Meeting KI-Postdocs: ‘What Machines Shouldn´t Do'

"Meaningful Human Control over the Evaluative"
(Dr. Scott Robbins)

In an increasingly autonomous world, it is becoming clear that one thing we cannot
delegate to machines is moral accountability. Machines cannot be held morally accountable or
responsible for their actions. This becomes problematic when a machine has an output that has a
significant impact on human beings. Examples of machines that have caused such impact are
widespread and include machines evaluating loan applications, evaluating criminals for sentencing,
targeting and killing people, driving cars, and whatever digital assistants are supposed to do. The
question that governments, NGOs, academics, and the general public are (or should be) asking is:
how do we keep meaningful human control (MHC) over these machines? I argue that keeping
meaningful human control over machines cannot be achieved without restricting the decisions we
delegate to machines to the descriptive.