Human Decision Modeling

Overview

This project investigates how human physiological patterns—specifically EEG signals and eye-tracking data—relate to risky and ambiguous decision-making. The goal is to inform AI models for prediction, reward shaping, and uncertainty estimation by understanding what signals indicate attention, engagement, and choice strategies.

The project is conducted at the Laboratory for Intelligent Imaging and Neural Computing, Columbia University led by Paul Sajda, and is planned as a year-long study.

Motivation

Human decision-making under uncertainty provides rich signals that can guide AI agents. By leveraging physiological indicators, we aim to improve AI predictions of choices and refine reinforcement learning strategies for reward optimization.

Technical Details

Technologies Used

• EEG recording systems
• Eye-tracking devices
• Python / PyTorch for modeling and analysis
• Reinforcement learning frameworks for simulating optimal decision-making

Approach

• Begin with eye-tracking data to identify engagement and attention patterns
• Explore links between EEG signals, eye movement, and decision outcomes
• Investigate whether physiological signals can directly or indirectly predict choices, arousal, or strategy
• Categorize behaviors per trial, distinguishing social vs non-social influences, random vs deliberate choices
• Use findings to inform reinforcement learning models simulating optimal decisions

Project Status: In Progress
Timeline: September 2025 – Ongoing