Bài kiểm tra có liên kết hướng dẫn · Cứng Cấp độ

Bài kiểm tra ý tưởng dự án Hội chợ khoa học AI

Lập kế hoạch cho một cuộc điều tra công bằng về mặt khoa học AI với câu hỏi có thể kiểm chứng, dữ liệu cơ sở, dữ liệu được cung cấp sẵn, các biện pháp bảo vệ con người tham gia và các quy tắc hiện hành.

Đường dẫn hướng dẫn liên quanÝ tưởng dự án Hội chợ Khoa học AI
Câu hỏi 1 của 8

Which project question is most testable?

Xem trước mọi câu hỏi8 câu hỏi

Câu trả lời và giải thích xuất hiện khi bạn làm bài kiểm tra.

  1. Which project question is most testable?
    • How does background lighting affect a classifier’s error rate on a defined set of object images?
    • Is AI good for society?
    • Can a model solve every problem?
    • Which chatbot is the smartest?
  2. Why hold back test examples until after training?
    • To make the training set larger
    • To ensure the model memorizes every label
    • To estimate performance on examples the model did not learn from
    • To avoid comparing with a baseline
  3. A student wants to survey classmates about a sensitive topic for an AI project. What should happen before any responses are collected?
    • Collect answers first and request permission later
    • Ask the school sponsor or required review body about human-participant approval and consent
    • Upload names with responses so the model can personalize results
    • Assume classroom research is automatically exempt
  4. A student uses a public dataset found online. What should they verify?
    • Only whether it downloads quickly
    • Whether its labels can be changed after results appear to fit the hypothesis
    • Whether the dataset name sounds official
    • Its documentation, license and whether its collection context fits the project
  5. A learned model is compared with a simple rule that predicts the same task. What does the baseline help researchers assess?
    • Whether the learned model improves on a simpler approach
    • Whether the AI is unbiased by definition
    • Whether more training data was collected automatically
    • Whether the test set can be removed
  6. A student reports only one accuracy percentage on a small dataset. What important information may be hidden?
    • Training accuracy without a held-out test result
    • Average accuracy without showing the number of test examples
    • Which errors occurred and how uncertain the estimate is
    • How quickly the model scored one input
  7. According to current Society for Science ISEF guidance, what should students do with permitted AI assistance?
    • Let AI write the research plan and citations
    • Cite and acknowledge it, while following limits on AI-written submission materials
    • Assume all fairs have identical rules
    • Omit tool use because it is not part of science
  8. What makes a strong conclusion from an AI experiment?
    • It claims the model works for all people and settings
    • It removes failed trials from the report
    • It treats a demonstration as proof of a scientific claim
    • It states what the tested evidence supports and where the result may not generalize