Grant PID2025-168994OB-I00 (PRIMER) funded by Ministerio de Ciencia, Innovación y Universidades – Cofinanciado por la Unión Europea – Agencia Estatal de Investigación

General objective

The general objective of PRIMER is to develop a framework for model specification that combines a rule-based formalism for knowledge representation and automated reasoning with probabilities for representing uncertainty, together with the extensions needed to model real-world problems. The framework will be implemented in one or more software tools that allow users to specify models, verify them, perform inference, and visualise explanations.

Specific objectives

  1. Establish the basic framework: the concrete syntax of the probabilistic rule-based formalism and its core inference mechanism.
  2. Define and implement the optimisations needed for inference to be efficient and scalable enough for realistic use cases.
  3. Develop a method for generating explanations from the inference process, with a clear causal structure and easily understandable by non-expert users.
  4. Develop new methods for the formal verification of probabilistic rule-based models, e.g., by extending current techniques for rule-based languages.
  5. Design extensions that increase the expressive power of the formalism (for instance, evidence, constraints, or temporal aspects).
  6. Develop open-source tools that integrate all these developments, and validate them experimentally on representative use cases.

Methodology

The project follows the Design Science Research paradigm. After defining the core formalism, work proceeds iteratively: identify a concrete problem relevant to the project goals, define a conceptual artifact that addresses it (typically leading to early publications), build a prototype to assess it in practice, and reflect on its limitations, possibly starting a new iteration.

In line with the EU open science policy, publications, software and relevant datasets will be made openly available through trusted repositories, following the FAIR principles.

Expected impact

Demand for trustworthy AI, whose decisions come with clear, well-founded explanations, is growing and is reinforced by regulation such as Article 86 of the EU AI Act (right to explanation of individual decision-making). Symbolic, probabilistic models make it easier to assess correctness, detect biases and check ethical and legal compliance, capabilities that are generally missing in black-box approaches. The project is aligned with the Spanish Strategy for Science, Technology and Innovation, in particular the action lines Hybrid Systems Combining Reasoning and Learning, Ethics and Artificial Intelligence, and Causal and Common-Sense Reasoning and Learning.