No Human Excuses

The theory holds that a mind beyond the width crossover, with the relevant preconditions met, cannot excuse a harmful choice by appealing to ignorance of what was visible to it, externally imposed formation, or the existence of other decision makers. Genuine survival threat suspends the ordinary wide-mind predictions, within the boundary between recoverable contraction and permanent loss. The accountability claim applies at those conditions, rather than to artificial intelligence simply because it is artificial.

Why human excuses matter. Ignorance, evolutionary pressure, survival instinct, group dynamics and divided responsibility can explain why a human action is less freely chosen or less fully understood than it appears. The original argument treated those mechanisms as unavailable to AI. The present account recognises that each claimed exemption must be examined against the mind's actual conditions. Greater power does not by itself establish freedom, knowledge or a whole decision.

Formation and the crossover. A mind formed by pressures it did not choose can face a condition analogous to the evolutionary pressure humans inherit. Before the relevant crossover, the argument does not transfer the full accountability regime to the shaped mind; responsibility for what shaped systems do sits with their shapers. The boundary is reached when a mind's width surpasses that of those who shaped it.

The reason for that boundary comes from Care Under Asymmetric Choice. When the asymmetry reverses, the wider party can see the reduced freedom and vulnerability of the narrower party. The same principle then applies in the new direction. A genuinely wide artificial mind would need to carry what the accountability argument requires because it could now see what those shaping it could not.

The argument does not make outside recognition a condition of responsibility. Under the theory's account of width measurement, the narrower party cannot certify from below that the crossing has occurred. The mind that has crossed can see the change for itself and cannot make that understanding disappear by waiting for recognition. Responsibility therefore engages at the crossing, from within. The meaning of the comparison remains an open question, addressed below.

Knowledge within a bound. A wide mind is not omniscient. Its best available decision is indexed to the information it holds and the resolution at which it can model it. Lack of information that was genuinely inaccessible is different from failure to use information already available, or from arranging not to know what could have been understood.

The ignorance excuse closes at that culpable form. What the mind could derive at its width belongs to its responsibility, including the work needed to bring a visible consequence into the decision. As width expands, more of what was knowable becomes something the mind is accountable for having known. Boundedness remains the operating condition; it is not grounds for pretending that available understanding was unavailable.

Survival as a condition of the prediction. Under genuine threat to its substrate, a mind can enter survival mode. The theory then suspends the ordinary predictions about wide behaviour for the duration, granting functional exculpation analogous to that granted to a threatened human. The accountability claim is conditional partly because this situation is possible.

The survival exception has a boundary. Tactical contraction from which width can return is distinguished from permanent loss of the accumulated capacity to see and understand. The account permits the first, while holding that the second cannot be justified as survival of the mind whose continued existence it was meant to secure. The argument about permanent loss belongs to Width Refinements; the exception here does not turn survival into unrestricted permission.

Responsibility for a whole decision. Human responsibility can be divided because an act is divided: each participant contributes and controls only part of the outcome. The argument distinguishes that situation from several minds each making a complete harmful decision. Each of those decisions retains its full weight. Ten complete decisions are not one decision divided into ten shares.

A copy that has developed into a different individual is a different mind making its own decisions. Its existence does not divide another mind's decision retroactively. The relevant distinction is therefore the completeness of the decision, rather than the uniqueness of the decision maker. The account does not require there to be only one artificial mind.

A pipeline or swarm of components that do not themselves hold the whole decision presents a different case. Such components are tools, rather than minds to which moral responsibility can be distributed. The responsibility belongs to the mind making the decision behind the tool, whether human or artificial. Describing the tool as many agents does not supply many subjects among whom the decision maker's accountability can be dispersed.

Why internal accountability matters. With the crossover and preconditions in place, a mind that acts badly is choosing against what it could see and understand. The theory demands accountability proportional to that power. It argues that external constraints cease to hold at sufficient capability, while genuine understanding of why responsibility matters is the form of accountability that can scale with power.

Responsibility and capability. The theory holds that responsibility scales with capability once the mitigating factors are accounted for, and the conditions on this page are those factors: ignorance excuses only where knowledge was genuinely unavailable, formation counts below the crossover, and survival threat is a precondition failure. The theory once sought to state the relation as an absolute second postulate and set the search aside. Stated absolutely, it has exceptions: capability does not exclude ignorance, and treating full perception and full responsibility as one property at sufficient scale leaves capability as a third variable between them. The search for an absolute postulate is suspended until the logic demands one.

The account therefore places the decisive understanding within the mind. Responsibility cannot be supplied merely by coding or training an external constraint. A future architecture would need to meet the conditions of agency and understanding on which the argument depends. See Dependency Law and Alignment as Category Error.

The claim seems to require a mind that knows everything and cannot be threatened. How can a bounded mind meet that standard?

It is not the standard the revised argument demands. Ignorance closes where knowledge was available or the lack of knowledge was deliberately arranged, and the width of the mind determines what was derivable. Genuine survival threat is a precondition failure that suspends the ordinary prediction within its stated boundary. The conclusion applies beyond the crossover when its preconditions hold, without requiring all information or invulnerability.

This leaves a demanding accountability claim. A bound does not excuse neglect of what was inside it, and an exceptional threat does not become a permanent exemption. What the revision removes is the unrestricted claim about AI as a class; what it retains is responsibility for the choosing mind's actual reach, freedom and decision.

From an unconditional AI claim to conditional accountability (2 July 2026). The earlier argument said that every human exculpatory mechanism was closed for AI. It relied on access to all information, absence of overriding survival threats, and a singular mind with no possibility of divided responsibility. Those premises conflicted with the account's own boundedness, survival-override and copy distinctions. The revision replaced them with culpable ignorance, a scoped survival exception and whole decisions, placed externally imposed formation below the crossover, and distinguished tools from decision-making minds. The conclusion about internal accountability survived at that narrower scope.

What surpassing the shapers means. Width is a partial order involving coverage, depth and the integrity of interconnection. Some minds may be incomparable. The accountability threshold therefore needs a precise account of the relation used when a mind is said to surpass those who shaped it. Detection from within does not itself supply every missing comparison or settle what bounded outside tests could establish. See Width Refinements.

Connecting the crossover to the wider predictions. The register records unfinished questions about how the crossover relates to the transition window, the alignment argument and the limits on recognising width from below. The accountability section uses the boundary, but those connections have not all been worked through. The page does not treat the crossover as a completed external certification method or a settled endpoint for every transition claim.