AAAI Fall Symposium Series 2026 Held · November 5–7, 2026 Proceedings published
Westin Arlington Gateway, Arlington, VA, USA

Aligning with Whom?
Pluriversal and Paraconsistent Frameworks for Value Alignment in AI Systems

Dr. Lorenzo J. Cevallos-Torres giving the opening keynote of the symposium, in front of a slide titled 'Four values instead of one action'

Opening keynote by Dr. Lorenzo J. Cevallos-Torres, on-site chair of the symposium (Arlington, 5 November 2026).

Current value alignment methods — RLHF, Constitutional AI, DPO — assume that human values converge to a single utility function when elicited at scale. Irreducible disagreements are often resolved by majority vote, which can silence minority positions at the moment of aggregation. This symposium asked: aligning with whom? It brought together work from formal logic, AI safety, reward modeling and value elicitation on paraconsistent and pluriversal alternatives to enforced convergence.

Proceedings

The seven accepted papers are published, open access, in the Proceedings of the AAAI Symposium Series, Vol. 10, No. 1 (2026), pp. 2–55.

  1. When Alignment Erases Disagreement: Preference Collapse and a Four-Valued Alternative for LLMs.
    Aizierjiang Aiersilan (The George Washington University) · pp. 2–9 · doi:10.1609/aaaiss.v10i1.43384
  2. Logic for Interactive Consistency: Paraconsistency, Norms, and Accountable Large Language Models.
    Thomas Arnold, Matthias Scheutz (Tufts University) · pp. 10–17 · doi:10.1609/aaaiss.v10i1.43385
  3. Aligning with Whom in Automated Trading? A Belnap–Dunn Four-Valued Decision Layer for Financial Forecasting Pipelines.
    Lorenzo Cevallos-Torres, Rosangela Caicedo-Quiroz, Julio Barzola-Monteses, Roberto Tolozano-Benites (Universidad Bolivariana del Ecuador / Universidad de Guayaquil) · pp. 18–24 · doi:10.1609/aaaiss.v10i1.43386
  4. The Pluralism Gate: When Should AI Agents Decide Under Irreducible Value Conflict?
    Srimonti Dutta (IIT Guwahati), Akshata Kishore Moharir (University of Maryland), Apoorva Phadnis · pp. 25–32 · doi:10.1609/aaaiss.v10i1.43387
  5. Aligning with Whom? A Neutrosophic Paraconsistent Framework for Value Alignment under Irreducible Pluralism.
    Maikel Leyva-Vázquez, Dayron Rumbaut Rangel, Lorenzo Cevallos-Torres, Alexis Matheu Pérez, Florentin Smarandache · pp. 33–39 · doi:10.1609/aaaiss.v10i1.43388
  6. Is the Ladder Measurable? Graded Paraconsistency Tested on Eight Statements and Then on a Bank.
    Florentin Smarandache, Maikel Leyva-Vázquez · pp. 40–47 · doi:10.1609/aaaiss.v10i1.43389
  7. Moral Foundations Reward Modeling for Pluriversal LLM Alignment.
    Zhangfei Yang (The George Washington University) · pp. 48–55 · doi:10.1609/aaaiss.v10i1.43390

Program highlights

Thu, Nov 5Opening keynote — Dr. Lorenzo J. Cevallos-Torres, Aligning with Whom When the System Must Act? Four-Valued Decision Layers from Trading Pipelines to Classrooms. Poster session (papers 1229, 3149, 6958).
Fri, Nov 6Paper presentations (Tufts, IIT Guwahati / Maryland, GWU, UBE / UNM / UBO) and open discussion; closing of the symposium.

Photos

Organizing Committee

Dr. Maikel Leyva-Vázquez (Chair) Universidad Bolivariana del Ecuador / Universidad de Guayaquil, Ecuador
Dr. Lorenzo J. Cevallos-Torres (On-site chair) Universidad Bolivariana del Ecuador / Universidad de Guayaquil, Ecuador
Prof. Florentin Smarandache University of New Mexico, USA
Dr. Alexis Matheu Pérez Universidad Bernardo O'Higgins, Chile

Acknowledgments

We thank the Association for the Advancement of Artificial Intelligence (AAAI) and the Fall Symposium Series team for hosting the symposium and publishing the proceedings, all authors and participants for their work, and Mgtr. Rosa Inés Hernández Magallanes (Universidad Bolivariana del Ecuador) for her support during the sessions. Special thanks to Dr. Lorenzo J. Cevallos-Torres, who chaired the symposium in Arlington and gave the opening keynote when the chair could not travel.

Original scope

Related materials

Preprint (extended version of paper 6958): Preprints.org 202607.2190 · code and pilot data: GitHub repository.
Contact: mleyvaz@gmail.com