Learn / AI Literacy Bridge / Unit 2: Machines that learn from data
Data is the curriculum
For a learning system, the data is the curriculum. Whatever is in it — and whatever is missing — shapes what the model learns.
Three questions to ask about any dataset:
- Who or what is represented? A skin-condition model trained mostly on light skin will be less accurate on dark skin. A voice assistant trained on a few accents will mishear others.
- Who labelled it, and how? Labels are human judgements. If labellers disagree, or were rushed, or brought their own assumptions, the model inherits that.
- Is it still current? Language, prices and behaviour change. A model trained on old data can be confidently out of date.
Good AI teams spend more time on data than on algorithms: collecting it carefully, documenting where it came from, checking it for gaps, and measuring performance separately for different groups of people.
When you meet an AI system, a powerful question is simply: “What was it trained on?” The answer often explains both its strengths and its blind spots.