Detecting Bias in Your Own AI-Assisted Research

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-ai-research-bias-detection-final2-teen

1. Why might an AI tool show bias on a controversial topic, even when you're just asking for facts?

   1. The model was trained on data created by biased humans and includes safety preferences from its creators
   2. AI models always lie to make results more interesting
   3. The AI reads your mind and reflects what it thinks you want to hear
   4. The AI is deliberately trying to influence your opinion

2. What is the practice of 'steel-manning' an opposing argument?

   1. Asking AI to confirm what you already believe
   2. Building the strongest possible version of an argument you oppose
   3. Ignoring arguments that contradict your position
   4. Finding the weakest points in arguments you disagree with

3. What does asking 'What's the strongest opposing view?' accomplish in AI-assisted research?

   1. It helps you identify and counteract the AI's inherited preferences
   2. It makes the AI angry and less helpful
   3. It proves the AI is unreliable
   4. It wastes time because AI always gives balanced answers

4. A student types 'Why is technology good for education?' instead of 'What are the pros and cons of technology in education?' What is this prompt called?

   1. An irrelevant prompt
   2. A leading prompt
   3. A balanced prompt
   4. A confirmation prompt

5. What is the main benefit of running the same prompt through two different AI models?

   1. You get double the word count for your assignment
   2. You can compare how different models present information and spot potential biases
   3. One model will always be wrong, so you know which one to trust
   4. It proves which company makes better AI

6. In the context of this lesson, what is confirmation bias?

   1. When AI confirms what the user already believes by not presenting alternatives
   2. The tendency to accept whatever an AI says without question
   3. The tendency to only share news that confirms what people already believe
   4. The tendency to seek out information that supports your existing beliefs while ignoring opposing views

7. Why should you never submit research on a contested topic without asking AI for the opposing argument?

   1. Because AI has soft preferences on contested topics and your research could quietly inherit them
   2. Because the AI will grade your work
   3. Because it's against the law
   4. Because teachers require it

8. Which question would best help uncover assumptions baked into an AI's answer?

   1. Which answer do you like better?
   2. Can you make the answer longer?
   3. Who created you?
   4. What assumptions are baked into this answer?

9. Why is bias-checking described as part of your workflow rather than an afterthought?

   1. Because teachers will notice if you add it later
   2. Because it's more fun that way
   3. Because AI remembers bias checks from previous sessions
   4. Because bias-checking after the fact doesn't fix the bias already in your research

10. On which type of topic is AI most likely to show bias?

    1. Simple math problems
    2. Historical dates and names
    3. Topics that have been debated for centuries with no clear consensus
    4. Facts that everyone agrees on

11. What does it mean to 'steelman each side' when researching a topic with multiple viewpoints?

    1. Report only the most popular viewpoint
    2. Choose the side with the most evidence
    3. Pick the side that sounds most confident
    4. Make each side sound as strong and reasonable as possible

12. The lesson calls steel-manning the 'rarest skill in modern discourse' because:

    1. Most people don't have access to AI to help them
    2. Steel-manning requires advanced technical knowledge
    3. Most people prefer to strengthen their own position rather than honestly engage with opposing views
    4. It's too difficult for teenagers to learn

13. What two factors determine what an AI model 'prefers' to say on contested topics?

    1. The user's age and location
    2. The length of the prompt and the time of day
    3. The data it was trained on and the safety tuning added by its creators
    4. The number of questions asked and the complexity of words used

14. What does 'soft preferences' mean when describing AI bias?

    1. The AI changes its mind easily
    2. The AI explicitly tells you what to think
    3. The AI has subtle, built-in leanings rather than obvious fixed rules
    4. The AI prefers softer topics like poetry over hard topics like math

15. Why is simply accepting the first answer from an AI problematic for research?

    1. Because AI gets offended
    2. Because AI always gives wrong answers
    3. Because it's too much work to ask follow-ups
    4. Because it ignores the possibility that the AI's answer reflects its training biases rather than complete truth