Data Literacy and Data Management for AI
What to know
Data literacy means reading, questioning, and using data responsibly. Know the data lifecycle: collection, cleaning, storage, access, use, retention, and deletion. Understand structured vs unstructured data, data quality dimensions (accuracy, completeness, consistency, timeliness, uniqueness), and the garbage in, garbage out principle - model quality is bounded by training and input data quality.
Key concepts: metadata (data about data), data lineage (where data came from and how it changed), master data, data warehouse vs data lake, and schema. Understand bias in data - sampling bias, historical bias, missing groups - and how it propagates into model outputs. Know basic statistics vocabulary: correlation vs causation, sample vs population, and base rates.
Traps
Questions test whether you confuse correlation with causation, or treat a large dataset as automatically representative. Another trap: assuming cleaning is a one-off task rather than continuous monitoring. Watch for options that ignore data provenance when an AI output is challenged - without lineage you cannot audit or correct it.
Worked example
A churn model performs well overall but poorly for a small customer segment. Investigation shows that segment is underrepresented in historical data. The fix is not a better algorithm but better data coverage and segment-level evaluation. Rule of thumb: check who is missing from the data before tuning the model.