Researchers developed an AI system that predicts the likelihood people will spread misinformation based on words they use