will bring on stage the following elements: a) Data pipeline modernization and AI data readiness strategies; b) most innovative developments in synthetic data adoption for AI; c) Impact of data governance and quality initiatives on AI utilization and d) examples and pilots representing a diversity of relevant sectors, so that the audience gets a practical view of what the most advanced techniques are, and the real impact in the corresponding industrial environments.

The portfolio of projects funded by the EC on “Efficient trustworthy AI – making the best of data”, including PANDORA (lead), AI-DAPT, RAIDO, MANOLO, EXTRA-BRAIN, and TURING, will create a dynamic context where main outcomes so far will be informed and shared with the audience, practical examples will be presented, challenges and gaps will be discussed openly, and feedback will be collected for BDVA SRIDA updates and execution plans of the projects.

The workshop will consist of a panel where speakers will feature insightful presentations and will discuss the topics above, including audience engagement.

Expected Outcomes:

Insights into the current developments and portfolio of solutions working on making the best of data for efficient and trustworthy AI. Advantages and limitations.
Clear exposure of pilots and practical use cases in different industrial environments
Collective work addressing elements like trustworthy AI in operation, adaptive and resilient AI systems, alignment with EU regulation or adoption at scale.

Agenda

16:00-16:10 Introduction to the session: objectives and relevance (Nuria de Lama, IDC, PANDORA)

16:10-16:55 Conceptualisation and Operationalisation of “Trustworthy AI”: Insights from projects and Panel discussion

Representatives from PANDORA, TURING, RAIDO, AI-DAPT, MANOLO and EXTRA-BRAIN will provide a short overview of their projects, highlighting the achivements so far vis-à-vis the topic of the session. In particular, each of them will point out how they conceptualise and operationalise “trustworthy AI.”

Each project will tackle “HOW” they:

Understand efficient trustworthy AI (conceptual framing)
Operationalise it (methods, pipelines, tools, mechanisms)
Measure it (KPIs, validation methodologies, pilots)

Projects will be asked for content through targeted questions in a discussion panel that will enable the audience to step in for clarifications or add-ons.

Participants:

PANDORA: Maria Pateraki (NTUA)

TURING: Konstantinos Tserpes (ICCS-NTUA)

RAIDO: Christos Vasilakis (Metamind Innovations MINDS)

AI-DAPT: Carlos Agostinho (Uninova)

MANOLO: Julia Palma (CeADAR)

EXTRA-BRAIN: Pawel Herman (KTH)

16:55-17:00 Wrap-up and next steps