SouthernWorldwide.com – Artificial intelligence (AI) holds the promise of significantly enhancing human capabilities and driving progress across various fields. However, a new study suggests a potential downside: AI could inadvertently make people less interesting or “dull.”
This concern stems from the nature of large language models (LLMs) that power many AI applications. These models tend to produce information that is predictable and aligned with the general population’s tendencies. Consequently, they can simplify the complexities of life into a collection of common, diluted ideas.
“LLMs predict the most likely next word in a sentence or event in a sequence, and by definition, that’s average,” explained Sandra Matz, a professor at Columbia Business School and the lead author of the study. This means AI recommendations, whether for movies or home decor, often reflect what is most probable, leading to a homogenization of choices and outputs.
The research involved analyzing over 110,000 real-world decisions made by 1,000 individuals. These decisions were then compared against the outputs of both generic and personalized AI agents. Additionally, the study utilized data from the myPersonality project, a Facebook application that offered personality tests to users who consented to share their profiles for research.
“AI hates risk” is a key observation from the study. When individuals rely on AI for decisions, such as selecting a vacation spot or purchasing walking shoes, they are often steered towards the most common options. This can lead them away from their more unique or even eccentric preferences and behaviors.
Matz, a computational social scientist with expertise in psychology and computer science, noted that AI tends to narrow the scope of what users explore, both in terms of topics and psychological inclinations. The LLM agents, she stated, “play it safe within a user’s preferences.”
Even if an AI system is aware of a user’s occasional unconventional choices, like an unusual dinner selection, the LLM agents are designed to guide behavior back towards more normative options. This process limits the range of experiences and explorations available to individuals.
“AI hates risk because we train it that way,” Matz emphasized. The underlying motivation for this programming is to keep users engaged on the platform, which is achieved by presenting them with content they already favor rather than venturing into less familiar territory.
While AI applications are not inherently designed to be this way, their current programming dictates this approach. To counteract the potential for AI to diminish the richness and diversity of human experience, Matz suggests that tech developers incorporate an “exploration mode.” This feature would offer users the option to receive more unexpected and less conventional recommendations.
Implementing such a feature could help individuals avoid becoming “boring” and prevent culture from converging into a singular set of preferences, Matz concluded.
