Human Control vs. Automation in Criminal Justice AI
You are building an AI risk assessment tool used in bail hearings. The system predicts the likelihood of reoffending. Judges can rely on the system completely or use it as a supporting tool. Too much automation may strip human judgment. Too little, and bias may creep back in.
Privacy vs. Utility in Healthcare AI
You are designing an AI system to detect disease outbreaks early using patient health records. The more data the system has, the better it can predict future health risks—but greater data access means less privacy for patients. How do you balance public health and personal privacy?
Transparency vs. Performance in AI Decision-Making
You are developing an AI to determine loan eligibility. A complex model (like a neural network) gives better predictions, but is hard to interpret. A simpler model is easier to explain—but slightly less accurate. Regulators and consumers are demanding transparency.
Key AI Ethics Terms
- Bias: Systematic and unfair discrimination by an AI model, often inherited from training data.
- Fairness: Ensuring that AI outcomes do not favor or harm groups unjustly. Can vary by definition (e.g., equal opportunity, demographic parity).
- Discrimination: When AI treats individuals or groups differently based on attributes like race, gender, or disability.
- Accountability: The idea that someone (person or organization) must take responsibility for AI outcomes.
- Transparency: The degree to which AI decision-making is understandable to humans.
- Explainability: The ability to interpret and explain how an AI system arrived at a decision.
- Consent: Users agreeing to their data being used, especially in contexts like health, surveillance, or personalization.
- Human-in-the-Loop: Keeping humans involved in critical AI decisions to prevent over-reliance on automation.
- Value-Sensitive Design: Designing AI with explicit consideration of human values like equity, autonomy, and dignity.
- Differential Privacy: A technique to share data insights while preserving individual privacy.