Policy-Driven AI in Dataspaces: Taxonomy, Explainability, and Pathways for Compliant Innovation

Privacy and data sharing require trade-offs: The paper identifies that techniques like Federated Learning, Differential Privacy, and Homomorphic Encryption can protect privacy in AI dataspaces, but each comes with performance costs—meaning organizations must balance privacy protection against factors like processing speed, system costs, and model accuracy.

Current regulations lack standardized compliance tools: While frameworks like GDPR and the EU AI Act exist, the paper highlights that there are no standardized methods to measure and validate whether AI systems actually comply with these regulations, creating uncertainty for organizations trying to follow the rules.

Explainability remains a major gap in collaborative AI: The research identifies that explaining how AI decisions are made becomes significantly harder in federated ecosystems where data and processing are distributed across multiple organizations, yet this transparency is essential for responsible AI deployment.

A unified framework is needed for policy-driven AI: The paper proposes that practitioners need a clear taxonomy and standardized benchmarks to navigate the complex landscape of privacy techniques and regulatory requirements, rather than making decisions in an ad-hoc manner.

European data initiatives offer a pathway forward: The paper suggests that emerging European projects like GAIA-X and Eclipse EDC could help establish common standards and infrastructure for trustworthy AI dataspaces that balance innovation with privacy and compliance requirements.