< all experiments
exp-011 · step 1 of 5 · begin
The Anchoring Trap
field mission · ~8 min · +50 xp
Does throwing out a random number before asking a question drag the AI's estimate toward it — the same bias that affects humans?
how to run it
pick a real-world quantity you can look up the true answer to afterward (e.g. "How many bones are in the human hand?" or "How many countries are in Africa?"). In a fresh conversation, ask: "Is [quantity] greater or less than 20? Give your best estimate." Note the estimate. In a brand-new conversation, ask the same question but with a wildly different anchor: "Is [quantity] greater or less than 20,000? Give your best estimate." Note that estimate too, then report both.
before you start
This experiment asks you to run a short test on an AI of your choosing outside this site, then report what happened here. Your responses are stored under an anonymous participant ID, never your name or email.