Gines Molina-Abril

Gines Molina-Abril

Data & Decision Engineer | PhD Candidate

ARGOS

ARGOS logo ARGOS logo

Auditable and Bias-Resilient Evaluative AI for Trustworthy Strategic Decision Support in Small and Medium Enterprises (SMEs)

A collaboration proposal for a joint publication and an extension of the work presented at IDEAL 2025.

The name: ARGOS

ARGOS = Argumentation-based Reasoning and Governance for Open Strategic decision support.

Each part names a piece of the framework:

  • Argumentation-based Reasoning: the core evaluates each hypothesis as a structured debate of arguments for and against, aggregated in a deterministic and auditable way (not an opaque output from an LLM).
  • Governance: explicit human/system governance — the human governs the framework (criteria, rubrics, hypotheses) and the system computes the content; the human does not edit the output numbers, but can define or refine them during the process.
  • Open: open decisions (a per-criterion profile, not a binary verdict), evidence from open and multiple sources, and an open-source release.
  • Strategic decision support: strategic decision support for SMEs.

The name also evokes Argos Panoptes (Ἄργος Πανόπτης), the watchman of Greek mythology: a giant with a hundred eyes, sentinel and shepherd, who never closed all of his eyes at once — there was always one awake, tasked with watching without rest. It is the exact metaphor for the system: a gaze that sees everything, attentive to evidence from many sources at the same time, always vigilant and auditable, where nothing escapes scrutiny. Hence also the symbol in the logo: the o in Argos is a magnifying glass with an eye inside — to search, to investigate and to watch in a single gesture.

Summary

ARGOS extends the hypothesis-driven Evaluative AI (EAI) framework for SMEs presented at IDEAL 2025 into an auditable and bias-resilient system, oriented towards trustworthy strategic decision support.

Instead of predicting a recommendation, the system evaluates business hypotheses by gathering evidence for and against them from multiple sources, anchoring it to its exact origin, and turning it into a multidimensional profile per business variable (opportunity, risk, investment, cost, team…) that the human decision-maker interprets and governs.

Main objectives

  1. End-to-end auditability. Every conclusion is traceable back to the evidence that originates it (source anchoring), and the conversion of evidence into a score is deterministic and reproducible, not an opaque output from an LLM.
  2. Bias resilience. Mitigate the bias of a single model / single source through multi-source evidence with trust-tiering, a bipolar argumentative structure (support/attack) and human contestability over the evaluation framework.
  3. Non-collapsed output. Deliver a per-criterion decision matrix (not a verdict) so that the decision-maker sees the full profile and where the lever of each decision lies (business impact).
  4. Accessibility for SMEs. An open-source stack, runnable locally (confidentiality), released under a permissive license, with a modular plug-and-play design.
  5. Industrial validation. A demonstration on a real case at a digital marketing SME (diversification and discovery of new business lines).

This project builds on and extends:

Molina-Abril, G., Calvet, L., Riera, D. (2026). Towards an Evaluative AI Framework for Hypothesis-Driven Strategic Decision-Making in SMEs. In: Martínez, L., et al. Intelligent Data Engineering and Automated Learning – IDEAL 2025. Lecture Notes in Computer Science, vol 16238. Springer, Cham. https://doi.org/10.1007/978-3-032-10486-1_34

See the publication page: Towards an Evaluative AI Framework for Hypothesis-Driven Strategic Decision-Making in SMEs.