failure is not publishable, so the scientific literature can be misleadingly optimistic

Just because an approach is published, doesn't mean it really works.

Because, if ten different researchers try an experiment, and it works for just two of them, only those two successful studies will appear in print; the others will not. Anyone reading the literature will see only the successful studies, and conclude that the approach just works!

(While this problem is generally self-correcting, such corrections can take some time.)