Abstract
This survey asks for your predictions about an experiment on AI memory and belief updating. We recently recruited 1,000 U.S. participants for an experiment where they made predictions about Donald Trump's approval ratings.
The experiment first had participants discuss their views about Trump's economic and trade policies with an AI interviewer. Participants then made incentivized predictions about whether Trump's approval would increase or decrease on these topics. Before making predictions, they were randomly assigned to either make predictions without AI assistance ("No AI"), make predictions with the option to use an AI assistant with access to their interview ("Memory On"), or make predictions with the option to use an AI assistant without access to their interview ("Memory Off").
We study how this form of AI memory of users affects usage of AI, AI predictions, human belief updating, and subjective confidence. This survey will ask you to predict treatment effects on each of these outcomes. The survey should take about 10-15 minutes to complete, and we will offer a $5 payment for completing the study as a token of our thanks. We recommend you take the survey on a larger device (computer or tablet).
Study ID sspp-2026-0044-v1
Authors
Michael Thaler, Ze Wang
Discipline
Economics
Field
Political Economics, Experimental Economics, Behavioral Economics
Completion Time
15 Minutes
Close Date (UTC)
Oct. 31, 2026
Citation
Thaler, Michael, and Ze Wang. 2026. "AI Memory, Confirmation Bias, and Polarization." Social Science Prediction Platform. September 5.
https://socialscienceprediction.org/s/x6ybfz
Incentive Details
Incentive Type
Flat rate for participation
Incentive Size
$5.00
Recipient Pool
Everyone receives the incentives