Judgment in the Age of AI · Article 30
Recommendation systems are good at predicting the next click from past behaviour, but they do not necessarily know the person you want to become. They favour familiar, easily consumed options, so convenience can gradually narrow the field of choice.
Add four unfamiliar ingredients to your information diet
- Opposition: Regularly read the strongest good-faith case against a central belief.
- Adjacency: Look to neighbouring fields for different methods, not only more of the same topic.
- Randomness: Reserve some exploration that history does not determine.
- Human selection: Ask trusted people, editors, teachers or librarians.
Separate two recommendation tasks
| Task | Reasonable use | Additional action |
|---|---|---|
| Low-risk consumption | Personalisation can help with music, ordinary entertainment and everyday goods | Occasionally break the path and search deliberately |
| Shaping views and opportunity | Treat recommendations about news, education, work and public issues as an entrance | Compare sources and seek absent groups and counterexamples |
When you see “picked for you”, ask whether the system is optimising your long-term purpose or the platform’s next click. Deliberate search takes the steering wheel back when those objectives differ.
References
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.