That's where your supervisor comes in. In most cases, they should be an expert in the field, and guide you towards a useful and novel problem.

Moreover, I am not suggesting you don't look at other papers at all. But google scholar and some quick skimming of abstracts and papers you find should suffice to check if someone has already done the work. If you start fully reading more than a handful of papers, your ideas are already locked in by what others have done, and it becomes way harder to produce something novel.