CHAPTER 04 / INFORMATION
Where does the information an AI system uses come from?
Data and knowledge
Data are recorded observations, documents or other representations of information. AI systems use data during development and may receive additional information through prompts, retrieval and feedback.
Different roles for information
Training data help shape a model’s parameters. Evaluation data help test its performance. At use time, a prompt and its context supply the current task and relevant information. Keeping these roles distinct helps explain what a system can know and how it can be checked.
Preparation and permission
Collection is only the beginning. Teams may clean records, remove duplicates, label examples and examine whether the data reflect the intended population. Rights, privacy, provenance and access permissions affect what can be used. More data alone does not guarantee a better result.
Retrieval and feedback
Retrieval searches an external collection and brings selected information into the application’s context. It can make a source available for inspection, but selection errors or unsupported summaries remain possible. Feedback can reveal problems; it becomes training data only if a development process uses it that way.
Keep this connection in mind.
Ask where information came from, whether it is appropriate to use and whether the output is supported by it.
- Retrieval-enabled applications can retrieve from Information collections
- Machine-learning models trained using Training data
Sources for this introduction
- Google · Machine Learning Glossary ↗
Models, parameters, training, inference, datasets and evaluation.
- NIST · AI Risk Management Framework ↗
Human context, evaluation, accountability and risk management.
Draft teaching text, prepared with AI assistance. Larry’s editorial review is pending. Read the editorial approach.
EXPLORE RELATED AREAS