Darek Barefoot responds
Lewis's philosophical instincts were good. He didn't develop his thought adequately in this case, but I think he sensed a genuine incoherence problem. I will admit that I have struggled with it mightily.
The key to the argument is that it pertains to naturalistic assumptions that occur epistemologically prior to speculations about the relationship between true beliefs and advantageous ones. Any natural account of reason begins with the assumption of processes that have no inherent sensitivity to truth; specifically, truth cannot be seen as the goal of the great sequence of chemical reactions that constitute evolution. If we stipulate beforehand that we owe our beliefs exclusively to processes that have no teleological dimension, then whether we realize it or not we are left to bootstrap ourselves up to coginitive reliability when it comes to our thoughts about the relationship between natural processes and truth.
Natural selection, specifically, can be sensitive to truth only indirectly. Where beliefs are concerned nature's imperative is their advantageousness in the struggle for survival. The natural perspective must put advantageousness before truth in evaluating human cognition, and therein lies the snag.
To illustrate, suppose that our only way to detect fire was by the smoke that usually accompanies it. Our only fire detector in that case would be a smoke detector. But while a smoke detector might keep us away from fire, it would be nonsensical to try to use a smoke detector alone to discover or verify that smoke usually accompanies fire. In fact, if our sole means of detecting fire were smoke, it is hard to imagine how we could know about fire. To know that smoke comes from fire, we must have some means of detecting fire other than simply by the smoke it produces.
If natural selection is the sole architect of human cognition, the brain is a detector of advantageous behavioral options. Our beliefs are screened only for their contribution to survival. The overlap between advantageous beliefs and true beliefs would be like the overlap between smoke and fire. The only thing we can know about our beliefs, given this account, is that they are advantageous. To know that advantageousness and truth usually go together when it comes to beliefs, we would have to have some means of distinguishing between advantageousness and truth. And that, in turn, would require the ability to detect the two values separately, just as with fire and smoke.
In fact, without the ability to distinguish truth we could not even know that our beliefs were advantageous, since that itself requires truth-detecting ability. The account based on natural selection alone reduces to a purely behavioral theory excluding knowledge of any kind. To make room for our knowledge--of natural selection or anything else--we have to bring in a non-naturalistic source of human reason.
Victor, you proposed the counter-example of supernatural beings who planted thoughts in our minds. But for such an example to be comparable to natural selection, you would have to propose that such beings not only plant thoughts in our minds but that they do so without any conscious regard for the truth or falsehood of the thoughts they are planting. Given those conditions it would indeed be incoherent to speculate that some blind mechanism happens to be aligning the planted thoughts with genuine reason. To know such an alignment could occur we would have to know genuine reason first-hand.
The argument against natural selection as the architect of reason is not a skeptical threat for the same reason that the liar's paradox is not a skeptical threat. If a speaker says to a listener, "I never tell the truth," it is not grounds for either party to conclude that the speaker never tells the truth. It is grounds for both parties to conclude that the speaker has not expresssed himself coherently. The speaker in the present case is naturalism, and to be taken seriously naturalism at a minimum must pass the test of coherence.