Update: I will clarify what I have written above as I realize it is unclear. You cannot solve the problem of discrimination (in non trivial cases of true and f

Update: I will clarify what I have written above as I realize it is unclear. You cannot solve the problem of discrimination (in non trivial cases of true and false, of good and bad) through a deterministic solution, as the epistemological debate did not solve the general problem. You cannot train a function approximator (e.g. an ANN) as a Discriminator through supervised learning, because the problem remains such for human judgement. Creating a Discriminator through unsupervised learning, I'd like to see how one would frame a proposal; and anyway, if we could create reliable filters, the main question - as usual for a progressive approach to AI - would be to have the oracle in the system teach us instead that knowledge that what could not achieve with good old thinking.