AI and False Memories (Bachelor’s Thesis)

Overview

An experimental study investigating how interactions with AI chatbots can influence the formation and persistence of false memories in humans. The research was part of my bachelors thesis and compared different methods of questioning, including traditional surveys, prescripted chatbots, and generative AI agents powered by large language models. The study also investigates the effect of factors like education, familiarity with chatbots and experience with similar incidents on the formation and persistence of false memories.

Role & Contributions

• Co-first author on research paper submitted to Science Advances
• Led system implementation and integration:
  • Developed custom chatbot interfaces using JavaScript and GPT-4
  • Created web interfaces for survey delivery and data collection
  • Implemented data pipeline for storing conversation data in Google Sheets
• Contributed to experimental design and methodology
• Conducted statistical analysis of results

Key Findings

• Generative chatbots led to 3x more false memories than control conditions and 1.7x more than surveys
• 36.4% of participants developed false memories through AI interactions
• False memories induced by AI showed remarkable persistence after one week
• Individual factors like interest in crime investigation and AI familiarity correlated with increased susceptibility

Impact & Implications

This work reveals important considerations about AI’s influence on human cognition and memory, with significant implications for the deployment of conversational AI in sensitive contexts like law enforcement, healthcare, and education. The findings highlight the need for careful design and ethical guidelines when developing AI systems that interact with human memory. Research paper submitted to Science Advances (August 2024)

To know more about this work:

• Code: GitHub
• Project: Official Project Page