Study raises concerns about AI health-prediction models trained on unreliable datasets

Some AI models designed to foretell changeable and diabetes hazard whitethorn beryllium based connected datasets whose origins cannot beryllium verified, according to caller research. The study, published successful BMC Medicine and led by researchers astatine QUT and the Australian Center for Health Services Innovation (AusHSI), examined 2 wide downloaded wellness datasets hosted connected Kaggle, an online level for sharing datasets and machine-learning resources, marketed arsenic "the world's AI proving ground."