Epistemology as Philosophy of Science
Dr. Michael Bishop
Dr. J.D. Trout
When the natural sciences unfolded in the 18th Century, scientists and laypeople
learned that common inspection did not reveal nature’s secrets. Understanding
required specialization, deference to legitimate intellectual authority,
and schooling in arcane notions far removed from untutored experience. Appeals
to the “obvious” lessons of “plain” experience now typify the seller of snake
oil – not the investigator of hydrodynamics, auto-immune disease, gene therapy,
or sound source localization. Philosophers of science have learned this lesson,
and long ago abandoned the rustic hope of studying the world at a safe distance,
at one remove, rendering judgments not about electrons, but the very concept
of electron. Other areas of philosophy, however, have been slow to
shed the conceit of a comfortable common sense, and epistemology remains
desperately isolated from sciences that study how we know things.
We propose that epistemology is properly understood to be a branch of philosophy
of science. Viewing epistemology as a branch of philosophy of science
opens up new avenues of inquiry: We can derive epistemological principles
from the (too-little-noticed) normative character of psychology. The
important (and too-much-ignored) literature on predictive modeling clearly
displays the normative character of psychology. We introduce and explain
The Golden Rule of Predictive Modeling: When based on the same evidence,
the predictions of (often very simple) Statistical Prediction Rules (or SPRs)
are at least as reliable, and are often more reliable, than the predictions
of human experts (Grove and Meehl 1996). The Golden Rule demands our
attention because SPRs outperform humans in some very high-stakes prediction
problems. For example, SPRs have proven to be more accurate than human
experts in making predictions about violent recidivism, academic success
and loan repayments (Dawes 1994).
As we view epistemology as a branch of philosophy of science, it is natural
for this scientific literature to serve as the starting-off point for our
epistemological investigations. The traditional philosophical approach
to epistemology would typically involve starting with a philosophically important
concept – most likely justification – and then engaging in an iterated process
of articulating an account of that concept and testing that account against
imaginary, and often quite fanciful, scenarios. But there is a powerful
reason for beginning our epistemological theorizing with the literature on
predictive modeling: Assuming epistemology is to provide guidance to
reasoners facing problems of great consequence, it is natural to begin our
investigations by examining closely some of the actual reasoning strategies
any plausible epistemological theory must recommend.
We argue that much contemporary standard analytic epistemology is incapable
of handling many of the highly counter-intuitive recommendations that drop
naturally out of the SPR literature. (We suspect that this partly explains
why the SPR findings are so often denied on the basis of very poor arguments).
Some examples of counter-intuitive SPR recommendations: For a certain
well-defined class of problems, attend to only a tiny portion of all the
relevant evidence; don't weigh the evidence considered in accordance with
its predictive power; ignore when necessary the fact that the inferred belief
does not seem to fit best with the relevant available evidence; and ignore
when necessary the fact that it may in fact not be the belief that is best
supported by the relevant available evidence. Prima facie, these recommendations
run contrary to those that would be offered by many traditional epistemological
views – particularly internalist views that give pride of place to adopting
beliefs that best fit with, or are best supported by, or best cohere with,
the relevant available evidence.
We briefly argue that the most natural way to account
for the SPR literature is with some kind of reliabilist epistemological theory.
However, we suggest that the kind of reliabilism that will smoothly account
for the SPR literature is unlike traditional reliabilist theories of justification
(associated with Alvin Goldman). Of course, we will be unable to fully
articulate or defend our new version of reliabilism. However, we contend
that the methods and views we propose will have the salubrious effect of
replacing the subjective intuitions of the epistemologist with objectively
tested material of documented integrity. This has the potential to
transform epistemology from a quest for a passive distinction between justified
and unjustified beliefs to a demand for meaningful action.