Guidelines for Responsible AI · Topic 6 of 6
Explainability Tools and Human Centered Design
Use SageMaker Model Cards, AWS AI Service Cards, Clarify and partial dependence plots to document and explain models, and apply human centered design.
- 4 min
- SageMaker Model Cards record the intended use, data, evaluation and limits of a model.
- AWS AI Service Cards describe the intended use and limits of AWS AI services.
- SageMaker Clarify explains predictions through feature influence and detects bias.
What it is
Several tools and habits help you document a model and explain its behaviour. They support the goal of transparent and explainable AI that the previous page described.
Human centered design adds a way of thinking: explanations should be built for the people who use and are affected by the AI.
How it works
Amazon SageMaker Model Cards
Amazon SageMaker Model Cards create a single record of a model. A card can hold the intended use, the training data description, the evaluation results, risk ratings and any limits or notes. Model cards support governance and audits because they keep the facts about a model in one place.
AWS AI Service Cards
AWS AI Service Cards are published documents about certain AWS AI services. They describe the intended use cases, limits, design choices and best practices for responsible use. They help you decide whether a managed service suits your purpose, without needing to inspect the model.
Amazon SageMaker Clarify
Amazon SageMaker Clarify explains individual predictions and overall model behaviour by showing which input features influence results most. It also detects bias, as the earlier page described. Feature importance is a common form of explanation for tabular models.
Partial dependence plots
A partial dependence plot (PDP) shows how the predicted result changes as one feature changes, while the others are held steady. For example, a plot can show how the predicted price rises as house size grows. PDPs help people see the direction and size of a feature's effect.
Open source models, data and licences
Transparency also depends on what you know about the model and its data.
| Item | Why it matters |
|---|---|
| Open source models | Design and sometimes weights can be inspected, depending on the licence |
| Data documentation | You can see where training data came from and what it covers |
| Licensing | You know what you are allowed to do with the model and the data |
| Provenance records | You can trace sources for audit or dispute |
Open does not always mean fully open. Licences differ, so read the terms.
Tools at a glance
| Tool | Main use |
|---|---|
| SageMaker Model Cards | Document a model's purpose, data, results and limits |
| AWS AI Service Cards | Understand intended use and limits of AWS AI services |
| SageMaker Clarify | Explain predictions and detect bias |
| Partial dependence plots | Show a feature's effect on predictions |
| Open source models and data records | Allow inspection and traceability |
Human centered design for explainable AI
Human centered design starts from the needs of people. The key ideas are:
- Know the audience: A data scientist, a manager and a customer need different explanations.
- Design for clarity: Use plain language and simple visuals, not raw numbers alone.
- Allow control and feedback: Let users question, correct or override results.
- Show uncertainty: Say how confident the system is.
- Support human oversight: Keep people in the loop for important decisions.
- Test with real users: Check that the explanation helps them make decisions.
- Include diverse people: Involve different groups in design and testing.
Putting it together
A responsible team documents the model in a card, explains predictions with suitable tools, tells users what the AI can and cannot do, and gives them a path to challenge results. This builds trust and also supports compliance.
What to remember for the exam
- SageMaker Model Cards record the intended use, data, evaluation and limits of a model.
- AWS AI Service Cards describe the intended use and limits of AWS AI services.
- SageMaker Clarify explains predictions through feature influence and detects bias.
- A partial dependence plot shows how a feature affects the prediction.
- Open source models, data records and licences affect how much you can inspect and reuse.
- Human centered design tailors explanations to the audience.
- Keep humans in the loop and allow feedback and challenge.
- Show confidence and uncertainty to users.
For a question about a central record of a model's purpose, data and limits, choose SageMaker Model Cards. For a published description of an AWS service's limits, choose AWS AI Service Cards.
Common mistakes
Mixing up Model Cards and AI Service Cards. Model Cards document your own models. AI Service Cards describe AWS services.
Confusing Clarify with Model Cards. Clarify explains and tests. Model Cards document.
Giving the same explanation to every audience. A customer needs a simple reason, not technical plots.
Assuming open source means no licence limits. Check the terms.
Treating documentation as a one time task. Update it when the model or data changes.
Check yourself
What does a SageMaker Model Card hold?
A record of a model's intended use, training data, evaluation results, risks and limits.
What does a partial dependence plot show?
How the predicted outcome changes as one feature changes while the others stay fixed.
Name one principle of human centered design for explainable AI.
Examples are tailoring explanations to the audience, showing uncertainty, and keeping humans in the loop.
Topic 6 of 6 in Guidelines for Responsible AI
Test yourself on Guidelines for Responsible AI
The full AWS AIF-C01 question bank has 849 questions, with timed practice exams and an explanation for every answer.
This study guide is independent and is not affiliated with, endorsed by, or sponsored by AWS. Vendor names, certification names and exam codes are trademarks of their respective owners and are used for identification only.