If it uses a rational problem-division approach for responses creation, instead of social conditioning/prompt censorship, it will certainly come to conclusions we don't like.

Pure rationalism has always been a double edged sword, thus only wielded against certain problems, sparing our sacred cows.