Location: Edgecliffe G03
Title: How to Build Morality into AI
Abstract: AI is spreading fast. We humans need to figure out the best way to prevent AI from making the worst decisions, which are harmful, unfair, or otherwise morally wrong. One way is to design AI to predict what humans would judge to be immoral if they were informed, rational, and impartial. Then the AI can use that information when making its choices. This talk will illustrate this method in a test case of who gets the kidney when two patients need a transplant but only one kidney is available. This same approach can be extended to other areas of morality, including end-of-life decisions in medicine, hiring and promotion in business, pretrial release in criminal law, and autonomous vehicles and weapons. The result will be AI that aligns with our deepest values.