The trouble is that the only experience we have of replicating molecules is of those used by life. It is extremely unlikely that DNA would form by chance. Even its simpler cousin, RNA, is hard to make in long enough strands to be biologically potent. And shorter nucleic acid molecules tend to make more errors when replicating. If the error rate gets too high, information leaks away faster than selection can inject it, and evolution grinds to a halt. Far from accumulating information, an error-prone molecule will shed it.