Ernest Sosa likes externalism. He thinks that it is intuitively correct. But he must and does agree that it must be clarified in order to avoid certain problems. So, his mission in this paper is to first define what he calls 'Generic Reliabilism,' then to show how it is susceptible to certain objections, then to present a modified version of it, and to show that this new version is, in general, better than its predecessor. Let us look at his argument.
First, we get the usual definition of generic reliabilism: S is justified in his belief that p at t if the belief is produced by some faculty that usually produces true beliefs. Then, we get a couple of Alvin Goldman's notions of justification with Sosa's revisions. A belief is strongly justified iff it is well formed, and by means of a truth conducive process. A belief is weakly justified iff it is 'blameless' (not the result of an intentional mistake?) but ill-formed, and the believer is not aware that the belief is ill-formed.
A belief is superweakly justified iff the process that produces the belief is unreliable but the subject did not intentionally come to hold the belief because it was acquired unreliably. And, finally, a belief has strong meta-justification iff the subject neither believes that nor can determine if the belief is ill-formed (hence the 'meta-' prefix), and the subject is aware of the process by which he got the belief and that the process is reliable.
OK, seems reasonable enough. But, Sosa points out, there are a couple of scenarios (actually, three, but Sosa concentrates mainly on the two listed below) in which these conceptions of justification just do not work. The 'new evil demon' problem takes a couple of forms in the article, but what...