🇩🇪 Deutsche Version: Folgenbezogener Einwand

Note: The ethical judgments on this page refer exclusively to the action — never to the person who performs it or who came into existence through it. Cf. Note on Ethical Judgments.

A consequence-based objection is directed against effects, risks or attendant circumstances of a practice and lapses when these lapse. It is remediable by procedure — and is therefore regularly answered by a set of rules. Its counterpart is the act-based objection, which is directed against the action itself, independently of circumstances, diligence and consequences, and is remediable by no procedure.

The distinction between the two is not classificatory but practical: it decides whether an objection can be cleared away by improvement of the procedure. The consequence-based objection is not thereby the weaker one. It is the more effective one — because it is addressed to an address that can answer.

Delimitation: objection to a practice, not to a concept

The page Objection treats something else: objections to a concept of person, that is, theoretical objections in the controversy over what a person is. Here it is a matter of objections to a practice, that is, to an acting. The distinction between consequence-based and act-based takes hold only in the second case; applied to a concept it would make no sense, for a concept has no consequences in the sense in which an action has them.

Ontological classification

is a subclass of: objection to a practice
presupposes: remediability by a procedure
is typically answered by: governance framework
stands in connection with: responsibility gap, nominal oversight
to be distinguished from: act-based objection
not to be confused with: objection to a concept of person

The typical cases

The objections that are actually brought forward in the debate about algorithmic decisions are overwhelmingly consequence-based. Five stand in the foreground.

Biased training data. A model that was developed on a non-representative population systematically disadvantages those whom the data do not depict. The objection lapses as soon as the data basis is broadened and the bias examined.

Lacking explainability. An output whose coming about no one can retrace withdraws itself from examination. The objection lapses as soon as the load-bearing features are disclosed.

Diffusion of responsibility. Between manufacturer, operator and user the attribution disperses until no one has performed the decision, although it was performed — the responsibility gap. The objection lapses as soon as a named office vouches and can vouch.

Inadequate information. Whoever does not know that and how a system was involved in a decision concerning him has not effectively consented. The objection lapses with an act of information that names this circumstance.

Unequal access. A procedure that stands open only to those able to pay distributes its benefit unjustly. The objection lapses with the extension of access.

In all five the same pattern shows itself: one can state a condition under which the objection disappears. Precisely in this lies the touchstone for the assignment — let one think the objected-to effect entirely away and ask whether anything remains. If nothing remains, the objection was consequence-based.

What a consequence-based objection co-signs

Here lies the point peculiar to this page: a consequence-based objection often concedes more than it says. It criticizes a practice and confirms it in the same stroke, because it proceeds from its fundamental admissibility — otherwise its own demand would be pointless.

This becomes clearest with unequal access. Whoever objects that a procedure stands open only to the wealthy already presupposes that the procedure is in principle of benefit — one does not lament the unequal distribution of an evil. The objection, however warranted it is under its own presupposition, treats the wider availability of the practice as a good. It is thereby no longer a merely procedural statement but contains a substantive judgment about the matter.

Less conspicuous, but of the same construction, is the objection from bias. Whoever demands that a selection model must rest on representative data demands that it select better — and has thereby approved the selection itself. Whoever demands better explainability demands that what happens anyway become retraceable. A framework of consequence-based demands is therefore never neutral towards the fundamental question; it answers it tacitly in the affirmative.

That is no reproach. It is a piece of information about range: whoever raises a consequence-based objection should know what he co-signs, and whoever hears only consequence-based objections should not conclude that no others obtain.

Mixed cases

Real objections frequently occur mixed, and the assignment is then an act of interpretation, not a reading-off. The objection to nominal oversight, for instance, is in form consequence-based — it objects to an organizational deficit and would be remediable by examinable systems and sufficient time. In substance, however, behind it there may stand the thought that a decision of this kind may not be handed over to a machine at all, no matter how well the oversight is organized — and that would be an act-based objection.

Whoever assigns an objection therefore makes a decision of interpretation for which he owes reasons. The assignment is not the reading-off of a characteristic on the objection but a proposal as to how it is to be understood — and this proposal is accessible to criticism. In particular, the speaker himself is not always the best source of information: people often bring forward consequence-based reasons because these connect better in public debates, while their actual misgiving is act-based. The suspicion is to be handled reciprocally; it must not serve to impute a hidden dogma to every consequence-based objection.

The strongest objection to this assessment

The most serious counter-objection is directed against the subordination that resonates in this presentation: if consequence-based objections tacitly approve the practice, they appear as the weaker instrument, as haggling over details, while the real question lies elsewhere. This reading is false, and for two reasons.

First, harms are real. Biased models disadvantage real people, inadequate information devalues real consents, unregulated attribution leaves real affected parties without an address. An objection that remedies such things has achieved more than a position of principle that changes nothing. Whoever dismisses consequence-based criticism as mere bargaining material spares himself the effort that actually has effect.

Second, the consequence-based objection has a structural advantage that is often not credited to it: it is examinable. It names a condition by which its success or failure can be measured, and it compels the other side to verifiable commitments. The act-based objection cannot do this — it can only be contested or acknowledged, and where it is contested, the conversation ends. The consequence-based objection keeps it open.

What remains open: neither advantage changes the fact that a consequence-based objection cannot represent an act-based one. Whoever rejects on fundamental grounds and argues on tactical grounds only in consequence-based terms, because that finds a hearing more easily, risks that his demands are met and his position thereby disappears — he has negotiated away what he actually wanted to contest. Honesty here demands that both be said side by side: what must be improved if the practice continues, and what one holds fast to even when everything has been improved.

Was den Menschen zum Menschen macht

Übersicht

Consequence-Based Objection

Delimitation: objection to a practice, not to a concept

Ontological classification

The typical cases

What a consequence-based objection co-signs

Mixed cases

The strongest objection to this assessment

See also

Graphansicht

Inhaltsverzeichnis

Backlinks