- Duration
- 5 min
The Responsibility Gap
When an autonomous system causes harm, where does the responsibility go?
A deployed system causes a harm no one intended. Distribute responsibility across the people and the system involved. Then compare your allocation with the classic philosophical positions and see whether a gap remains.
The experiment
In 2004 Andreas Matthias argued that learning machines create a “responsibility gap”: their behaviour is not sufficiently under the control of any human for that human to be blamed, and the machine is not a fit subject of blame either. The argument has structured the debate ever since.
Here you are given a scenario and a hundred units of responsibility. Allocate them. The instrument then shows how your allocation compares to the positions in the literature — and how much responsibility, on your own view, ended up nowhere.
What to notice. Most people leave a residue unassigned. Ask whether that residue is a genuine gap or a sign that you are looking for a single culprit where the correct answer is a distribution.
Other instruments
The Mind Detector
What are you actually measuring when you decide something has a mind?
Run experimentShip of Theseus, Digital
At what point in a gradual replacement do you stop being you?
Run experimentThe Epistemic Chain
Where does responsibility for a belief live when it arrives through a machine?
Run experiment