When the Map Runs Out

A graduate student teaches a neural network to drive a racing game and discovers why watching an expert is not enough. This episode follows that insight through Waymo’s twenty million miles to the two kinds of uncertainty, and why more data solves one and can never solve the other.

Sources mentioned:

• Reinforcement learning, explained with a minimum of math — Timothy B. Lee, Understanding AI
• A Reduction of Imitation Learning to Structured Prediction — Stéphane Ross et al., the DAgger paper
• Waymo One: a year of firsts — Waymo
• The Waymo Driver — twenty million miles of real-world driving

Written by Tabrez Syed. Narrated by an AI voice. A Mandalivia production.