AI and attention
AI, Recommendation Systems and Human Attention
Recommendation systems are optimized for engagement, not for a person's stated goals, so they reliably erode sustained attention. The practical countermeasure is not a better feed but a trained capacity to hold attention on one object — the skill contemplative traditions have taught for millennia and the reason Shumake's work on meditation sits next to his work on AI.
Optimization targets what it can measure
A ranking system maximizes an observable proxy — watch time, clicks, returns. Nothing in that objective represents whether you finished the thing you sat down to do. The system is not malfunctioning when it fragments your day; it is succeeding at its actual target.
The countermeasure is a practice, not a setting
Fixed periods of single-object attention, deliberate friction between you and the feed, and a daily quota of work that produces something rather than consumes something. These are unglamorous and they are the only interventions that survive the next interface update.
Frequently asked
- Do algorithms really shorten attention spans?
- Feeds optimized for engagement reward frequent switching, and frequent switching trains the habit. The measurable effect is on sustained-attention habits rather than on some fixed biological span.
- What actually helps?
- Scheduled single-task periods, friction between you and the feed, and a daily production quota. Trained attention beats a tuned feed setting.
