Artificial Intelligence and Personality: Large Language Models’ Ability to Predict Personality Type

Abstract

Micro-targeting, which is when messages are specifically tailored to an individual based on the information that can be derived from their digital footprint, has become a prevalent practice in digital spaces. This communicative approach often centers around personality, the rationale being that individuals resonate with messages differently depending on their personality type. There has been research into whether machine learning methods can improve the detection of personality types, but insufficient analysis has focused on whether large language models enhance this technique even further. And so, this study examines the ability of GPT-3.5 and GPT-4, two of the models underpinning OpenAI's ChatGPT, to classify personality type. Using a publicly available dataset collected through the ‘PersonalityCafe’ forum, this paper found that GPT-3.5 and GPT-4 were able to perfectly classify 73% and 76% of the sample's personality type, respectively, simply by analyzing an individual's 50 most recent tweets—a level which is significantly better than random guessing and other machine learning models. This has important implications for communication scholarship because if LLMs enable actors to detect personality types more accurately, micro-targeting could become even more powerful and widespread. There is also a possibility for misuse by bad actors and privacy threats, so understanding these mechanisms is crucial for building safe artificial intelligence systems and safeguards against this.