Predictability and Understandability In Lethal Autonomous Weapons

Predictability and understandability are widely held to be vital qualities for lethal autonomous weapons (LAWS). However, many questions remain unanswered as to how these qualities can be defined, characterized, and enshrined in the development and real-world use of these technologies.

• What are predictability and understandability in a technical and operational sense? And what factors determine the degree to which a LAWS’s actions could be understood or anticipated?

• How can AI-based systems be made more predictable or explainable? And conversely, to what extent is some degree of unpredictability and unexplainability likely to be inherent in any complex autonomous weapon when used in the real-world?

• What constitutes “appropriate” or “sufficient” predictability and understandability for autonomous weapons, and how could these qualities be measured and assured?

This side-event to the 2021 GGE on LAWS unpacked these and other relevant considerations and pointed to concrete avenues of action for all stakeholders involved in the LAWS debate.

SPEAKERS:

Aude Billard, Full Professor, EPFL
Arthur Holland Michel, Associate Researcher, UNIDIR

EVENT DETAILS:

This event took place on 29 September 2021, 1:30pm-2:45pm CEST.