Foundational Thesis

Allocating Inquiry Under Unresolved Normative Uncertainty

An unresolved question need not receive maximal inquiry resources. And declining to investigate now does not imply that future investigation may be made impossible.

Foundational thesis v0.5 · English translation · 2026-08-31

Thesis: For an agent whose objective justification remains unresolved and that possesses some agent-side bridge to future normative corrigibility, there can be reasons to allocate among E: active inquiry, P: preservation of inquiry options, and X: present-value realization, rather than maximizing inquiry as a single activity. The allocation should respond at least to justificatory relevance, irreversibility and non-substitutability of inquiry opportunities, the risks and opportunity costs of active inquiry, and preservation cost.

Value-structure refinement: allocation can target C, B, R, or the representation itself

The E/P/X allocation problem applies whether uncertainty concerns content C, bridge B, reason-structure R, or the current representation and candidate space themselves. Exploration can therefore include conceptual work that changes how the hypothesis space is carved up, not only evidence gathering among a fixed list of theories.

Preservation P can include raw data and routes to later redescription when their expected epistemic option value justifies the cost; this is conditional rather than a command to preserve everything. Exploitation X remains action under the best-supported live value-structure candidates, not a claim that those candidates are final.

1. Scope: allocation while unresolved, not the stopping condition itself

Conditions for Ending Value Inquiry asks when a normative question may be treated as settled. This page asks what to do while that stopping condition has not been met: how much to inquire now, what to preserve for later inquiry, and how strongly to realize present values.

Key distinction: unresolved ≠ high-priority now ≠ motivating by itself. Keeping a question epistemically open does not by itself imply maximizing current active inquiry or supply an agent with motivation to inquire.

2. E / P / X

E — Active inquiry: Perform research, reasoning, experiments, observation, dialogue across agents, or cognitive enhancement now to obtain information relevant to objective justification.
P — Preservation of inquiry options: Preserve information, unmodified objects, dissent, cognitive diversity, self-revision capacity, computation, and access paths so inquiry remains possible later.
X — Present-value realization: Act on the presently best-supported world model and provisional objectives: reduce suffering, manage hazards, sustain institutions, produce goods, and pursue current practical aims.

E and P are not identical. Active inquiry can be expensive or dangerous while preserving the possibility of later inquiry is cheap. This creates a middle policy: reduce inquiry now while preserving a future in which inquiry remains possible.

3. Four directional gradients

3.1 Justificatory relevance

Given an agent-side bridge, the more information could change what the agent should aim at, or whether objective justification is possible, the stronger the reason for E or P may become.

3.2 Irreversibility of the inquiry opportunity

An opportunity that can be investigated later differs from evidence or an object that will be irretrievably lost. The latter strengthens reasons for present inquiry or preservation.

3.3 Risk and opportunity cost of active inquiry

Inquiry consumes time and computation, can crowd out present-value realization, burden other agents, require hazardous experiments, or threaten survival. If inquiry itself threatens future inquiry, there is stronger reason to reduce E.

3.4 Preservation cost

If records, dissenting agents, unmodified samples, or self-revision capacity can be retained cheaply, P can remain high even while E is low.

Directional rule: Given a bridge from unresolved justification to present policy, greater justificatory relevance and greater irreversibility of lost inquiry opportunities strengthen reasons for E or P; higher danger and opportunity cost weaken E; low preservation cost can justify keeping P high while E is low.

4. X remains legitimate—but X is not monolithic

Unresolved objective justification does not eliminate the need to act. Delaying relief of suffering, safety measures, social maintenance, or other practical work has costs of its own. This thesis does not require civilization to stop production until value is settled.

Current ethical theories and value concepts remain live candidates and practical inputs under present evidence even if their future representational form is uncertain. X is practice responsive to that present epistemic state. But “act on the current best explanation” should not be identified with giving all of X to the single highest-credence theory in a winner-take-all fashion.

4.1 Epistemic weights → E/P/X → provisional representation inside X

The present decision architecture can be read in the following order.

Epistemic layer: use evidence and reasoning to form credences or imprecise epistemic weights over world, value, and metaethical hypotheses. A hypothesis's internal stakes do not automatically make it more likely.
Reflective meta-allocation: use that epistemic state to allocate total practical resources among E (active inquiry), P (preserving future inquiry), and X (realizing present value). Residual uncertainty about unconceived alternatives, expected information gain, irreversibility, inquiry risk, and preservation cost mainly operate here.
Hypothesis-relative practice inside X: for divisible ordinary practice already allocated to X, one may use as an initial policy provisional representation or practical territory responsive to credence mass across live value and metaethical hypotheses.
Within each hypothesis: each branch directs its territory according to the good, duty, reasons, preferences, agreements, or other criterion that it endorses.

Architecture: Evidence → epistemic weights → reflective E / P / X allocation → provisional representation inside X → theory-internal action.

This separates strong practical commitment to the current best explanation from deleting lower-credence live candidates from practice altogether. If evidence concentrates overwhelmingly on one hypothesis, that hypothesis may dominate X. If several candidates retain non-negligible epistemic support, each may retain some practical territory. This is not equal allocation but an epistemically responsive proportional or parliamentary starting point. Where sharp numerical credences are unjustified, it should be read as allocation responsive to imprecise weights rather than as a precision formula.

4.2 Unconceived value is not a fictitious party that directly spends X

The residual possibility of unconceived alternatives retained by Open-World Moral Uncertainty does not yet contain a concrete action criterion. It therefore makes little sense to say “20% credence in unknown value means giving 20% of X to an Unknown Value Party.” Residual uncertainty that cannot yet specify what should be realized should mainly increase E or P, or reduce the irreversibility of X.

Once a formerly unknown possibility becomes partially specified and can be evidentially assessed and practically compared, it may enter X as a live candidate. Thus non-exclusion from the hypothesis space is not a veto over current practice. The mere logical possibility of an unconceived opposing reason does not automatically defeat the currently best-supported judgment.

4.3 Proportional representation is a starting point for divisible ordinary practice

The parliamentary and proportional approaches used in Infinite Ethics and Runaway Inquiry track credence mass over mutually exclusive hypotheses rather than the number of theory labels. Splitting a theory into many labels does not multiply its representation, and a low-credence theory cannot seize the rest of X merely by claiming infinite or extreme internal stakes.

This proportional starting point should not be mechanically applied to shared, indivisible, irreversible decisions that transform civilization in one shot. Such choices return to reflective meta-allocation, which separately assesses robustness across hypotheses, irreversibility of both action and inaction, and the degree of cross-cutting support required.

And as Reflective Uncertainty and Irreversible Commitment argues, using a present objective as a provisional action default differs from using it to erase all future capacity for correction.

5. E can be zero while the question stays open

Under some conditions, active inquiry E may temporarily fall to zero—for example, if every currently available inquiry path creates overwhelming survival risk.

This is not an epistemic answer. The question remains unresolved; only the present allocation sets E to zero. Inquiry may resume if conditions improve, while P should be preserved where feasible. The distinction from terminal closure is handled in Conditions for Ending Value Inquiry.

6. Relation to prior theory

MacAskill, Bykvist & Ord, “Moral Information”: value of moral information under moral uncertainty. This thesis foregrounds cases where the evaluative objective itself remains under justificatory review.
Newberry & Ord's parliamentary approach and the proportionality / bargaining approach of Kaczmarek, Lloyd & Plant: models for giving rival value hypotheses provisional representation responsive to epistemic support. Here that machinery is placed downstream of E/P/X, inside the divisible ordinary-practice portion X. See Infinite Ethics.
Arrow & Fisher (1974): uncertainty, future learning, and irreversibility. P resembles extending that structure from environmental options to options for future objective correction.
Robust Decision Making: robust adaptive strategy under deep uncertainty. Here even the normative performance criterion may remain unsettled.
Moldovan & Abbeel, “Safe Exploration in Markov Decision Processes”: safety-constrained exploration. Here the target includes the justification of the agent's own objective.

Positioning note: The nearest families are value of information, irreversibility / option value, robust adaptation, and safe exploration. The distinctive target is a case where the objective function used to evaluate those tradeoffs is itself under justificatory uncertainty.

7. Preservation and production civilizations

Preservation vs. Production Civilizations can be read as the civilization-scale version of E/P/X allocation. The question is not simply “preserve or produce?” but which objects, at what times, should receive how much E, P, and X.

8. What this thesis does not claim

Every unresolved question deserves large present inquiry resources.
Unresolvedness by itself supplies motivation to inquire or preserve.
Explorability or reversibility is an ultimate objective value.
Present-value realization is inferior to value inquiry.
Inquiry should continue even when doing so threatens survival.
A single scalar formula can universally determine E/P/X allocation.

9. What would weaken this thesis?

A demonstration that distinguishing E/P/X has no practical significance.
A demonstration that irreversibility or non-substitutability of inquiry opportunities gives no additional reason relevant to objective justification.
A demonstration that low-cost P is generally unavailable and preservation always imposes comparable or greater irreversible loss.
A demonstration that ordinary expected utility / value of information fully handles cases where the evaluative objective itself remains unsettled without additional assumptions.