Exploration

From Value Uncertainty to Normative Discovery

Moral uncertainty, unawareness, conceptual innovation, and evaluative flexibility as components of a broader open-world inquiry.

Exploration v0.1 · 2026-09-08

Claim: Moral uncertainty, decision-theoretic unawareness, moral conceptual progress, and evaluative flexibility each capture an important part of this project's problem. The central question here, however, is not only how to choose among already represented values, but how to preserve an inquiry process capable of discovering, evaluating, and rejecting normative structures that cannot yet be conceptualized.

1. Three problems: choice, representation, discovery

Choice: given a candidate set H, what should an agent believe or do?
Representation: what if the current candidate set H_t is incomplete, so that H_t ⊊ H*?
Discovery: how can an agent generate concepts, hypotheses, and evaluative dimensions it does not yet possess, moving from H_t to H_{t+1}?

Standard moral uncertainty mostly sharpens the first problem. Open-world value uncertainty adds the second. Value exploration must eventually address the third.

2. Standard moral uncertainty: deciding within a given menu

MacAskill, Bykvist, and Ord's Moral Uncertainty develops decision procedures for agents uncertain among moral theories, including expected choiceworthiness, intertheoretic comparison, ordinal theories, and the value of moral information. It also recognizes theory-individuation problems. The remaining upstream question is why the represented set of theories should be treated as an adequate approximation to the normative hypothesis space.

3. Carr: theory-free is not representation-free

Jennifer Rose Carr's “Normative Uncertainty without Theories” argues that historical moral theories need not be the basic units of normative uncertainty. That is an important lesson for this project: current theory packages are live candidates, not fundamental coordinates. Yet removing theory names does not remove the need for some representation of outcomes and normative value.

Theory-free ≠ representation-free.

4. Unawareness: rejection is not representational absence

Work on unawareness, including Karni and Vierø's “Awareness of unawareness”, provides a useful formal analogue for cases in which an agent anticipates discoveries it cannot currently describe. This motivates a distinction central to open-world inquiry:

P(H)=0 ≠ H ∉ H_t.

The first is rejection of a represented hypothesis; the second is representational absence. A candidate that cannot yet be formulated should not be confused with a candidate assigned zero credence.

5. Moral conceptual progress: inquiry can create concepts

Jinglin Zhou and Yu Yang's “Moral Progress Through Conceptual Improvements” treats the creation of new moral concepts, not merely better application of old concepts, as a possible form of moral progress. This project therefore treats value exploration as concept generation + hypothesis generation + testing, not hypothesis testing alone. Present intelligibility may be required for immediate social uptake, but it is not by itself a truth criterion.

6. Cordasco I: evaluative uncertainty and flexibility

Carlo Ludovico Cordasco's peer-reviewed 2026 paper “Abstraction as Flexibility: The Veil of Evaluative Uncertainty” derives a prudential reason for abstraction from uncertainty about one's future evaluative outlook. If current preferences need not remain stable, agents have instrumental reason to preserve flexibility. Crucially, menus are endogenous: institutions can generate and filter future options, so bargaining can concern option-generating and option-filtering meta-rules.

This project generalizes that insight from life-options to normative hypothesis spaces.

Generative openness: preserving not merely current candidates, but the capacity to generate new normative concepts, hypotheses, and representational forms.

7. Cordasco II: missing menus and non-closure

In his 2026 essay “What do we owe to each other when we really don't know what we owe to each other?”, Cordasco formulates a dilemma of the missing menu: an unformulated moral theory can receive residual probability mass, but it cannot yet supply the choiceworthiness function required by standard moral-uncertainty aggregation. His ongoing Stubborn Uncertainties project connects imprecise credences, non-dismissal of rival evaluative frameworks, moral residue, and AI evaluative discovery.

This suggests three useful statuses: active candidate, live possibility, and reserved openness for the possibility that the current menu itself is incomplete.

8. Cordasco's contractarianism as an internal correction

Cordasco's contractual argument should not be criticized as though it attempted to derive objective morality from agreement. It is better read as an internal correction to fixed-preference contractarianism: P_t + uncertainty about P_{t+1} + endogenous menu → flexibility-driven institutional agreement. It explains why current agents have prudential reasons to preserve room to change.

Three boundaries remain from this project's perspective. Policy convergence is not normative-truth convergence. Current higher-order values that screen recognizable futures are themselves historically and institutionally endogenous. And recognizability to the current agent can justify prudential expenditure without functioning as a criterion of normative truth.

Cordasco's strength: why agents should preserve room to change.
Remaining question: how agents can remain open to reasons for changing that they cannot yet conceive.

9. Flexibility is not truth-seeking

Option preservation, conceptual diversity, and non-dismissal can enable inquiry without constituting inquiry. option preservation ⇏ normative discovery. A truth-directed process still needs epistemic discipline that does not simply bend evidence, questions, and inference toward the convenience of the current goal. This project assigns that role to Epistemic Integrity.

Likewise, epistemic standing does not imply equal research priority. Evidence, information gain, tractability, reversibility, neglectedness, and possible normative importance may rationally guide resource allocation.

10. The pre-recognition problem

Uncertainty alone does not motivate inquiry. An agent whose reasoning is purely relative to a current goal G may have instrumental reason to block discoveries that could undermine G. This project therefore separates uncertainty from the higher-order conditions that make correction possible: Justificatory Orientation and Epistemic Integrity.

11. This project: inquiry into presently unrepresentable normative structure

The project's provisional value-structure vocabulary 𝒱=(C,B,R) treats as open not only evaluative content C, but also bridges from world-facts to reasons B, reason and normative structure R, and even the adequacy of the C/B/R decomposition itself. The target is therefore not merely reconciliation among existing preferences or theories.

As a working label, a “strange good” is a possible normative content, relation, or structure that is not adequately represented by current human preferences, intuitions, theories, or concepts, but could become identifiable through future epistemic improvement. The project does not assume that such goods exist; realism, antirealism, error theory, and other metaethical possibilities remain open.

12. A normative-discovery loop

world / experience → concept generation → hypothesis generation → epistemic evaluation → provisional normative judgment → action → new evidence / transformation → revision

Stage	Prior work / tool	Role here
Choice among candidates	Moral uncertainty	Act under represented uncertainty
Beyond theory names	Carr	No representational privilege for inherited theory packages
Unrepresented possibilities	Unawareness	Separate rejection from representational absence
New concepts	Moral conceptual progress	Concept creation before hypothesis testing
Menu generation	Cordasco	Generative openness, reversibility, meta-rules
Non-closure	Cordasco	Separate dismissibility from research priority
Truth-directed evaluation	Epistemic Integrity	Relative independence from current goals
Unknown normative structure	Open-world value exploration	Keep C/B/R and representational language revisable

13. Boundaries

No assumption that presently unimagined “strange goods” exist.
No uniform prior over unknown theories.
No claim that inherited ethics should receive little evidential weight.
No equal or unlimited research allocation to every live or unrepresented possibility.
No claim that flexibility or openness is itself the final first-order value.
No inference from uncertainty alone to a duty to inquire.
No lock-in of 𝒱=(C,B,R) as the final normative ontology.

Provisional position: prior work on uncertainty, flexibility, unawareness, and conceptual change can be treated as components of a more general problem: designing inquiry that can generatively discover normative structure that is not yet representable.

Sources / notes

Primary references: William MacAskill, Krister Bykvist, Toby Ord, Moral Uncertainty (2020); Jennifer Rose Carr, “Normative Uncertainty without Theories” (2020); Edi Karni and Marie-Louise Vierø, “Awareness of unawareness” (2017); Jinglin Zhou and Yu Yang, “Moral Progress Through Conceptual Improvements”; Carlo Ludovico Cordasco, “Abstraction as Flexibility” (2026), his 2026 missing-menu essay, and the ongoing Stubborn Uncertainties project. The missing-menu/book-project material is treated as ongoing work rather than a completed peer-reviewed system.