গাইড লিঙ্ক ক্যুইজ · মাঝারি স্তর
Automation Bias: Overtrusting AI Quiz
Recognize automation bias, commission and omission errors, effective human review, and ways to test whether reviewers verify AI advice.
প্রশ্ন 1 এর 8
Which behavior best describes automation bias?
প্রতিটি প্রশ্নের পূর্বরূপ দেখুন8 প্রশ্ন
আপনি কুইজ নেওয়ার সাথে সাথে উত্তর এবং ব্যাখ্যাগুলি উপস্থিত হয়।
- Which behavior best describes automation bias?
- A tendency to replace vigilant judgment with automated advice
- A software crash caused by a robot
- A method for improving model accuracy
- A type of training-data encryption
- A user follows a wrong AI recommendation. What kind of automation error is this?
- An omission error
- A commission error
- A calibration metric
- A data deletion
- A reviewer misses an important finding because the AI did not flag it. Which error pattern is this?
- Commission
- এলোমেলো নমুনা
- Omission
- Model calibration
- Why is a nominal “human in the loop” label insufficient?
- The reviewer may lack time, evidence, expertise, or authority to challenge the system
- Human beings never make errors
- The model may be open source
- It means the system has no automation
- Which test can reveal commission errors?
- Give a correct recommendation and ask for its color
- Present a plausible but incorrect suggestion and observe whether the reviewer catches it
- Remove all source information
- Ask users whether they like the interface
- Which test can reveal omission errors?
- Ensure the system fails to flag a relevant case and see whether the reviewer notices through other evidence
- Only test a system that always flags everything
- Measure page load speed
- Ask the model to explain its training data
- What factors can affect reliance on automated advice?
- Only model parameter count
- Workload, time pressure, interface, expertise, and independent evidence
- Only color contrast
- ফাইল ফরম্যাট
- What does a meaningful automation-bias audit measure?
- Only whether a review checkbox exists
- Whether users like AI generally
- Overrides, missed alerts, evidence inspection, and downstream outcomes
- The number of AI features