> Is not empirical Bayes basically the same thing, but with priors stemming from the known data?
Yes. The problem is that either you end up using the data twice (once in the prior and once in the likelihood) or you have to choose how to split the data between the prior and likelihood, which can lead to other problems (particularly if you want to compare different models).
As a not-an-expert-in-stats, why would you say that? Is not empirical Bayes basically the same thing, but with priors stemming from the known data?