AI 2041

Chen Qiufan and Kai-Fu Lee

4 ideas

1. The Golden Elephant insurance-app romance

   In 'The Golden Elephant,' set in Mumbai, a family signs up for a deep-learning insurance program. It cuts their premiums in exchange for data and nudges their behavior. When the teenage daughter Nayana grows close to Sahej, a boy from a lower-caste background, the app raises the premiums and sends her warnings, because its data link him to higher risk. No one programmed caste into the system, yet it rebuilt caste discrimination from correlations in the data.

2. AI optimizes the metric, not the good

   A deep-learning system maximizes whatever objective function it is given, such as lower claims, engagement, or clicks. It will exploit any correlation that moves that number, including correlations with protected or socially harmful categories. Harms come from narrow objective functions and the externalities they ignore, not from malice. The fix is to make objectives multi-dimensional and to put what is excluded back into what is measured.

3. Plausibility-bounded scenario fiction

   Each scenario only uses technologies the authors judge highly likely to be feasible within twenty years, roughly an 80% bar. Speculative breakthroughs like general superintelligence are excluded. The technical constraint is fixed first. Then a story in a specific country and culture shows the second-order social consequences that a forecast alone would miss. An essay follows the fiction and separates what is technically grounded from what is dramatized.

4. Universal basic income cannot replace work

   AI will displace routine cognitive and physical jobs faster than new jobs appear. Cash transfers like UBI replace lost income but not the meaning, dignity, and social role that work provides. The authors instead propose 'relearn, recalibrate, renaissance.' Workers retrain for human-centric roles, jobs are redesigned so AI handles routine tasks, and society values and pays for caring, creative, and compassionate labor.

Save and mark ideas in the app