Week 4: Responsible Power
Day 22: Bias, Safety, and Limitations
Goal
Understand why AI systems can be unfair, unsafe, or overconfident.
Learn
AI can reproduce bias from training data, miss minority contexts, overgeneralize, refuse legitimate requests, or comply with harmful ones. Safety work reduces risk but does not remove it. Responsible use means testing, checking edge cases, using inclusive examples, and keeping humans accountable.
Behind the scenes
AI tools are products wrapped around models, data, prompts, retrieval, safety systems, and user interfaces. The better you understand the wrapper, the better your results get.Example
Example: in a real workflow, this idea helps you decide how to use AI carefully. For this lesson, connect the goal to one task you already do: understand why AI systems can be unfair, unsafe, or overconfident..
Practice
Ask an AI for advice on a sensitive topic, then critique what context it missed and what disclaimer or source would improve it.
Checkpoint
Checkpoint
You can identify limitations instead of assuming neutrality.