Grounding Truth in Non-monotonic Reasoning

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 250 papers94 conceptsfrozen 2026-10-04

In classical logic, the statement "A is true if B is true, and B is true if A is true" is perfectly consistent. If both are true, the logic holds. However, in artificial intelligence and common-sense reasoning, this kind of circularity is a bug, not a feature. If an AI believes there is a fire because it sees smoke, and believes there is smoke because it assumes there is a fire, it has "bootstrapped" a belief out of thin air.

To build reliable agents, researchers have developed a suite of techniques to ensure groundedness—the requirement that every belief must be supported by a non-circular chain of justifications reaching back to foundational facts.

The Problem of Floating Beliefs

The challenge arises primarily in non-monotonic reasoning, where agents must make assumptions and revise them as new information arrives. Early work by Jon Doyle in The Ins and Outs of Reason Maintenance identified that systems tracking justifications often fall into "self-supporting" loops. Without a formal requirement for finite grounding, a system might maintain a set of beliefs that are internally consistent but logically invalid because they lack external roots.

Robert C. Moore formalized this in the context of an agent's self-reflection. In Semantical Considerations on Nonmonotonic Logic, Moore argued that an "ideally rational agent" should not just be consistent; its beliefs must be epistemically grounded in initial premises. This prevents the inclusion of arbitrary, self-fulfilling beliefs that arise simply because the agent assumes they are true.

Enforcing a Derivation Order

How do we mathematically forbid these loops? The most intuitive solution is to require that every truth has a "timestamp" or a level in a hierarchy.

One powerful approach is ordered completion, developed by Fangzhen Lin, Yan Zhang 0003, and Yi Zhou 0013. In Ordered Completion for First-Order Logic Programs on Finite Structures, they introduced auxiliary "comparison predicates" that track the order of derivations. For a belief to be valid, it must be supported by premises that were established at a strictly lower level in the order. This mechanism effectively turns a potentially circular graph of justifications into a well-founded tree.

Similarly, Yi-Dong Shen and Kewen Wang proposed well-justified FLP semantics in FLP Semantics Without Circular Justifications for General Logic Programs. By integrating a level-mapping formalism, they ensure that every atom in an "answer set" (a valid state of the world) is derived through a non-circular sequence of steps, filtering out models that rely on self-supporting loops.

Detecting Loops with External Support

In Answer Set Programming (ASP), a popular paradigm for solving complex constraints, the problem of circularity is often handled through loop formulas.

As Fangzhen Lin and Yuting Zhao demonstrated with the ASSAT system, standard SAT solvers can be used to find answer sets if we incrementally add constraints that break these loops. The core idea is external support: a set of atoms is only valid if at least one atom in that set can be proven using information from outside that set.

Joohyung Lee 0002 generalized this in A Model-Theoretic Counterpart of Loop Formulas, showing that "externally supported sets" are the fundamental bridge between classical logic and the more restrictive "stable model" semantics. If a group of atoms only justifies itself, it is "unfounded" and must be rejected.

From Truth to Process

This research direction represents a shift in how we define "correct" reasoning. Instead of viewing logic as a static set of truths, researchers like James W. Goodwin argue for a process theory. In A Process Theory of Non-monotonic Inference, Goodwin suggests that theoremhood should be replaced by the existence of a computable, non-cyclic "current proof."

This focus on the path to a conclusion is what makes these systems explainable. As Tran Cao Son and Enrico Pontelli note in Justifications for Logic Programs under Answer Set Semantics, a "cycle-safe" grounding is what allows a system to provide a "declarative trace"—a step-by-step explanation that a human can follow without getting lost in a hall of mirrors.

By enforcing groundedness, these researchers have moved AI from simple consistency to rationality, ensuring that every conclusion an agent reaches is built on a solid foundation of evidence rather than the circular logic of its own assumptions.

Papers in this version

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 77 papers
  2. Version 1 · 2026-10-04 · 50 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_circular_justification_9bd2ac4d,
  title = {circular justification},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/circular-justification/rdr_9bd2ac4d99664e93be2e19f415f09447},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-21},
}

This exact version (version 1, frozen text):

@misc{lacuna_grounding_truth_in_non_monotonic_reasoni_9bd2ac4d_v1,
  title = {Grounding Truth in Non-monotonic Reasoning},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/grounding-truth-in-non-monotonic-reasoning/rdr_9bd2ac4d99664e93be2e19f415f09447/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-21},
}