SAN FRANCISCO — A study of how three popular artificial intelligence chatbots respond to queries about suicide found they generally avoid answering questions that pose the highest risk to the user, ...