Guidelines for responsible AI cheat sheet
All 6 AIF-C01 flashcards for Guidelines for responsible AI on one page, grouped by exam objective. Print it or save it as a PDF for a last-minute read.
Explain the development of AI systems that are responsible
- Overfitting vs underfitting?
- Overfitting: accurate on training data but poor on new data, because the model memorised instead of learning. Underfitting: inaccurate on both.
- Bias vs variance?
- High bias means the model underfits: it's inaccurate on training and test data. High variance means it overfits: accurate on training data, poor on new data.
- Key legal and business risks of generative AI?
- Intellectual property claims, biased outputs and hallucinations. Each one can cost you customer trust.
- Why does a balanced dataset matter?
- Models fit the groups they see most. If a group is under-represented in training data, predictions for that group are less accurate, and biased.
Recognize the importance of transparent and explainable models
- Transparent vs black-box models?
- A transparent model, such as linear regression, shows how each input drives the answer. Complex (black-box) models need extra tools, such as feature importance scores, to explain them.
- What are AWS AI Service Cards?
- AWS's own documentation for its AI services: intended use cases, limitations, responsible AI design choices and best practices.