L6.09
Feedback Loops and Preference Entrenchment
Cards in this group · 5
- L6.09.1The system only observes items it showed; unshown items never receive positive feedback
- L6.09.2Early accidental clicks get amplified into durable profile features
- L6.09.3Users adjust behaviour to get the recommendations they want, so behavioural data lose representativeness
- L6.09.4Breaking the loop needs active exploration; the cost is immediate and the benefit is delayed
- L6.09.5Offline evaluation on historical logs systematically favours the policy that produced those logs