Trust and safety
Accuracy, deception, privacy, bias and attacks.
The chapter at a glance
AI claim
Original source
Date and context
Independent confirmation
Publish or correct
In this chapter
- 1How do you check an AI answer?For an important claim, do three things: find the source, check the original and look for independent confirmation.
- 2Why can AI be biased?Models can repeat inequalities found in their data, their goals and the way they are used.
- 3What does AI mean for privacy?With AI, both the training data and the data we type in while using it matter.
- 4How can someone abuse an AI agent?An attacker can hide instructions in a document, message or web page that an agent wrongly treats as commands.
- 5Deepfakes: how is a fake recording made?AI can create or alter an image, a sound or a video so that it looks like a real person or event.
- 6How do we measure whether AI is any good?Quality has to be measured for a specific task, group of users and set of conditions.
- 7Why do we rely on AI too much?People can trust a system that sounds sure of itself too much, or only check the result they expect.
- 8How is AI tested before use?Beyond the usual measurements, a system can be deliberately exposed to difficult and unexpected cases.