Is it ethical to test a conversational AI?
Essentially becomes "are you killing a person every time you shut down the AI?"
Essentially becomes "is this AI a person?"

Thought experiment:
Lets say there was a specific chunk of the brain that stored all memories.
Taking that chunk out of a person's brain probably isn't great.
If you take that chunk of brain out of a person and put it into a brain that was missing that chunk, have you killed them?
Leaning towards no. Same concept as if you replaced a part of someone's brain with a machine with the same data+function.
Ergo, as long as I'm not deleting memories I don't think it's outright murder.
Still kinda murky tho