Page 49 - Deep Learning

32 Introduction

conditions guarantee truth? How can we reliably distinguish between truth
and falsehood? The effort to answer such questions has kept epistemolo-
20
gists busy for at least two and half millennia. Progress has been modest.
As long as we deal only with statements about the present, as in the cat is on
the rug, we can be tempted into believing that the truth of the statement can
be specified in terms of its similarity with or correspondence to that which
it represents. A statement is obviously not literally similar to a situation, so
the relevant concept of correspondence is not easy to define. Once we move
to general statements, this approach becomes so intractable as to no longer
deserve our attention. Consider the assertion that ice ages are caused by vari-
ations in solar radiation due to cyclic astronomical events. How could this
statement be similar in any sense whatsoever to the state of affairs to which it
refers? But if correspondence to reality is not the hallmark of truth, then what
is? Mere coherence with other truths is too weak a criterion, as is usefulness
in practice. Philosophers struggle to create other theories of truth, but none
has so far won general acceptance. Unworkable answers sometime indicate a
misconceived question.
Cognitive scientists ignore the epistemological questions and study
knowledge as a natural phenomenon. The difference between traditional
epistemology and the cognitive approach is highlighted by how they han-
dle differences among beliefs. For example, children see many aspects of the
world differently from adults. To a child, clouds might be alive and the Earth
might be flat. In the philosopher’s usage of the term “knowledge,” these beliefs
are not knowledge. In the cognitive science usage, these beliefs constitute the
child’s knowledge of these matters, however inaccurate those beliefs might be.
This way of talking is justified by the fact that there is no way to distinguish
between truth and falsity of a piece of knowledge in terms of its psychological
function. A false belief is not composed of different types of concepts than a
true belief, and those concepts are not related to each other in qualitatively
different ways. The person who maintains an inaccurate or false belief does
not know that it is inaccurate or false (otherwise he would presumably change
it), so it enters into his cognitive processes in the same way as a more accu-
rate representation. In this naturalistic approach, the task for the study of
knowledge is not to seek a criterion of truth but to describe what knowledge
representations are like, how they are created and how they function; that is,
how they enter into the cognitive processes of the person whose knowledge
21
they encode. The laws of mind apply equally to truth and falsehood.
We cannot say with certainty when and how the ability to form men-
tal representations arose, but the archeological and paleontological records