148: The Trolley Problem

Episode Summary

Executive Summary: The episode uses the trolley problem to explore how self-driving cars force society to encode moral decisions into algorithms. It contrasts intuitive public responses to sacrificing one life to save many with people’s reluctance to own a vehicle that might sacrifice its passenger, highlighting the gap between abstract ethics and personal self-interest. The host argues these decisions must be addressed collectively and in advance, including rules against discriminatory outcomes.

Main Topics: The trolley problem as a moral thought experiment: The host explains the classic trolley dilemma: choosing between letting five workers die or diverting the trolley to kill one worker instead. Bridge variant and intuition about direct harm: A second version adds a large person on a bridge; most people reject pushing them, even though it saves five, because it feels like direct murder. Public attitudes toward autonomous vehicle ethics: The episode notes a mismatch between what people say a car should do in a moral dilemma and whether they would buy a car programmed to sacrifice its passenger. Self-driving cars and programmed ethical decisions: Autonomous vehicles are framed as real-world systems that will need built-in rules for unavoidable crash scenarios and life-or-death tradeoffs. Need for ethics in data science and AI: The host emphasizes that these decisions cannot be left to individual programmers and should be considered at company, national, or global levels. Anti-discrimination principles for autonomous systems: Germany’s reported law is cited as an example of regulating autonomous vehicles so they do not discriminate based on protected or social characteristics. Neurological and research-based study of moral judgment: The host points to Radiolab’s deeper reporting, including brain research using MRI to study how people respond to these dilemmas.

Key Arguments: People usually endorse utilitarian outcomes in abstract form, preferring to save five lives even if one person dies. When the harm is made direct and personal, many people reject the same utilitarian choice because it feels morally like murder. Autonomous cars will require predefined ethical decision rules because they may face unavoidable accident scenarios. There is a contradiction between wanting vehicles to make the “right” moral choice and refusing to buy a car that might sacrifice its passenger. Ethical programming for AI and self-driving cars should not be left to ad hoc developer judgment; it requires broader societal and policy discussion. Autonomous vehicles could potentially discriminate using data about occupants or pedestrians, so regulation is needed to prevent unethical outcomes.

Data Points: Workers in classic trolley scenario: 5 - Five workers are on the track in the standard trolley problem Alternative victim in classic trolley scenario: 1 - One worker dies if the lever is pulled to divert the trolley Survey response on lever version: 90% - Radiolab street survey: most people said they would pull the lever Survey response on bridge version: 90% - Radiolab street survey: most people said they would not push the person off the bridge Radiolab episode length: 42 minutes - Driverless Dilemma episode mentioned at the end Radiolab episode release date: 27 September 2017 - Date given for the Driverless Dilemma episode Self-driving car public availability: By 2021 - Host says self-driving Ubers were expected and public ownership would arrive around then

Pivotal Quotes: "you can push the large person off the bridge onto the tracks and he will die, but he will stop the trolley and therefore save the other five people" — Host: Explaining the bridge version of the trolley problem "Most people said they won't buy a car like that." — Host: Describing public reluctance to own a self-driving car that might sacrifice the passenger "these questions need to somehow be addressed in advance" — Host: Arguing that ethical rules for autonomous systems must be decided before deployment

Implications: Self-driving cars will force society to formalize moral rules for unavoidable harm. The episode suggests regulators, companies, and the public must decide now how to balance utility, fairness, and personal risk before autonomous AI systems become widespread.