Discussion Prompt
Initial post: find and analyze one AI fairness/bias case, ideally one that your peers have not yet posted about.
In your post, please:
- Describe the specific issue (what system has a problem? for whom? etc.), including a link to a reputable source.
- Explain why it’s biased/unfair using clear criteria (e.g., disparate performance across groups, or disparate standards used between groups), including quantitative evidence if you can find it. Note that there are different definitions of what constitutes bias or fairness. A classic example is “affirmative action”: some people see it as a way to correct for past discrimination, while others see it as a form of discrimination itself. So you’ll need to be clear about what you mean by “fair” or “unfair.”
- Discuss why this matters (e.g., what are the potential real-world impacts on affected groups?)
- Consider key stakeholders in this situation:
- Who’s affected directly and indirectly?
- Who has power to make changes?
- Who benefits from the current system?
- Acknowledge real-world constraints:
- What technical limitations affect potential solutions?
- What business or economic factors are relevant?
- What tradeoffs might be necessary?
Then, respond to some peers’ posts. In your responses, you might:
- Evaluate whether your peer’s criteria for “unfair” makes sense (maybe take a different perspective)
- Suggest an additional impact or affected group your peer didn’t consider
- Propose a potential solution or mitigation
Posts should:
- Be understandable without referring to these instructions
- Cite sources where possible.
- Be written clearly, for an educated but non-technical audience.
Sources for Cases
Ironically, you can ask an AI for examples of AI bias! (but dig in to make sure it’s not making stuff up—which is another problem with AI that we’ll study later in the course).
A few sources you might consider:
- CS Education resources
- Academic and Nonprofit Organizations
- Industry reports
- News articles
- Scholarly articles (e.g., from conferences like ACM FAccT)
Some Things to Think About
(If you find any of these interesting, feel free to bring it up in your post.)
- What sorts of decisions might AI systems make more fairly than humans – or vice versa? Give specific examples of situations, explain why your choice could be more fair, and be specific about what you mean by “fair” in each situation.
- Do you think that social media algorithms are biased? Why or why not? Cite evidence where possible.
- Suppose you’re hired to develop an AI system that might help identify people at risk of mental illness. What issues of fairness or bias might you be concerned about, and what might you do about them?
- The article cited mathematical proofs about the impossibility of fair decision-making by anyone, whether machine or human. Do you believe those results, or are they missing something?
- Suppose a friend was denied a car loan by an algorithm, and thinks he was being unfairly discriminated against. What would you tell your friend to help them understand their situation? What evidence might you want to collect to help your friend make a strong discrimination case against the loan company?