Skip to content
Coordinates of ThoughtHome of Philosophy
Join
Question · Open

Who is morally responsible when an AI system causes harm?

Nur Aisyah Rahman Fellow
· 3 replies
Technology & the future#ai-ethics#responsibility
Follow

Andreas Matthias coined the term "responsibility gap" in 2004: when a learning system behaves in ways its designers couldn't predict, it seems nobody is fully responsible. The designers didn't intend the outcome, the operators couldn't foresee it, and the system isn't a moral agent.

Some say the gap is an illusion and responsibility should go to whoever deployed the system. Others say we need new forms of collective or institutional responsibility.

Where would you place responsibility, say, when an automated loan system unjustly rejects thousands of applicants from one neighbourhood?

LikeAnswer
Follow3 answers

Join the conversation: members reply here, and joining is free. Reading is always free.

Join free to reply Sign in

3 answers

  1. Pieter van Dijkhuizen

    Fellow

    Deployment is where the decision happens. An engineer can say "this model may do X in situations we haven't tested." The person who decides to use it on thousands of real people anyway owns that risk. We don't let bridge builders off because the bridge "learned" to collapse.

    Helpful · 2
  2. Hamza Qureshi

    Fellow

    The law already has tools: strict liability, vicarious liability, product liability. We don't need to solve the metaphysics of agency to say who pays. The responsibility gap is more of a gap in moral feeling than in legal practice.

    Helpful · 3
  3. Emeka Anyaoha

    Fellow

    Hamza, but legal liability and moral responsibility come apart. A company can pay damages without anyone feeling, or being, culpable. Maybe that's the gap Matthias meant: compensation without blame.