L1-M4 / lesson 1 of 2

Recognise harm beyond a wrong answer

An internal team reviews a fictional recommendation tool.

10–15 min · 0/4 steps complete

Start here / A workplace decision

One accuracy number hides different impacts

An internal team reviews a fictional recommendation tool.

The average success rate looks strong.

Before you read / Think it through

In “One accuracy number hides different impacts”, what would you check or do first?

Pause and form your answer, then compare the reasoning. This reflection is not scored.

Compare the reasoning

Ask who experiences the errors and what those errors do to people.

The ideas below give you a method for this decision. Then work through the rest of this case and try the decisions yourself.

Part 1 of 4 Learn the ideas

Choose a topic0 of 5 read

Bias in inputs and outputs

Training data and design choices can create uneven errors.

A proxy can reproduce a protected characteristic indirectly.

In the workplace

A postcode can correlate with disadvantage without explicitly recording ethnicity. Look at effects rather than only field names.

Remember this

Plausible output still needs context and fairness review.

The workplace case

YOUR CASE / RISKS BEYOND ACCURACY

A manager receives a convincing voice message apparently from the finance director requesting an urgent supplier payment. Separately, HR proposes an AI video tool to infer candidates' emotions. Marketing wants to publish an AI image resembling a competitor's logo.

Quick reference / five key ideas

Bias in inputs and outputs: Plausible output still needs context and fairness review.
Unequal errors: Escalate consequential disparities before deployment.
Human judgement: A signature alone is not meaningful oversight.
Rights in inputs: Text-and-data-mining exceptions have conditions.
Rights in outputs: Vendor terms do not remove third-party rights.