When people disagree about whether an AI system is acting rightly, they are usually using different moral theories without realizing it. This module gives you the vocabulary to tell them apart. Consequentialism, and its best-known form utilitarianism, judges an action by its results. Deontology, associated with Kant, judges it by whether it follows duties and respects people as ends in themselves. Virtue ethics, from Aristotle, asks what a person of good character would do.
The lessons take the three theories in turn, then see them argued over in Michael Sandel’s famous trolley-problem lecture. The module ends by applying them to a real AI question, how a driverless car should be programmed to act in an unavoidable crash, where each theory points to a different answer.
Most AI ethics guidelines also draw on a shorter list of principles taken from bioethics, namely beneficence, non-maleficence, autonomy and justice, plus explicability for AI. As you watch, notice how those principles echo the theories above. Watch the lessons in order, then continue on to Bias & Fairness.
Explains the principle of utility and the difference between act and rule utilitarianism, the main form of consequence-based ethics.
Video 10minCovers hypothetical and categorical imperatives, the universalizability principle and treating people as ends in themselves rather than mere means.
Video 10minIntroduces virtue ethics, including the Golden Mean, moral exemplars and the idea of eudaimonia, or flourishing.
Video 9minHarvard professor Michael Sandel opens his course with the trolley problem and other dilemmas that pit consequences against duties.
Video 55minIyad Rahwan shows how the trolley problem becomes a real design question for self-driving cars, and why different moral theories give different answers.
Video 13min