How would it “know” not to do this. It can only provide you with a synthesis drawn from its training data. If your problem is within its training set, it’s not really research (which is what I assume “science” means). If its outside its set, you’re getting statistically generated noise.