A task is the problem the model is asked to solve. The clause names the common families: regression, classification, clustering, anomaly detection, dimensionality reduction and others.
What the base framework establishes
The clause explains that a task identifies the problem to be solved with the model, that a system can have more than one, and that a trained model is applied to the problem instead of a function written by hand. Setting up a task means setting out the problem, the format of the data and the features. It then describes the families: regression predicts a continuous value; classification assigns an instance to one or more classes, binary, multi-class or multilabel; clustering groups similar instances without a human-defined target; anomaly detection finds unusual instances; dimensionality reduction reduces the feature space; and other tasks include segmentation, translation, speech and image generation. Structured prediction is offered as a way to connect several of them.
Why it matters to an AI management system
The task definition is the first place a governance review can catch a mismatch between the promise and the method. A model asked to classify a case is not the same as one asked to generate the next text. Generative systems make this urgent, because the task is open-ended and the output space is not a fixed set of classes.
The public draft summary names no change to 6.2 Task. The final text could still touch it.
Evidence a reviewer samples
| Category | Artefact |
|---|---|
| Task definition | The problem, input format, output format and success measure. |
| Task-to-outcome map | Why this task answers the business question and what it cannot answer. |
| Multi-task note | Where one system holds several tasks and how their outputs are combined. |
One question to prepare for. What exactly is the model asked to produce, and how would you know it did well?
Common gaps
- The task is described as the product feature, not as a learning problem.
- The output format is undefined, so the evaluation cannot be reproduced.
- A generative task is documented as if it were a classification task with a fixed label set.
Prepare before publication
Write a task definition for each model: problem, input, output and the measure of success. For a generative model, state that the output space is open and describe the evaluation approach.
Dates that matter
No separate date attaches to 6.2; the record dates are the stage date, the planning estimate and the CEN expectation stated above.
Cross-references
6.3 Model · 6.5 Tools · 8.1 General
Source
Source for 6.2: ISO/IEC 23053:2022, clause 6.2. The statement is read from the public base sample at 6.2 and is written in our words. The two official records are linked on this page so any line can be checked.
Not legal advice; does not certify. A reading of the framework and the evidence a reviewer commonly samples, not a substitute for the standard or your auditor's rules.