Small Skews, Repeated at Scale
A free lesson from Swipe AI · AI Ethics & Safety, level 2 · updated October 2026
By the Swipe team, founded by Colin · how we make lessons
Feedback loops rarely announce themselves as one big biased decision — they usually start as something tiny. A one-percentage-point difference in how often two groups are shown a particular option is easy to miss in a single snapshot of the data. Run that same small skew through thousands or millions of interactions, and it stops being noise.
Scale is what turns a minor skew into a measurable, self-reinforcing gap. Each cycle — output shapes outcome, outcome becomes new data, new data retrains the system — adds another small nudge in the same direction. None of the individual nudges look alarming on their own; the pattern only becomes obvious after looking at outcomes across a long stretch of time, not a single snapshot.
Cycle 1: System shows option A slightly more often to Group 1 Cycle 1: Group 1 picks A more often, simply because it was shown more Cycle 2: That pick becomes "evidence" Group 1 prefers A Cycle 2: The system leans further toward showing Group 1 option A Cycle 3: The gap between groups widens again ...repeat, and the small starting skew keeps growing
Why does a tiny initial skew in a feedback loop matter over time?
A feedback loop turns a small, easy-to-miss skew into a measurable gap by repeating the same nudge across many ____.
Individual nudges within a feedback loop usually look alarming even in a single snapshot of the data.
Free ai tools
Keep your streak going in Swipe AI
Swipe through every level of AI Ethics & Safety on your iPhone — with XP, stars, streaks, a personal advisor and the full track beyond these free levels.
Get the Swipe apps →