Scientists create “Centaur” AI that mimics human psychological quirks, irrationality

An international team of scientists has created Centaur, a ChatGPT-like AI system that can participate in psychological experiments and behave as if it has a human mind. Published in Nature, the research demonstrates how large language models trained on 10 million psychology experiment questions can help cognitive scientists better understand human cognition by mimicking both our rational decisions and cognitive quirks.

What you should know: Centaur represents a new approach to studying the human mind by creating AI that replicates human psychological patterns rather than trying to surpass them.

• The system was trained specifically on psychology experiment data to mirror human responses in cognitive tests.
• Unlike current AI systems that excel at single tasks, Centaur aims to replicate the full spectrum of human mental processes, including our biases and limitations.
• The research addresses a key limitation in current AI development, where systems can perform complex tasks like chess mastery but fail at basic human-like reasoning.

Why this matters: Current AI systems remain “very distinguishable from the human kind” despite impressive capabilities, creating an opportunity to use AI as a tool for psychological research rather than just commercial applications.

• While companies like OpenAI and Meta race toward artificial general intelligence without a clear definition, scientists are taking a different approach by focusing on understanding rather than replacing human cognition.
• The research could accelerate psychological studies by providing a consistent “digital human” for experiments, potentially leading to faster breakthroughs in understanding how our minds work.

The big picture: Modern AI excels at specific tasks but lacks the integrated, flexible intelligence that characterizes human thinking.

• A chess-playing AI system remains “helpless behind the wheel” of a car, illustrating the narrow focus of current AI capabilities.
• AI chatbots can make surprisingly simple errors, like allowing illegal chess moves, despite their sophisticated language abilities.
• Cognitive scientists have developed sophisticated theories about learning, memory, and decision-making that require extensive human testing to validate.

Key details: The system leverages decades of cognitive science research to create a more human-like AI behavior pattern.

• Cognitive scientists have identified consistent human behavioral patterns, such as preferring certainty over risk even when it means “forgoing a chance to make big gains.”
• When offered a guaranteed $1,000 versus a potentially larger but uncertain payout, people typically choose the certain option.
• These predictable human quirks and biases are what Centaur has been designed to replicate in experimental settings.