IEEE Technology and Society Magazine - Summer 2013 - 28

gamblers end up losing against the
house, because when the oscillations between wins and losses are
large, most people quit when they
run out of money. Of course, that
is exactly the wrong time to quit,
but a gambler with limited funds
has no choice. A good example of
this is the "sure win" strategy that
is often espoused by roulette players. The idea is to simply double
your bet after every loss, and start
over after every win. If winning
pays double, then your expected
winnings are always positive, as
long as you don't quit immediately
after a loss, but double down as
your strategy requires. Yet, most
people lose anyway, because you
need a large sum of money to be
able to continue doubling your bet,
and when you run out of money,
you can't continue. So, the critical risk is the risk of a catastrophe, which is running out of funds,
not the expected risk by playing
indefinitely [31].
The gambling example has
many real world analogs. Nassim
Taleb argues in his book The Black
Swan that this is a fundamental problem in financial markets
where expected risk and return is
commonly used to evaluate investments. But the rare financial crises may be the determining factor
for most investors and businesses.
Such rare events are not usually
factored into risk/return calculations because they are difficult to
predict, and even impossible if they
never happened before. Yet, they
could be devastating to investors
and businesses, since a single crisis
can force them out of the market
after a big loss, and at precisely at
the wrong time to quit. Taleb gives
the example of not predicting the
possibility of a black swan, since
in Europe all swans were white,
until black swans were discovered
in Australia. A black swan type of
crisis, which has never been seen
before, is impossible to predict.
Yet, financial engineering technologies are constantly introducing
28

new financial instruments that create new risks of catastrophe, which
are impossible to predict, because
they have never happened before
since the technologies did not exist
previously [31].
More importantly, statistical
analysis focuses on precision and
correctness in a narrow domain,
under strict assumptions, at the
expense of intuitive concepts and
general principles that are roughly
correct but apply broadly. The
precision and reliability of analytical models become liabilities
if they lead to overconfidence
and overgeneralization. Specificity
and precision do not imply broad
applicability. On the contrary, analytical models are often very brittle,
and when their (not necessarily realistic) assumptions are not satisfied,
they can fail dramatically. Moreover, the abstractions of the analytical models increase the opaqueness
of their assumptions, and raise the
danger of overgeneralizing. Broad
qualitative insights on the other
hand are often roughly correct
and lack precision, but apply more
generally and with fewer and more
transparent assumptions. For example, during the 2008 financial crisis,
former Federal reserve chairman
Alan Greenspan famously acknowledged that the financial markets had
failed to obey some basic assumptions made in their analytical models, such as self-correcting, and
accurately judging their own longterm self-interest. Without those
basic assumptions, most analytical
models could not reliably predict
market behavior, even in the aggregate. With constantly changing
financial engineering technologies,
the ability to predict the behavior
of financial markets, and to judge
the impact of specific Federal
reserve actions is even more suspect [16], [31].
Similarly, extreme and rare
rewards are often ignored in statistical models, but in fact they
might be the primary drivers of
human behavior. This might be the

reason why people buy lottery tickets, although their chances of winning are very low, and the expected
value of the ticket is always less
than the cost. But money does
not have linear utility. The value
of a large win is so high that it
dominates the decision process.
This is the same reason why people
commit property crimes, although
they are likely to be caught and
punished, and the expected value
of the crime is less than the cost of
punishment. But the value generated from crime can be so high in
the rare case of not being caught,
or in the brief period before being
caught, that it dominates the decision process, especially because
the cost of punishment comes later
and gradually over a long period
of jail time. Especially for those
whose lives are not very rewarding and meaningful, the cost of
punishment may be minimal,
since a prison life may not be terribly different from a difficult life
in a crowded and dangerous inner
city slum. Crimes of passion, sex
crimes, and drug crimes all have
this quality; and punishment may
not be an effective deterrent, as
high rates of recidivism also confirm. Punishment for crime may
be merely an opportunity to take
revenge and to get psychological satisfaction. If the purpose of
punishment is deterrence, the system of punishment has to be reassessed, as technologies of crime
change the rewards and risks, but
also the technologies of crime prevention such as monitoring and
tracking become more effective.
Deterrence requires an emphasis
on prevention, not punishment
after the fact [2], [17].
Finally, statistical analysis in
social sciences often focuses on
a typical or average subject, but
very few people are actually typical or average, so the results may
not apply perfectly to most people. Average marriage lasts seven
years, but that may not be relevant
to most people who do not have

IEEE TECHNOLOGY AND SOCIETY MAGAZINE

SuMMEr 2013

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - Summer 2013