Both of these are considered forms of confounding (when something "extra" from the person/environment affects the results of an experiment).

For a situation that is classified as Experimenter Bias - that would be when the Experimenter (when watching the displayed behaviours OR the symptoms of a patient), sees things because of knowledge they possess about who is recieving "which kind" of treatment. So for example, because she knows that Patient #4 is receiving medication and Patient #5 is receiving a placebo - she interprets signs of "getting better" more readily in patient 4 than in 5. In this case - her knowledge about what is actually going on (or not) changes what she sees.

With demand characteristics - during interactions between an experimenter and participant, each one can give off subtle cues about what is the "better" kind of response. So for example, the experimenter leans in, asks followup questions, and seems more interested when an unusual kind of response is elicited. The client learns "OH, this is the kind of thing he wants" and starts responding more in that direction (giving more elaborative and unusual responses). The experimenter then reacts to the increasing psychopathology and re-engages in those behaviors (leaning in, etc.). So they are both accidentally reinforcing the other's behavior, increasing the chances of those behaviors rather than what would normally happen.