Source
Floridi (2016), LIT-899, read in full in NOTE-699. It is the second half of a pair with Floridi 2013 (LIT-901, read in NOTE-698), which defined distributed moral actions and left responsibility for them open, and it takes its agent conditions from Floridi & Sanders 2004 (LIT-898).
What was actually shown
That intention is not closed under causal implication, direct or distributed (p. 4), so no one intends an outcome that arises from interactions which are each neutral (pp. 3–4). That moral value can be assessed on the states of the affected system without information about the agents (pp. 5–6). Then, as a proposal, a five-step procedure (p. 7): identify the outcome, identify the network causally accountable for it, back-propagate responsibility so that each node is “prima facie equally and maximally responsible”, correct, and repeat. A node may override the default by showing no involvement (p. 8), and common knowledge of the rule is meant to make the repetition unnecessary (p. 8). The precedents offered are strict liability in tort and two Dutch Supreme Court rulings (pp. 8–9). In the 1948 cyclists ruling all three were held fully responsible, because each could easily rectify the situation. In the 1931 boats ruling only the fourth boat was responsible, because the other three could not easily do so.
Nothing in the paper could have come out the other way. It contains no test. The cases were chosen to illustrate it, and one of them goes against the default.
What this does not say
- Not that the allocated responsibility is moral responsibility in the usual sense. It is causal accountability turned into a corrective signal for agents that learn. The paper says it is “not … blaming or punishing” (p. 10). Floridi & Sanders 2004 called this accountability and kept “responsibility” for agents with intentional states (NOTE-701). This account is no rival to THEORY-121 or THEORY-147 on what moral responsibility requires. It is about a different thing.
- Not that the network is an agent or bears responsibility. Responsibility “remains ‘theirs’ (agents’) not ‘its’ (network’s)” (p. 8). The account bears on QUESTION-009 only in showing that allocation can proceed without an answer to it. It is compatible with THEORY-139, on which corporations and members can both be responsible, and it does not engage that account.
- Not a criterion of causal relevance. The theory of causation is deferred (fn 15, p. 6), and the permissiveness the paper grants in fn 11 is left to the override clause.
- Not that full and equal allocation is better than graduated allocation. The paper’s own remedies for risk aversion move towards allocation “proportionally to the ability of the agents to avoid the negative outcome” (p. 11). That is what THEORY-149 found in commons institutions that last.
- Not tested on artificial or hybrid nodes. The bot case is deferred (fn 20, p. 9).