THEORY-561: If moral evaluation has thresholds below which effects are negligible, a system of agents can do good or evil that none of its members' interactions does, and the same thresholds make an environment tolerant of small evils and inert to small goods; on the examples offered the non-additivity lies in the evaluation, as a step function of summed impact, not in the actions

Source

Floridi (2013), “Distributed Morality in an Information Society”, LIT-901, read in full in NOTE-698.

What was actually shown

A model, drawn in Figures 1–3, with four examples and a policy programme.

• The phenomenon (p. 729). Distributed morality covers “moral actions that are the result of otherwise morally-neutral or at least morally-negligible … interactions among agents constituting a multiagent system, which might be human, artificial, or hybrid”. “Can ‘big’ morally-loaded actions … be the result of many, ‘small’ morally-neutral or morally-negligible interactions? I hold the answer to be yes.”
• Thresholds (pp. 730–731). Left unchecked, consequentialist promotion of every action to good or evil leaves no neutral actions, which Floridi calls “too implausible to be acceptable”. The fix is the morally negligible, modelled as two thresholds “the value of which can be left unspecified”. The result is moral inertia: “most actions are morally neutral and tend to stay that way”.
• Resilience and inertia (pp. 732–733). Possible evils that stay below threshold make environments “morally resilient”, “fault-tolerant”. Possible goods that stay below make them “morally inert … potential goodness can be too weak to become actual goodness”. “It is only by aggregating and merging individual courses of action that a moral difference is made” (p. 733).
• The examples (pp. 734–736). The tragedy of the commons is named as negative distributed morality and not discussed. The positive cases are cause-marketing, the Co-operative Bank’s charity cards (£3 million to Oxfam, 1994–2007), JustGiving and peer-to-peer lending.
• The programme (p. 736). Aggregate possible goods, fragment possible evils, and build “moral enablers”. Their ensemble is infraethics, a first-order framework of expectations, attitudes and practices (trust, respect, privacy, transparency) that facilitates moral action without being morally good in itself (pp. 738–740).

The first claim follows from the model once thresholds are granted (NOTE-698, C1, moderate). Nothing else could have come out otherwise, since nothing is tested.

What this does not say

• That distributed morality is the moral counterpart of distributed knowledge. In the epistemic case different contents (P∨Q and ¬P) combine logically into one neither member holds to any degree. Every moral example sums contributions of one kind (pounds, purchases, loans, bike rides), each already carrying “some moral value” (p. 733). What makes them distributed morality is only that evaluation is a step function of the sum. No aggregation operator for actions is given (fn. 2) (NOTE-698, limitations).
• Where the line falls. A thousand £100 donations, each above threshold, are not distributed morality; a million 25p contributions, each below it, are. Nothing but the unspecified threshold separates them.
• That it happens without design. The definition’s “invisible hand” (p. 728) fits the commons. All four positive examples are aggregators somebody designed, and the paper says that “unless A and B interact properly” the outcome does not emerge (p. 733).
• That fragmentation works. It is named and never illustrated.
• That enablers are morally neutral. The abstract says so; the body says only that infraethics “is not necessarily morally good in itself”, concedes that the separation “never occurs in reality” (p. 739), and lists as enablers things usually called values.
• Who is responsible. Responsibility for distributed moral outcomes, promised by Floridi & Sanders 2004 (LIT-898), is not given here. It is the subject of the 2016 paper (LIT-899, THEORY-207).
• That the “supra-agent” is an agent. No criterion of agency is applied to it; it is an agent by fiat at a level of abstraction that hides the members (fn. 1). It bears on QUESTION-009 only by showing how freely the word is used.

Connections

• THEORY-207 (Floridi 2016) proposes allocating responsibility for such outcomes in full to every causally accountable node, and takes the phenomenon stated here as its premise.
• Ostrom’s commons institutions (LIT-719, NOTE-563) argue against Hardin’s tragedy, which Floridi takes as his paradigm of negative distributed morality. In his terms her monitoring, sanctions and communication are infraethics and fragmentation; THEORY-149 states what she found.
• The distributed-knowledge example is the epistemic face of List and Pettit’s supervenience result (THEORY-127), on which the aggregation used can make a group depart from what every member holds. Floridi does not take that step.

Note of 2026-10-29: what holds it back

Read in full (NOTE-698). The title already claims only the conditional the threshold model yields, and the reading of the examples. The promote_when asks for the unconditional account Floridi asserts. The record holds neither a threshold rule stated before the cases nor a case of heterogeneous, individually neutral acts combining non-additively; every example is a designed aggregator summing like contributions. Promoting the conditional would need a kind-1 reading to carry the unconditional form and its promote_when (ADR-047), and it would free nothing: THEORY-207, its one dependent, is held by its own unmet promote_when.