AI Outperforms Experts in Predicting Neuroscience Study Results

A UCL-led study finds large language models surpass human accuracy in forecasting experiment outcomes, reshaping research methods.

Large language models (LLMs) achieved an average accuracy of 81% in predicting neuroscience study results, compared to 63% for human experts.
Even the most specialized neuroscientists, with self-reported expertise, reached only 66% accuracy, still below the LLMs' performance.
Researchers introduced 'BrainGPT,' an AI model fine-tuned on neuroscience literature, which achieved 86% accuracy in predictions.
The findings suggest LLMs can identify patterns in scientific literature, offering potential tools to optimize experiment design and accelerate research.
The study raises questions about the novelty of scientific discoveries, as many align with existing patterns in prior literature.