Suppose that some condition, E, is external in one of the senses delineated above and is a non-redundant part of some sufficient condition, C, on justification. It seems very likely, then, that it would be possible to construct a case in which, for some person S and some propositions p and q,
i) S satisfies E with respect to p but not with respect to q;
ii) S satisfies all the other parts of C, with respect to each of p and q;
iii) S does not satisfy any sufficient condition on justification other than C, with respect to either p or q;
iv) It seems to S that he is in the same epistemic position with respect to p as he is in with respect to q, and S has no reason for suspecting that either proposition is more justified or more likely to be true than the other.
So, for example, maybe S's belief in p is formed by a reliable process and S's belief in q is formed by an unreliable process. What's wrong with that?
If Reliabilism is a correct theory, then [assuming that her beliefs about dogs and unicorns are produced by processes that differ significantly in terms of their reliability] Susan would be justified in believing that the purple unicorn exists, but unjustified in believing that the dog exists. Internalists will find this counter-intuitive. To see why, consider things from Susan's point of view. It seems that, if one has adequate justification for believing that p and none for believing that q, it follows that one rationally should (or at least may) believe that p while refraining from believing that q. In addition, it seems that, at least normally, including in scenarios like the above, a rational person might recognize and report their own doxastic situation. Thus, Susan might say something like the following to one of the people in her virtual world; call remarks of this form the Absurd Speech:
I seem to be aware of a dog, just as I seem to be aware of a unicorn. These two experiences seem equally reliable to me, and in general, seem alike in all epistemically relevant respects. However, I believe that there is a unicorn, and I do not believe that there is a dog. I have no reason to think that the unicorn experience is any more likely to be accurate than the dog experience; I just accept the content of the one and not the other, for no apparent reason.
He then remarks, "On Reliabilism, the above would be a rational thing for Susan to say, or, more importantly, would (if true) be a report of an epistemically rational state of mind. But the Absurd Speech is not a rational thing to say. Nor is this a matter merely of the propriety of asserting what the Absurd Speech asserts; even to think to oneself what the Absurd Speech says would be a mark of irrationality."
There are two things that are strange about this--the reliabilist isn't defending a view of rationality, but justified belief. Moreover, the force the example seems to depend upon views about second-order justification and the example doesn't make clear whether the subject has adequate justification for falsely believing claims about the epistemic status of her first-order attitudes. Let that pass. Notice that we can do similar things with action. We can have someone intend to perform an action on Monday and an action on Tuesday where:
i') S satisfies some condition, E, with respect to the first action, A1, but not the second, A2;
ii') S satisfies all the other parts of C, with respect to each action;
iii') S does not satisfy any sufficient condition on justification other than C, with respect to either action;
iv') It seems to S that he is in the same practical position with respect to A1 as he is in with respect to A2, and S has no reason for suspecting that either action is more justified or more likely to be right than the other.
We can get similarly absurd speeches out of the agent. Should we conclude, then, that justified action is internalist in this way?
Suppose the agent sincerely and reasonably believes she's obliged to do A1 and A2. I can't then see how her obligation could be to do something else. But, it seems that the rationality of the agent's judgments about whether to perform A1 and A2 will be a function of how things appear to her and it will be a contingent fact whether genuine reasons appear to her to be reasons at all, whether the stronger of the available reasons will seem stronger, or whether non-reasons appear to the agent to be reasons. If any of this happens, the agent's actions won't be justified, but the agent will be able to engage in the same speech acts as above. So, what gives? Either the absurdity of the above speeches are not a good guide to justification or anyone who thinks that they are will have to deny that there are normative truths that we can miss if we deliberate rationally. That view is quite clearly incompatible with the sort of objectivist view about morality that Huemer defends in other places.
From bad to worse.
Suppose you judge that you ought to A. Suppose the judgment is reasonable. Are you permitted to do other than A? If you're not permitted to do other than A, then you're obliged to do A. Someone who thinks of justification in such a way that it's impossible for a case in which (i')-(iv') are satisfied could say that you're justified in doing other than A. But, if you're obliged to do what you rationally judge you're obliged to do, what do you say about cases like these:
(a) You judge that you ought to A rather than any of the alternatives to A-ing. You judge that you ought to B rather than any of the alternatives to B-ing. What you don't realize is that A-ing is an alternative to B-ing and vice-versa (i.e., can't do both).
(b) You judge that you ought to A but you can't A for reasons inaccessible to you. Either 'ought' doesn't imply 'can', or you can't reasonably believe you ought to A if you can't A. Both are bad.
Internalists need to show that you can't get from Huemer's internalist principles to the view that you ought to A if you rationally judge you ought to A. The problem is that there are cases where you ought to do what you rationally judge you ought. The difference between such cases would be an external condition.