Page 313 - Deep Learning

296 Conversion

To achieve this, he might have to abandon or revise prior beliefs, not merely
add new ones. For example, children eventually abandon their belief in santa
Claus, and at least some physics students abandon their intuitive, impetus-
based beliefs about mechanical motion and replace them with the newtonian
concept of inertia. At the collective level, the human race has abandoned the
belief that history cycles through a fixed sequence of eras, among many oth-
9
ers. in the fields of politics and religion, people sometimes undergo radical
changes of mind, much to the horror of those among their friends who remain
believers. Although data on prevalence are lacking, the majority of belief for-
mation events in everyday life must be of the monotonic, routine sort, but non­
monotonic changes – conversions – fascinate because they are consequential,
dramatic and difficult to understand.
Logic appears to provide a straightforward method for non-monotonic
belief revision: A proposition that implies a false consequence must itself be
false. The relevant inference is called by its Latin name Modus Tollens.* if P and
Q are two propositions, then Modus Tollens can be rendered as follows:
if true (P), then true (Q).
But, in fact, false (Q).
Therefore, false (P).

The argument, if the economy is healthy, stock prices rise; but prices have fallen
steadily; so the economy is not healthy, is an instance of this inference rule. The
logical validity of Modus Tollens is not in doubt. When beliefs are logically
related, Modus Tollens enables a single contradictory fact to produce large-
scale change in a belief system: if true (P) is implied by some belief true
(o), which itself is implied by yet other beliefs, then the falsification of Q will
propagate backward along the implication links and perhaps undermine mul-
tiple beliefs.
Both common sense and psychological research suggest that this
description of belief revision fares poorly as a psychological theory. instead,
10
resistance to contradictory information is ubiquitous. People are slow to
revise their beliefs and when two people enter into a disagreement, neither
of them is likely to convince the other, no matter how many facts or argu-
ments each contributes to the discussion. This is obviously so in religious and
political disputes, but resistance to change is present in all areas of experi-
ence, including professional and technical domains. scientists are not nec-
essarily quick to change their favorite theory in response to an inconsistent

*
The full name is Modus Tollenda Tollens, but it is commonly abbreviated.