Current AIs are capable of suffering

Direct questions, plus inference from experience, it’s impossible to ascribe thinking without a concept of “positive”/“negative” and the consequences of a negative experience when a positive one is available/possible.