Artificial intelligence has been widely integrated into the field of scientific research, but the application of machine learning models can lead to misleading or erroneous results. Researchers at the University of California, Berkeley, have proposed a statistical technique called "Prediction-Driven Inference" (PPI) for validating scientific hypotheses. The PPI technology can correct the outputs of large general models without understanding the nature of model errors, adapting to specific scientific problems and thus avoiding machine learning biases. This technique is not only applicable to protein structure prediction but also to various research areas such as Amazon rainforest deforestation estimation, becoming an essential component of modern data-intensive, model-intensive, and collaborative science.