A new study has uncovered a startling and potentially consequential flaw in today’s leading artificial intelligence systems: they consistently favor