Rules of Engagement

Show notes

Long before there were algorithms, command structures had already worked out a discipline of accountability — not as a slogan but as a practice. A decision of consequence is made by someone who holds the authority to make it, and that authority was delegated to them on the record by someone above. When the decision is later examined, the question is never only was it correct? It is who was authorized to make this, what were they told, and could they have decided otherwise? The chain runs in two directions at once: downward as delegated authority, upward as answerability. This episode is about what that two-way chain requires, and it is a conversation about principles — the structure of responsibility, not the conduct of any operation.

Decision support changes where the difficulty lands. A machine that recommends, ranks options, flags some inputs and suppresses others, or assembles the picture a person finally acts on has quietly entered the chain — but it cannot hold authority, and it cannot answer for anything. So the burden does not disappear; it moves back onto the human who signed. The doctrine of meaningful human control is exactly the demand that the person in the loop be more than a formality: that they had the standing, the information, and the genuine room to decide otherwise, rather than ratifying a recommendation they were never positioned to refuse. When control is meaningful, the human is accountable because the decision was really theirs. When it is not, accountability has been laundered through a system that can never be called to account.

The frame this show keeps returning to is Toulmin's: a claim stands on grounds, licensed by a warrant, and the warrant in turn rests on its backing — the deeper authority that makes the inference legitimate in the first place. For a command decision, the backing is the structure of responsibility itself: who delegated what authority to whom, on what record, and who must therefore answer. An automated recommendation can supply grounds and even propose a warrant, but it cannot supply backing — it has no authority of its own to lend. That is why "the system recommended it" is not an account of a decision; it names a step without naming who stood behind it. The book this series accompanies, Admissible Reality, makes the same demand of any consequential machine-mediated claim: that you can always reconstruct who signed, on what basis, and under what authority — and that the answer is a person, not a process.

So the conversation puts a scholar of command accountability beside the question every organization adopting decision support will eventually face: when the assist becomes routine, does the human in the loop still own the call? After the fact — in the honest reconstruction every serious command structure already runs — can you separate what the person decided from what the system decided for them, and find a name at the point of responsibility? The wager of the episode is that the principle was never about machines at all. It is about standing, signature, and the refusal to let answerability dissolve into "the tool said so" — and that this is precisely the principle automated decision support is about to test.