Course
AI ethics: how a machine learns to decide
An AI ethics course taught as mechanism: how a computer copies a pattern from the past, what rides along in the examples, what is new when a machine makes the mistake, and who answers for it.
The shape of it
15 lessons
About 74 minutes in total. Your progress is kept on this device — no account, nothing sent anywhere.
Course complete
Put your name on it and print it, or save it as a PDF. Nothing is sent anywhere.
The lessons
What a machine learns
3 lessons · about 14 min01A pattern copied from the past
Tell apart a rule a person wrote from a pattern a computer found in examples, and say what the pattern is made of.
About 5 min
02The goal is a number
Spot the number a machine is told to raise, and tell it apart from the thing people wanted.
About 5 min
03Never sure, only a score
Work out what moving the line does to the two kinds of mistake, and who carries each.
About 5 min
What rides along in the data
4 lessons · about 20 min04The past is in the examples
Trace how a pattern copied from old decisions carries their habits into new ones.
About 5 min
05Who the camera never saw
Work out why a model is worst on the people it saw least, and read the gap in a table of errors.
About 5 min
06The label was a stand-in
Spot the gap between the thing a model was told to predict and the thing people meant.
About 5 min
07Fair has more than one meaning
Tell apart two meanings of fair, and work out why one score cannot give both when two groups differ.
About 5 min
What is new about a machine
4 lessons · about 20 min08One mistake, a million times
Work out how one wrong rule becomes thousands of wrong letters, and why nobody sees it at once.
About 5 min
09Nobody can see inside
Say what it means that nobody can read a model's reason, and spot where that matters most.
About 5 min
10Fluent is not true
Tell apart a sentence that reads well from one that was checked, and spot where a chatbot invents.
About 5 min
11Who did the work
Trace the hands behind a chatbot's answer, and name the ones that were never asked.
About 5 min
Who answers for it
4 lessons · about 20 min12A chain of hands
Spot why blame slides down a chain of hands, and who is left holding it.
About 5 min
13A rule written down
Sort uses of a machine by how much harm they can do, the way the first big law does.
About 5 min
14A human in the loop
Tell a real check from a signature, and work out what makes a person able to say no.
About 5 min
15An exam grade, taken apart
Take one real machine decision apart with every tool in the course.
About 5 min