The responsibility gap

The responsibility gap is the worry that learning machines can produce outcomes for which no one can fittingly be held responsible. Matthias argued that when a system’s behaviour is shaped by learning during operation, neither its manufacturer nor its operator can predict or control it in the way traditional ascriptions of responsibility require. Sparrow pressed the point for autonomous weapons, where the gap would leave no one answerable for a killing. Santoni de Sio and Mecacci distinguished four gaps, in culpability, moral accountability, public accountability and active responsibility, and argued that they call for different remedies, including meaningful human control.

Responses divide. Some deny the gap, holding that designers, deployers or users remain responsible under existing principles. Others accept it and seek to fill it, by distributing responsibility across many hands, as Nissenbaum’s account of accountability in computerised society anticipated, or by assigning it to institutions. A few propose that the systems themselves be held accountable in some functional sense.

I meet a version of the problem in two places. In the epistemology of AI it is often assumed that because deployers answer for a system, they stand behind its outputs as a speaker stands behind what she asserts. They do not. A deployer has not seen the output, does not believe it, and disclaims it in the terms of service. Responsibility for putting a system into the world is not authorship of its sentences. Derivative and distributed responsibility are real, and they are how we already treat certified instruments, but they leave no one in the position of an asserter. That is why treating a deployer as the principal behind a model’s assertions fails as a route to AI testimony.

In algorithmic welfare, the gap appears when a fingerprint fails to read and the failure is described as a glitch, an event without an author. The Constitutive Obligation Principle closes that gap by locating obligations of legibility, non-abandonment and remediation in the institution itself. They are non-waivable and persist whatever any individual has or has not authorised. The failure is then not a hole into which accountability disappears but an obligation the institution did not discharge.