That is not a useful rebuttal if in fact it is impossible to guarantee that your AGI will not be a socialpath no matter how you program it.

Eliezer's position generally is that we should make sure everything is set in advance. Jacob_cannell seems to be basically saying that much of an AGI's behavior is going to be determined by its education, environment, and history, much as is the case with human beings now. If this is the case it is unlikely there is any way to guarantee a good outcome, but there are ways to make that outcome more likely.