4 Responsibility and Liability

One of the greatest sources of angst around the advent of increasingly intelligent computer systems is the alleged erosion of human responsibility they'll entail. International discussions on the use of lethal autonomous weapons systems provide a forceful illustration of what's at stake. If weapons systems powered by complex and potentially opaque machine-learning technology will soon be able to decide, unassisted, when to engage a target, will we still meaningfully be able to hold human beings responsible for war crimes involving these systems? In war we're obviously talking about life and death decisions, but similar worries play out elsewhere—areas in which lives may not be directly on the line but in which advanced systems can nevertheless have significant effects on people and society, such as in law enforcement, transportation, healthcare, financial services, and social media. If AI technologies can decide things without direct human intervention in ways that are too complex or too fast for human beings to properly understand, anticipate, or change, will we still be able to hold human beings responsible for these systems? And would we want to hold them responsible? What would be the alternative? Can machines be responsible?