“数字读心术”精准模拟和预测人类行为

Core Insights - The "Centaur" AI model developed by the Helmholtz Munich Research Center can accurately predict human behavior in psychological experiments, surpassing traditional cognitive models [1][2] - The model is based on a dataset called Psych-101, which includes over 10 million decisions made by more than 60,000 participants across 160 psychological experiments [1] - "Centaur" demonstrates exceptional generalization capabilities, providing accurate predictions even in novel scenarios [2] Group 1: Model Development - The model utilizes Meta's latest Llama 3.1 language model and is fine-tuned on a vast dataset of human decision-making [1] - Training the model took only five days on high-end computing processors [1] Group 2: Performance and Comparison - "Centaur" outperformed over ten other behavioral prediction models in 31 out of 32 tasks tested on new subjects [2] - The model's internal processing mechanisms align closely with human brain activity, indicating a similar information processing approach [2] Group 3: Applications in Psychology - The potential applications of AI in psychology are significant, including the development of reliable AI-based psychological counseling and emotional support systems [3] - AI can help uncover new psychological strategies and enhance scientific discovery in the field [2][3]