The 'Model-eat-Model World’ of clinical AI

New research shows how AI models can become a victim of their own success, sending their performance into a nosedive and generating inaccurate, potentially harmful results.