引導式測驗 · 中等 等級
歐盟人工智慧法案第 10 條中的資料治理要求測驗
測試本指南的《歐盟人工智慧法案》範圍、角色、責任、實際限制以及組織的實際部署注意事項。
問題 1 的 8
第 10 條要求供應商治理什麼?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- 第 10 條要求供應商治理什麼?
- Training, validation, and testing datasets for high-risk AI systems
- Only public datasets
- Only customer-facing prompts
- All personal data held by a deployer
- Which practice aligns with Article 10?
- Assuming a larger dataset is always fair
- Removing documentation after training
- Assessing relevance, representativeness, completeness, and errors in context
- Using one global quality score for every purpose
- 第 10 條是否將敏感資料處理普遍允許?
- Yes, whenever a model is high-risk
- No; it provides a narrow, safeguarded condition for bias detection and correction
- Yes, if a provider stores data securely
- No, it bans all bias testing
- Which duty can apply directly to a deployer’s input data?
- Create all of the provider’s training datasets
- Publish all input data
- Approve the model’s conformity assessment
- Ensure data under its control are relevant and sufficiently representative
- Why can a large dataset still be poor quality?
- Large datasets are automatically illegal
- Volume prevents validation
- It can omit populations or contain biased measures and labels
- A large dataset cannot have missing values
- What does representativeness depend on?
- The system’s intended purpose and context of use
- Only the number of rows
- Only the model type
- Whether the data are open source
- What should dataset records connect?
- Only a dataset name
- Only vendor contact details
- Only the final model score
- Sources, collection, exclusions, labels, transformations, and tests
- 僅有資料集文件是否能證明合法處理或公平性?
- Yes, if the dataset is complete
- No; those are separate questions requiring their own assessment
- Yes, when a model is accurate overall
- No, because documentation is prohibited