# Conformal Prediction Sets Improve Human Decision Making

**Authors:** Jesse C. Cresswell, Yi Sui, Bhargava Kumar, Noël Vouitsis  
**arXiv:** [2401.13744](https://arxiv.org/abs/2401.13744)  
**Date:** 2024-01-24  
**Summary page:** https://www.summarizepaper.com/en/paper/2401.13744v1/

## Key points

- Study titled "Conformal Prediction Sets Improve Human Decision Making" by Jesse C. Cresswell, Yi Sui, Bhargava Kumar, and Noël Vouitsis
- Investigates the impact of conformal prediction sets on human decision-making processes
- Machine learning models mimic human uncertainty signaling through calibrated prediction sets generated using conformal prediction
- Pre-registered randomized controlled trial involving human subjects to assess utility of prediction sets in aiding decision making
- Providing individuals with conformal prediction sets leads to improved accuracy compared to fixed-size prediction sets with equivalent coverage guarantees
- Value of quantifying model uncertainty using conformal prediction in facilitating decision-making processes involving humans and AI systems
- Emphasizes potential benefits of incorporating conformal prediction sets into human-in-the-loop decision-making scenarios and collaborative efforts between humans and AI teams

## Summary

In their study titled "Conformal Prediction Sets Improve Human Decision Making," authors Jesse C. Cresswell, Yi Sui, Bhargava Kumar, and Noël Vouitsis explore the impact of conformal prediction sets on human decision-making processes. The researchers investigate how machine learning models can mimic human uncertainty signaling through calibrated prediction sets generated using conformal prediction. They conduct a pre-registered randomized controlled trial involving human subjects to assess the utility of these prediction sets in aiding decision making. The results demonstrate that providing individuals with conformal prediction sets leads to improved accuracy compared to fixed-size prediction sets with equivalent coverage guarantees. This finding highlights the value of quantifying model uncertainty using conformal prediction in facilitating decision-making processes involving both humans and artificial intelligence systems. Overall, the research emphasizes the potential benefits of incorporating conformal prediction sets into human-in-the-loop decision-making scenarios and collaborative efforts between humans and AI teams by leveraging predictive tools that mirror human uncertainty signaling.