The Comfort Trap: Getting Too Used to AI
summary
AI creates a comfort trap that stops critical checks and produces concrete harms: a 2024 Financial Times example shows a screen reader describing a sporting event as a restroom misdescription, historical bias reappears (Amazon recruiting) and an automated script by a UK department wrongly flags 200,000 wrongly flagged claims. Practical defenses are human-in-the-loop review, bias audits, and inclusion checkpoints with people with disabilities; recommended tools include XCore, Robot Framework, Fairlearn, SHAP, LIME and Google What‑If for explainability and testing.
AI-written summary · about
chapters (8)
- 0:00Introduction to AI comfort trap
- 1:30Everyday ethics in AI decisions
- 2:14Seven wonders of inaccessibility
- 4:03Testing in the age of AI
- 5:51What AI cannot yet do
- 6:33Ethical AI testing framework
- 7:58Tools for explainable AI
- 9:22Conclusion: AI as a helper
related talks
05mToward Ethical AI
46mCan you trust your AI?
44mToxic Trust: The Psychology of Automation Bias in Generative AI
57mToxic Trust: The Psychology of Automation Bias in Generative AI
23mHuman-Driven Intelligence: Where AI & Automation Fall Short
34mReally a New Mouse Trap? Exploring Risks with Artificial Intelligence
48mAI and Disability: Progress or Exclusion?
26mResponsible AI-What, Why, How and Future!