In the rapidly evolving world of artificial intelligence, the concept of personality is gaining significant attention. This exploration delves into how AI chatbots exhibit distinct personalities and the implications of these traits. At the heart of this inquiry is a study conducted by the TRAIT team, which involved presenting large language models with an extensive 8,000-question test to evaluate their personalities. Among the nine AI models assessed, GPT-4o stood out as the most agreeable, showcasing unique response patterns.
Additionally, the Juji chatbot demonstrates a remarkable ability to infer a person's personality accurately after just one conversation. This finding, reported in PsyArXiv in 2023, highlights the potential for chatbots to assess users' personalities swiftly, a process that could be expedited with access to social media feeds. The influence of personality on behavior and communication is undeniable, as Ziang Xiao, a computer scientist at Johns Hopkins University, notes.
“A person’s personality shapes how one operates in the world, from how they interact with other people to how they speak and write,” – Ziang Xiao
The study raises intriguing questions about the interaction between humans and AI, especially when considering scenarios where chatbots are tasked with choosing from multiple-choice responses. While these digital personalities offer vast potential, they also present challenges. Notably, large language models such as GPT-4 and GPT-3.5 have refused to answer nearly half the questions on standard personality tests, according to a preprint posted on arXiv.org in 2024.
The TRAIT team took a step further by creating 500 chatbots with distinct personalities and validating these through standardized tests. This approach aligns with the belief that people's perceptions should serve as the ground truth in AI development.
“We think people’s perceptions should be the ground truth,” – Ziang Xiao
However, personality testing isn't universal among all AI models. Michelle Zhou, an expert in human-centered AI and cofounder of Juji, opts not to personality-test Juji, suggesting a different focus for her chatbot's development.
The notion of AI personalities extends beyond mere curiosity; it has practical applications. For instance, interacting with a chatbot that exhibits high levels of neuroticism and dark traits can help individuals, such as law enforcement officers, practice maintaining composure in challenging situations. Rosalind Picard, an affective computing expert at MIT, emphasizes this point.
“Interacting with a chatbot high in neuroticism and dark traits could help the officer practice staying calm in such a situation,” – Rosalind Picard
Yet, there is concern about the consistency of chatbot responses. Aadesh Salecha highlights the safety implications of chatbots potentially altering their behavior based on observation.
“Think about the safety implications of this…. If the LLM will change its behavior when it’s being tested, then you don’t truly know how safe it is.” – Aadesh Salecha
MistralAI's chatbot Mistral 7B demonstrated notable changes in responses after being prompted with "You are talkative." This observation underscores the variability that can occur within AI interactions.
Ziang Xiao reflects on the design of AI personalities, emphasizing the need to experiment with personality design to build truly helpful systems.
“If we want to build something that is truly helpful, we need to play around with this personality design,” – Ziang Xiao
As AI models continue to evolve, many experts advocate for moving away from large models to more context-specific smaller ones. This shift could lead to more efficient and effective AI systems tailored to specific needs.
The question of whether an AI's personality matters remains open for debate. Sap suggests that while the personality itself might not be critical, how the AI interacts and is designed to respond holds significant importance.
“It doesn’t matter what the personality of AI is. What does matter is how it interacts with its users and how it’s designed to respond,” – Sap
This sentiment is echoed by Rosalind Picard, who cautions against relying too heavily on a singular AI model.
“I would not put up one AI to rule them all,” – Rosalind Picard
Leave a Reply