When dealing with synergistic or human-in-the-loop systems, AI may offer liberal suggestions to human users as alternatives for decisions or action, but these can be combined with conservative warnings as to which suggestions are more or less reliable. That is, if the system has some form of confidence measure, then the liberal suggestions would include anything with even quite a low level of confidence, but any below a very high threshold would be marked as potentially unreliable (amber warning).
Used on page 482