Plotkin, Henry. 1994. Darwin Machines and the Nature of Knowledge. Cambridge, MA: Harvard University Press.
Plotkin approaches the problems of epistemology by identifying the mechanisms of knowledge acquisition; those mechanisms, he claims, situate a substantial amount of the study of knowledge within scientific rather than philosophical inquiry. From the point of view of "Universal Darwinism" (as proposed by Dawkins, et al.), ideas, intelligence, and genetics are all controlled by the same evolutionary mechanisms, each at their own level of a hierarchy.
The primary heuristic
Campbell (1960) produced a formulation of evolutionary theory which has come to be known as the "g-t-r heuristic." A heuristic is something "which leads to discovery and invention." The first stage of this process, generation (g), is a process or mechanism for introducing variations on the elements undergoing evolution. The second stage, testing (t), is a consistent selection process that winnows out some of the elements. The third phase, regeneration (r), is actually the same as the first, but stating it explicitly indicates that the cycle continues indefinitely.
Plotkin identifies the g-t-r heuristic operating at the genetic level as the primary heuristic for knowledge acquisition. Genetic evolution results in the physical embodiment of knowledge about the world: proteins encode "knowledge" about chemistry, eyes "know" about optics, flippers know about moving through water. Hard-wired instinctual behavior is a more obvious form of this type of knowledge.
Some animals have developed non-genetic mechanisms for encoding knowledge about the world, such as association and memory (i.e., intelligence). Most of these mechanisms do not have complex generative stages (such as reasoning and thought), and this is a weakness from the g-t-r heuristic perspective. Without a generative stage, whatever elements are encoded are either maintained statically, or winnowed out without descendents. A useful encoding might happen randomly, but there is no means of progressing from not-so-good to good. But as James Baldwin wrote in 1909:
there would be at every stage of growth a combination of congenital characters with acquired modification: natural selection would fall in each case upon this joint or correlated result; and the organisms showing the most effective combinations would survive the result is that variation would tell most when in the direction in which the accommodations were being made and found useful [emphasis in original].
If a particular association or memory, or anything involving learning, is consistently beneficial to a species, and some genetic variation makes it easier for the species to acquire it, or to prefer it to other knowledge, there will be evolutionary pressure to retain that variation. We can observe that even as babies, animals tend to learn things relevant to their species survival most easily. One of the most extreme cases of this selected learning is the example of honeybee dances.
We can observe even in humans that an infant mind is not a tabula rasa: a healthy baby attends to human faces, and a lack of interest is usually an indication of some mental problem. These effects guide all elements of human behavior and intelligence, even rationality: humans have a tendency to learn particular ways of estimating likelihood and not others, which makes us good at avoiding unlikely but very bad things, and terrible at resisting bad gambling bets.
The secondary heuristic
Even though intelligence operates within the constraints of the primary heuristic, the development of thought and rationality introduced a mechanism for variation into intelligence that in turn began an independent process of evolution. This conception of evolving intelligence has been around for some time, as in the writing of William James:
new conceptions, emotions, and active tendencies which evolve are originally produced in the shape of random images, fancies, accidental outbursts or spontaneous variation in the functional activity of the excessively unstable human brain, which the outer environment simply confirms or refutes, adopts or rejects, preserves or destroys selects, in short, just as it selects morphological and social variations [1880, emphasis in original].
Because this new type of evolution, the secondary heuristic, operates in a radically different timescale, it has little opportunity to affect the function of the primary heuristic, although language may be the result of such an interaction.
The tertiary heuristic
Plotkins argues that even beyond the secondary heuristic, there may be yet another level of evolution. In 1976, Dawkins proposed that the elements of genetic evolution are not species or individuals, but genes. An organism is not the goal of the gene, it is instead a conveyance for the gene until the next generation of genes is replicated. A network of genes manages to replicate successfully by assembling a good protective organism.
Dawkins also suggested that there is similarly a unit of cultural evolution, which he called a meme. A belief is a network of memes, varied and reproduced by human minds, that together manage to assemble a coherent entity and propogate themselves. Those beliefs that either aid the survival of their humans, or manage to propogate faster they kill them, survive (to put it fancifully). Science may be an example of an effective long-term meme network.
Baldwin, J. M. 1909. Darwin and the Humanities. Baltimore: Review Publishing.
Dawkins, R. 1976. The Selfish Gene. Oxford: Oxford University Press.
James, W. 1880. "Great men, great thoughts, and the environment," The Atlantic Monthly, 46(276):44159.
Further commentary is available regarding an evolution approach to epistemology.