New AI Model 'Centaur' That Predicts Human Behavior With 64% Accuracy Is Freaking Experts: 'We Just Crossed a Line'

Artificial intelligence (AI) has again stunned people with yet another innovation by creating a new AI language model capable of predicting human behavior. This system has accurately predicted humans' next move in several complex scenarios and has surpassed advanced cognitive models with its outcome. It is named "Centaur" by experts and was trained in just a matter of five days on a dataset called Psych-101, which contained details about human behavior in scenarios like learning games, memory tests, moral dilemmas, as well as risk-taking situations. Findings regarding Centaur have been published in the journal Nature.

While the model is being lauded, some have advised caution considering the implications. "We just crossed a line in AI that should make every behavioral scientist sweat," tweeted SmartRoad AI CEO Christopher Brya. "This system masters the game of being human across gambling, memory tasks, and problem-solving."

 "It outperformed 14 established cognitive models in 31 out of 32 tasks, even predicting behavior in scenarios it had never seen before," said Brya. We're looking at the potential end of expensive, slow human behavioral studies. Think about this: an AI that knows how you'll decide before you do. The researchers call it a tool for cognitive science. I call it the beginning of the end of human behavioral unpredictability. Your next therapist might be running you against Centaur first," 

Humans' Choice Prediction

Centaur reveals what pathway or option a human will choose in a particular situation. It showcases how a human will choose to learn a new skill, face unknown situations, or navigate complex decisions. The AI model can predict how a human will behave in any psychological experiment. For this capability, the AI has studied 10 million decisions from 60,000 human subjects across 160 experiments.

Through the dataset, the AI has figured out the pattern in which humans think, study, and then make choices. Surprisingly, all humans, even with their differences, somewhat follow the same pattern when it comes to thinking. "Not only do we routinely make mundane decisions, such as choosing a breakfast cereal or selecting an outfit, but we also tackle complex challenges, such as figuring out how to cure cancer or explore outer space," the researchers wrote.

Training of Centaur

The most stunning aspect of Centaur is the accuracy in its results. The study claims that the language model showcased 64% accuracy in predicting human behavior, according to Live Science. It was way more than any of the models in the past have been able to do. Researchers believe that they used the world's largest human behavior dataset for this language model in the form of Psych 101. Its training was no different from other AI models. Centaur was fed all the details of the experiment, what the subjects were told, the situation, and all the parameters. 

Then the model was asked what decision the human took. If the Centaur answered incorrectly, the model was fine-tuned to provide the correct response the next time. The process was repeated until Centaur started to give correct answers regularly. Now, the model can even answer about situations it has not trained for, and also provide reaction times of humans in several scenarios. "We've created a tool that allows us to predict human behavior in any situation described in natural language — like a virtual laboratory," Marcel Binz, a research scientist at the Helmholtz Institute for Human-Centered AI in Germany, said.

Possibility of this Model

Researchers believe that such a model could completely transform education, product design, marketing, and mental health treatment. They were surprised to find Centaur's internal workings had slowly become more aligned with a human brain's activity. The model has somehow started to mimic how the actual brain processes information. More analysis of this capability could shed light on the complicated functioning of the human brain. Currently, researchers know what decisions humans take, but they now want to understand how humans arrive at those calls. Currently, the team focuses on expanding Centaur to account for other diverse domains and populations in the world. The model is now publicly available.