SCCS-UAH Artificial Intelligence

—Research Focus and Activities

The SCCS-UAH group advances safe, empirically-grounded artificial intelligence for economic analysis, spatial systems, and public-interest applications. Our work combines agent-based modeling (ABM), large language models (LLMs) as decision-making agents, network science, and computational geography, with an explicit commitment to governance-first design, auditability, and replicability.

What we study

  • AI-for-Economics & Policy: We build ABMs where firms and households are controlled by LLM agents to study market competition (Cournot/Bertrand/Hotelling baselines), macro-fiscal adjustments, and redistribution mechanisms (e.g., digital-euro scenarios). Results are benchmarked against theory and validated on real datasets.

  • Synthetic Populations & Digital Twins: We generate and calibrate virtual populations (e.g., EPF/ECV-style consumption and income structures) to evaluate policies, shocks, and counterfactuals, including family-coupled decision rules and spatial heterogeneity (urban–rural, regional).

  • Semantic Dynamics & Knowledge Infrastructures: We develop pipelines to measure semantic drift (Spanish/English; Word2Vec/embedding-based) across dated corpora, and explore cross-lingual conceptual geometry for scientific discovery and AI evaluation.

  • Historical & Modern Networks as Computation Substrates: We reconstruct and analyze telegraph/rail/road networks to study diffusion, market integration, and accessibility, and connect these insights to contemporary infrastructure and mobility digital twins.

  • Safety, Governance, and Evaluation: We prototype “cognitive tripwires,” alignment metrics, and audit protocols for multi-agent LLM systems, stressing containment, risk mitigation, and transparent reporting.

How we work

  • Methodological stack: Python (Mesa for ABM, vectorized pipelines), Polars/Pandas, GPU/CUDA acceleration, rigorous calibration (Sobol, Bayesian/likelihood-free), and open, reproducible notebooks.

  • Data practice: FAIR principles; integration of official statistics (INE/Eurostat/World Bank), historical sources, Wikidata/OSM, and large behavioral datasets where licensing permits.

  • Validation: Theory-constrained simulation (recovering known equilibria under restrictive assumptions) plus empirical calibration and out-of-sample tests; ablation and sensitivity analyses.

  • Education & Outreach: “AI-first” pedagogy in Economics and Business, classroom debates with automated feedback (ASR+NLP), and hands-on seminars that connect students to real research code and datasets.

Current priorities

  1. LLM-driven market simulation with explicit institutional and spatial frictions.

  2. Population-level digital twins for mobility and household finance, with family coherence and privacy-preserving synthetic data.

  3. Metrics for AI progress and risk, combining semantic geometry, multi-agent stress tests, and governance playbooks.

  4. Historical network reconstructions as laboratories for diffusion, resilience, and accessibility policies.

SCCS-UAH’s mission is to translate AI advances into credible, auditable evidence for economic understanding and policy design—always prioritizing safety, accountability, and societal benefit.