The responsibility gap denotes the phenomenon that, in cases of harm caused by autonomous or learning systems, no personal agency can any longer be held accountable in the classical sense. The term was coined by Andreas Matthias in 2004 in Ethics and Information Technology and sharpened by Robert Sparrow in 2007 with regard to the military application of lethal autonomous systems.
The Structure of the Gap
When a self-learning or autonomously acting system causes harm, the classical imputation falls apart across several possible agencies, each of which, taken on its own, is not fully liable:
- The manufacturer/programmer. He designed the system but did not foresee the concrete action — particularly in machine learning, the system develops behaviors that were not explicitly contained in the training data.
- The operator. He deployed the system but did not authorize the concrete decision, because the system acts autonomously.
- The commander/principal. He set the operational framework but did not determine the individual action.
- The machine itself. It acted — but as a robot it is something, not someone. It cannot bear culpability, because it possesses neither freedom nor conscience nor reason.
The result: the harm has occurred, but no personal address of imputation bears full responsibility. Responsibility is structurally fragmented.
Why This Is Not a Technical Problem
The temptation is to close the gap technically — for instance through better logging, through a black box recorder for AI systems, or through formal liability rules. Personalist ontology shows why this can succeed only in part: responsibility is not a merely juridical construct but arises from the personal structure of action. According to the principle agere sequitur esse, only that being can act — and therefore only that being can be responsible — whose being is capable of bringing forth action: a being with reason, freedom, and moral self-determination. This is precisely what every machine lacks. Imputation can therefore always be traced back only to the person behind the machine — and where this tracing-back is structurally interrupted, a genuine gap arises, not a merely formal one.
Four Types according to Santoni de Sio and Mecacci
Santoni de Sio and Mecacci (2021) distinguish four manifestations:
- Culpability gap: No one has acted in a morally culpable manner.
- Moral liability gap: The obligation to make compensation cannot be unambiguously assigned.
- Public accountability gap: Institutional accountability remains unclear.
- Active responsibility gap: No one feels proactively responsible for the overall system.
Each of these gaps corresponds to an aspect of what classical ethics held together as the responsibility of a person — and what falls apart in distributed socio-technical systems.
The Military Sharpening
The gap shows itself most acutely in the case of lethal autonomous weapon systems. Sparrow argues: international humanitarian law (jus in bello) presupposes that every killing is imputable to a personally identifiable agency. Where this imputation becomes structurally impossible, the deployment is impermissible not only ethically but also under international law. From this he derives the prohibition postulate: systems that preclude such imputation must be banned. The UN General Assembly institutionally confirmed this position on 6 November 2025 (First Committee, 164 in favor / 6 against / 7 abstentions).
The Answer of Personalist Ontology
Personalist ontology responds to the responsibility gap not primarily with new rules but with a protective line: where responsibility would structurally disappear, autonomization is to be limited. In practical terms this means that degrees of autonomy must be chosen such that a person remains meaningfully imputable at all times (meaningful human control). The gap is then not a technical defect to be closed but a boundary at which the machine must come to a halt.
Ontological Classification
Typically arises in: Fully autonomous systems, learning systems without meaningful human control
Causally responsible for: Erosion of responsibility, hollowing-out of right
Counter-measure: Human-in-the-loop, meaningful human control
Chapter assignment: Chapter 5: Oblivion of the Person (German)
Sources: Generated by querying the Personhood ontology.
Further sources:
- Matthias, Andreas (2004): “The responsibility gap: Ascribing responsibility for the actions of learning automata.” Ethics and Information Technology 6(3), pp. 175–183.
- Sparrow, Robert (2007): “Killer Robots.” Journal of Applied Philosophy 24(1), pp. 62–77.
- Santoni de Sio, Filippo / Mecacci, Giulio (2021): “Four Responsibility Gaps with Artificial Intelligence: Why they Matter and How to Address them.” Philosophy & Technology 34, pp. 1057–1084.
- International Committee of the Red Cross (2021): ICRC position on autonomous weapon systems. Geneva.
- Heyns, Christof (2013): Report of the Special Rapporteur on extrajudicial, summary or arbitrary executions: Lethal autonomous robotics and the protection of life. UN doc. A/HRC/23/47.