# AI prediction leads people to forgo guaranteed rewards

**Authors:** Aoi Naito, Hirokazu Shirado  
**arXiv:** [2603.28944](https://arxiv.org/abs/2603.28944)  
**Date:** 2026-03-30  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2603.28944v1/

## Key points

- Study by Aoi Naito and Hirokazu Shirado on impact of AI on decision-making
- Research involved 1,305 participants using Newcomb's paradox
- Individuals believing in AI's predictive authority tended to forgo guaranteed rewards
- Over 40% of participants viewed AI as a predictive authority
- Participants passing up on rewards increased by a factor of 3.39 compared to random framing
- Resulted in reduced earnings ranging from 10.7% to 42.9%
- Influence of AI on decision-making persisted even when predictions failed
- Emphasizes complex interplay between human cognition and technological advancements
- Highlights need for deeper understanding of how AI shapes decision-making processes

## Summary

A study by Aoi Naito and Hirokazu Shirado titled "AI prediction leads people to forgo guaranteed rewards" delved into the impact of artificial intelligence (AI) on decision-making processes. The research involved 1,305 participants and utilized a behavioral implementation of the classic Newcomb's paradox to investigate how AI influences decision-making. The findings revealed that individuals who believed in the predictive authority of AI tended to constrain their decision-making, often opting to forgo a guaranteed reward. Over 40% of participants viewed AI as a predictive authority, resulting in a significant increase in passing up on guaranteed rewards by a factor of 3.39 compared to random framing. This pattern led to reduced earnings ranging from 10.7% to 42.9%. Interestingly, the influence of AI on decision-making persisted even when its predictions failed to materialize. The study sheds light on the complex interplay between human cognition and technological advancements and emphasizes the need for a deeper understanding of how AI shapes decision-making processes. It also highlights potential implications for future interactions with intelligent systems.