To Explore or To Exploit: Decoding Human Decision Making with R and Python

on YouTube ↗18m

" summary

Experiment uses a 10-item optimal stopping task run on Qualtrics with JavaScript/HTML/CSS, and compares human choices to a mathematically optimal agent and a cognitive instance based learning agent implemented in PyIBL (Python). Result: humans quit too early (many stopped on 72 while the max was 80), feedback—especially detailed—improves accuracy, and showing the underlying distribution reduces exploration and improves performance. Analysis, visuals and stats use R with Quarto, tidyverse and ggplot2; code and data are on GitHub.

AI-written summary · about

" chapters (4)

1. 0:00Introduction
2. 0:49Overview
3. 7:05Simulation
4. 10:47Results

" related talks

• 49mIncentivized ExplorationAlex Slivkins · Quantifying Uncertainty: Stochastic, Adversarial, and Beyond
• 15mAI-Driven Application OptimizationAlberto Falossi · DevConf.CZ 2023
• 01h 03mExperimentation in Data SciencePaul Meinshausen, Bargava Subramanian · The Fifth Elephant 2020
• 20mWhat can AI do for me?: evaluating machine learning interpretations in cooperative playShi Feng, Jordan Boyd-Graber · IUI 2019
• 44mRetrospective Search: Exploration and Ambition on Uncharted TerrainCan Urgun · Quantifying Uncertainty: Stochastic, Adversarial, and Beyond
• 35mMental Models to Use and Avoid as a Data ScientistJonathan Whitmore · PyData San Francisco 2016
• 32mRevisiting the Exploration-Exploitation Trade-Off in Bandit ModelsEmilie Kaufmann · Optimization and Decision-Making Under Uncertainty (Simons Institute, September 2016)
• 32mLearning from algorithms to make Human decisionsBrian Christian · Oredev 2017