Relative Utility - Behavioral Mathematics for Game AI - page 210

Furthermore, they may look to place the blame on the algorithm itself without
analyzing and testing the validity of the components. The point is that both the con-
struction of the algorithm and the values and weights that are used in the algorithm
are of equal importance. Later on, we will address both of these components in
more detail.
A PPARENT C ONTRADICTIONS
Another interesting issue to consider is the possibility that there might be times when
it is actually preferable to select a less attractive option. I use the preceding sentence
with the disclaimer that both the terms preferable and less attractive are entirely sub-
jective. As in the deadly outbreak example, when we analyze what is truly happening
at a mathematical level, a statement that initially looks like a contradiction may
actually become clearer.
This is different from the outbreak example above, where the options were the
same. The perception of the outcomes was what was flawed. In this case, we face a
choice where, on the surface, one option is mathematically superior to another—
in the short term. Once we place the choice into the entire framework, however, it
is exposed as actually being less preferable. This issue can be even more apparent
when placed in a continuum. For instance, as the situation changes, it would seem
that one choice would become more advantageous. Yet surprising results can occur.
G IFFEN G OODS
Interestingly, it seems that we humans often confuse ourselves by making situations
overly complex. Animals often seem to “get it.� One example that showed how an-
imals can grasp the sometimes elusive complexity involved in hedonic calculus was
an experiment conducted by Raymond Battalio and John Kagel. They showed that
laboratory rats were perfectly capable of analyzing the relative merit of controlled
supplies of pleasure and discomfort and making the correct decisions with that
information.
The experiment involved placing rats in a box equipped with the equivalent of
two vending machines. Each machine would dispense a measured amount of liquid
for the rats to drink for each press of the lever. They also changed how many lever
presses were available to the rats in each session in the box. With a moment of re-
flection, you can see how the variables in play correspond to familiar terms in the
world of economics. By capping the number of lever presses available to the rats,
they were setting the income level, that is, the most the rats had available to spend
in any one period. By adjusting the amounts of liquid that the respective lever