Page 346 - Deep Learning

10

Belief Revision: The Resubsumption Theory

… the act of judgment that leads scientists to reject a previously accepted theory
is always based upon more than a comparison of that theory with the world. The
decision to reject one paradigm is always simultaneously the decision to accept
another …
Thomas S. Kuhn 1

Beliefs can be expressed in discourse (I believe that …; In my opinion …), but
they are consequential primarily because they guide action. Only someone
who believes that the Earth is round would set sail for China by sailing west
from Europe, and only someone who believes that a nuclear chain reaction
is physically possible would try to build a nuclear power plant. In everyday
life, mundane actions are guided by myriads of beliefs of smaller scope: If you
believe that the movie starts at four o’clock, you will try to be at the theater
shortly before that time. Beliefs can inhibit action as well as guide it. In the
late 19th century, some argued that rockets could not work outside the Earth’s
atmosphere because the rocket exhaust would have nothing to push against.
2
If this belief had persisted, people would presumably not have built rockets for
space travel and there would have been no moon walks.
The outcomes of actions reflect back on the belief or belief system that
guided those actions. There can be no better way to become convinced that
the Earth is round than by circumnavigating it. The success of the action, the
triumphant homecoming, increases confidence in the relevant belief. The rela-
tion between action and belief is particularly visible in technology. The main
reason for believing in aeronautics is that airplanes fly, and the main reason for
believing in nuclear chain reactions is our ability to produce nuclear energy.
When beliefs change, actions change accordingly. Investors who become con-
vinced that the stock market is about to fall will move their money to less risky
investments. The third contribution of this book is a micro-theory that puts

329