OECD Framework for the Classification of AI Systems
A user-friendly framework that classifies AI systems and examines their risks across five dimensions (mnemonic PEDMT). Privacy sits under People and planet; training-data collection method sits under Data and input.
The OECD Framework for the Classification of AI Systems is a user-friendly framework for classifying AI systems and examining their risks. Its purposes: promote a common understanding of AI, inform registries or inventories, support sector-specific frameworks, and support risk assessment and risk management.
| Dimension | What it covers |
|---|---|
| 1 · People and planet | Who could be affected → human rights, environment, society. Privacy considerations sit in this dimension. |
| 2 · Economic context | Sector (healthcare, finance), business function and model, critical or non-critical nature, deployment, impact and scale, technological maturity. |
| 3 · Data and input | Data and/or expert input (human knowledge codified into rules). How data was collected, by machine or human, its structure and format. Covers training data and production data. |
| 4 · | Computational representation of the environment. Technical type, how it is built (expert knowledge, ML, or both), how it is used (objectives, performance measures). |
| 5 · Tasks and output | Tasks performed (recognition, forecasting), outputs and resulting actions, action autonomy, combined task-action systems like autonomous vehicles, evaluation methods. |
PEDMT → "People Expect Decent Model Transparency" → People & planet · Economic context · Data & input · AI Model · Tasks & output.
If a question asks where privacy fits in the OECD framework → People and planet. If it asks where training data collection method fits → Data and input.