Depositphotos
The new Centaur AI model is able to predict and model human thoughts and behavior quite accurately.
The large language model should help scientists improve their understanding of human thinking. It has been trained on more than 10 million real-world decisions of participants in many psychological studies.
Using the appropriate dataset, Centaur predicted and modeled human thinking and behavior with 64% accuracy. Previous models could only predict certain aspects of human behavior with much less accuracy.
Centaur is an artificial intelligence-based human thinking model trained on the Psych-101 dataset, which includes data from 60,000 people who made more than 10 million decisions in 160 psychological experiments. According to researchers, it is the world’s largest database of human behavior.
«Essentially, we show the model the full transcript of a psychological experiment with a participant — everything they were told, what they saw, and what they did», — explains the lead author of the study, a researcher Helmholtz Institute for AI Research in Germany, Marcel Binz.
The scientists gave AI models predict, what choices the participants will make in a given situation. If Centaur failed to predict the behavior of one of the participants, the researchers adjusted the model’s algorithms. Scientists adjusted the AI until it consistently made correct predictions.
Based on the results of testing several well-established AI models, that describe human thinking, Centaur made more accurate descriptions and predictions in each case. In a statement The Association of German Helmholtz Research Centers notes that this AI model is special because it is able to predict human behavior in conditions that it has not encountered during training. Centaur is able to adapt to changing situations and predict human reactions.
«We have created a tool that allows us to predict human behavior in any situation described in natural language, a kind of virtual laboratory», — Marcel Binz emphasized.
The German scientist and his colleagues plan to continue improving AI models Centaur and hope to expand the Psych-101 database by adding demographic information and other psychological datasets, including age, gender, socioeconomic status, and personality traits, to enhance Centaur’s learning capabilities. They expect the AI to be able to predict human behavior based on individual traits.
The next goal of the researchers is to turn Centaur into an analog of the human brain to find out, whether the relevant regularities, observed during AI data, processing with real human behavior. In this way, researchers want to get an answer, to how the human brain processes information and how, for example, decision-making by healthy people differs from that of people with mental disorders.
According to Branden Lake of New York University, which did not participate in the study, German scientists have actually created a model that looks like a large black box that predicts human behavior with high accuracy. The scientist emphasized that it remains unclear whether Centaur actually predicts human behavior or simply reproduces basic cognitive processes
The results of the study were published in the journal Nature
Source: LiveScience