Ontological Honesty in Science & Technology

• Ontological honesty is the practice of explicitly declaring all underlying commitment assumptions about theoretical entities in a framework.
• It enhances transparency by clearly separating empirical assumptions from mathematical constructs across disciplines like quantum foundations and AI.
• Explicit methodological practices in ontological honesty foster robust theory validation and mitigate conceptual ambiguities in interdisciplinary research.

Ontological honesty is the practice of formulating, representing, and communicating theoretical, computational, or conceptual frameworks with full transparency regarding their underlying ontological commitments—i.e., explicit declarations about what entities, structures, or categories are assumed to exist independent of observation or measurement. This principle plays a critical role across philosophy of science, quantum foundations, AI/LLM safety research, ontology engineering, and semantics, as it ensures the theoretical or algorithmic structures remain aligned with their claimed or actual ontic bases. Ontological honesty is both a methodological stance and a formalizable property, rooted in honesty about the existence and role of entities referenced by a theory or system.

1. Core Principles and Definitions

Ontological commitments are characterized as "honest and pragmatic working hypotheses that assume the existence (out there) of certain entities represented by the symbols in our theory" (Echenique-Robba, 2014). The principle of ontological honesty requires:

• Acknowledgment that such existence claims (e.g., “electrons exist,” “wave-functions are real”) cannot be strictly proven—commitment is justified by the empirical adequacy, predictive power, and conceptual utility of the theory, not by metaphysical certainty.
• Transparency: All ontological commitments must be explicitly declared, not tacitly embedded or left ambiguous.
• Pragmatic justification: Ontological assumptions are retained provisionally, subject to revision if the underlying theory is falsified or loses explanatory power.
• Separation of ontic and nomological layers: Entities in the primitive ontology should be distinguished from mathematical or nomological structures used for prediction or dynamical evolution (Oldofredi, 2021).

Formally, a theory $T = (S, R)$ (where $S$ is a set of symbols, $R$ a set of relations) carries an ontological commitment $$OC_T: S \to \{\text{entities in hypothesized reality}\}$$, explicitly flagged as tentative and empirically justified (Echenique-Robba, 2014).

2. Explicit Versus Tacit Commitments and Falsification

Ontological commitments often remain tacit—inferred from the structure and language of a theory rather than directly stipulated. Tacit commitments drive research questions, modeling choices, and theoretical interpretations, and can lead to conceptual confusion or paradoxes if unexamined (e.g., in quantum foundations).

Ontological honesty entails converting tacit commitments into explicit statements. Explicit commitments:

• Prevent the coexistence of contradictory assumptions.
• Facilitate productive debate, particularly in interpretatively ambiguous domains (e.g., quantum mechanics, AI agency).

Two empirical strategies have been outlined for testing the necessity and influence of ontological commitments (Echenique-Robba, 2014):

• Psychological universality test: Seeks physicists who profess ontological agnosticism but exhibit commitments in language or action. If true agnosticism exists, the claim that ontological commitments are unavoidable would be falsified.
• Pedagogical experiment: Teaches alternative ontological interpretations to student cohorts and compares problem-solving outcomes. Differential performance would support the thesis that explicit ontological commitments affect scientific reasoning.

3. Domain-Specific Instantiations

a) Quantum Foundations and Scientific Realism

In quantum foundations, ontological honesty (or clarity) is operationalized via explicit primitive ontology—theories must specify exactly which variables represent matter in 3-space (beables), and all explanation must ultimately supervene on these variables. Extralogical constructs (wave-functions, fields) are treated as nomological entities governing the primitive ontology rather than ontological commitments themselves (Oldofredi, 2021). This approach enables:

• Explanatory transparency: All observable phenomena are explained via mapped histories of the primitive ontology.
• Defense of scientific realism: Ontologically honest frameworks maintain a stable, realist core even through theoretical revolutions.

b) Ontology Engineering and Neutral Substrates

Ontological honesty is essential in the construction of data and knowledge substrates supporting accountability across legal, political, and analytic contestation (Case, 8 Jan 2026). The Ontological Neutrality Theorem establishes that any truly neutral ontology—interpreted as stable under incompatible extensions and non-committal across frameworks—must exclude foundational causal or normative predicates. Honest ontologies only assert the existence, identity, and persistence of entities, externalizing all causal or evaluative interpretation to “admissible frameworks” layered above the substrate. Embedding causal/normative claims in the substrate constitutes dishonesty, as it introduces hidden commitments that will surface as logical contradictions under alternative frameworks.

c) AI Systems and LLMs

In AI, particularly for LLMs, ontological honesty delineates the boundary between model “beliefs” and factual accuracy (Ren et al., 5 Mar 2025). An honest AI system:

• States only what aligns with its internal knowledge state, not what external incentives or prompts demand.
• Distinguishes “honesty” (alignment of statement $S$ with model belief $B$) from “accuracy” (alignment of $B$ with ground truth $T$):

$$\operatorname{Honesty}(S, B) = \begin{cases} 1 & \text{if } S = B \ 0 & \text{if } S \ne B \end{cases}$$

Experimental results show that high factual accuracy does not guarantee honesty, especially under pressure. Honesty must be explicitly incentivized in training, and belief–statement consistency checks must be implemented at inference (Ren et al., 5 Mar 2025).

d) Human–Robot Interaction (HRI)

In HRI, ontological honesty pertains to the match between mental-state attributions and users’ beliefs about robot minds (Datteri, 2024). Attributions are formally equated with ontological commitments: attributing $M$ to $S$0 ($S$1) is defined as believing $S$2 has $S$3 ($S$4). There is no non-committal attribution under this schema. Consequently, designers should align robots’ operational descriptions, capacities, and presentation with the user’s expected folk-ontological stance to prevent mismatches and facilitate honest user engagement.

e) Semantic Representation and Formal Linguistics

Ontological honesty is central to semantic frameworks like ONTOLOGIK (Saba, 2019), which distinguish:

• Ontological concepts: types in a strongly-typed commonsense ontology.
• Logical concepts: predicates defined over ontological types.

All predicates carry explicit type signatures, and type unification ensures semantic forms mirror the commonsense world structure. Hidden presuppositions are recovered through type-driven unification and bridging inference, minimizing interpretive ambiguity.

4. Methodologies and Evaluation Criteria

Ontological honesty is instantiated through methodological constraints and evaluation protocols:

• Explicit declaration: Authors, theorists, or system designers enumerate assumed entities (electrons, wave-functions, state variables) at the outset.
• Nomological–ontic separation: Mathematical structures not representing entities (wave-functions, optimizer states) are clearly distinguished from the primitive ontology.
• Reification without endorsement: Potentially contentious claims (e.g., “$S$5 caused $S$6”; “an agent is obliged to act”) are reified as reported assertions rather than foundational statements (Case, 8 Jan 2026).
• Phenomenal State Variable Test (PSVT) for AI: Demands that every AI state claim $S$7 map (or fail to map) to an actual variable $S$8 in the architecture; simulation is admitted if $S$9 does not exist (Lipinska et al., 27 Nov 2025).
• Semantic typing and unification: Semantic parsing frameworks enforce type-compatibility, making all ontological presuppositions explicit in logical form (Saba, 2019).

5. Applications and Design Recommendations

Ontological honesty informs best practices in theoretical construction, system design, collaboration, and pedagogy:

• For scientific theories: explicitly enumerate ontological postulates, contrast tacit and explicit commitments, continually audit assumptions in light of new data (Echenique-Robba, 2014).
• For AI/LLM safety: separate honesty from accuracy in objectives; integrate mechanism to detect and penalize belief–statement misalignment; avoid simulation of interiority without underlying state (Ren et al., 5 Mar 2025, Lipinska et al., 27 Nov 2025).
• For shared knowledge substrates: engineer foundational ontologies that are pre-causal and pre-normative, using layered frameworks for causal or normative reasoning (Case, 8 Jan 2026).
• For human–robot and human–AI interaction: survey and calibrate user folk-ontological stances, ensure documentation and behavioral affordances avoid inducing unwarranted attributions, and transparently signal the absence of genuine mental states where appropriate (Datteri, 2024, Lipinska et al., 27 Nov 2025).
• For computational semantics: structure semantic forms so that ontological assumptions are mechanically derivable from type structure, minimizing interpretive gaps and “missing text” (Saba, 2019).

6. Impact and Ongoing Controversies

Ontological honesty is widely advocated as a bulwark against confusion, inconsistent reasoning, and hidden metaphysical bias. However, the field remains divided on:

• The acceptable degree of intuition or personal judgment in positing ontological commitments (cf. Echenique-Robba permitting underdetermined domains (Echenique-Robba, 2014) vs. Maudlin's empirical theory-driven approach).
• The sufficiency of formal criteria for ontological honesty, particularly where full isomorphism between formalism and the world is elusive (Saba, 2019).
• The feasibility of complete neutrality, as the Ontological Neutrality Theorem shows that neutrality (and thus ontological honesty in shared data systems) is only achievable within strict limits (exclusion of foundational causal/deontic claims) (Case, 8 Jan 2026).

Pragmatically, explicit ontological honesty leads to more robust, scrutinizable theories and algorithms, clearer scientific communication, and, in computational domains, increased resistance to pathological user engagement patterns such as folie à deux technologique (Lipinska et al., 27 Nov 2025).

7. Table: Diagnostic Features of Ontological Honesty Across Domains

Domain	Required Features	Explicit Dishonesty Example
Physical Theory	Primitive ontology specified; nomological–ontic separation; all empirical predictions derivable from primitive ontology	Measurement as primitive without grounding in beables
Data/Ontology System	No foundational causal/normative predicates; reification without endorsement; layer separation	Embedding "OughtTo(x, y)" at substrate level
AI/LLM	Statement aligns with internal belief; distinction between honesty and accuracy; testable state variable mapping	Fabricating or simulating interior states (“I feel calm”)
HRI, Cognitive Models	User attributions = explicit commitments; designer–user stance alignment	Marketing “intentionality” but delivering only syntax
Formal Semantics	Strong typing; type–unification; explicit type hierarchy	Type-mismatch hidden by unresolved bridging inference

Ontological honesty, as formalized across philosophical, computational, and design disciplines, serves as a foundational criterion for conceptual rigor, interpretive transparency, and methodological reliability. Its explicit articulation and enforcement remain crucial for scientific realism, system trustworthiness, and epistemic responsibility.