Week 0 - Why AI Safety Exists
IN THIS LESSON
AI is Advancing Faster Than Control
Artificial intelligence is no longer confined to research labs. It diagnoses patients, drives cars, screens job candidates, moderates speech, and shapes public opinion. In 2023 alone, the FDA approved 223 AI-enabled medical devices, up from just six in 2015. Autonomous vehicles now deliver hundreds of thousands of rides per week across the U.S. and China.
Yet our ability to predict, evaluate, and govern failures is lagging behind. AI safety exists because we increasingly deploy systems we do not fully understand, cannot reliably explain, and struggle to shut down once deployed at scale.
This week introduces:
What we mean by “AI” (models, systems, and socio-technical deployments)
Why accelerating capabilities create novel risk profiles
The distinction between AI usefulness and AI controllability
The core premise of AI safety: preventing harmful outcomes before they become irreversible
We end with a key question: If AI is already embedded everywhere, what exactly could go wrong?
