By analyzing the three systems described above, we find that my earlier arguments that intelligent interfaces should be based on the intermediary interaction model were incorrect. A system can still volunteer informationwithout providing an intermediary interaction model. For example, one can imagine an SMA system that has a "suggestions" window rather than an MC window. The suggestions window volunteers information just like MC, but does not accept any natural language input from the user. Additional help can take the form of direct manipulation of the suggestion text (e.g. by highlighting key words and then invoking an additional help button).