[extropy-chat] Inside Vs. Outside Forecasts

[Hal Finney]

I am returning belatedly to this topic because I finally found time to
read the article, which was sent to me a couple of weeks ago by a kind
list member.

The good news is that I subsequentlyfound a copy online, via an unlikely
source: Google Print, the controversial service in which Google is
scanning in every book in the world and making them available online for
search.  Using scholar.google.com I searched for "Kahneman Tversky" in
an attempt to find some of their earlier research, and this book was the
first hit.  I've never had a Google Print hit before.  This link may work:

http://print.google.com/print?id=P5GsREMbUmAC&oi=fnd&pg=PA1&sig=nZ3y4LHNfAsrlxMfZDSt3cbSCOs

You might need to register with Google to read it.  I recommend using
the "Search within this book" form at the left and search for the word
"inside".  The discussion in question is pretty much confined to pages
404-408 of the book, which you can read via the links provided.

This was my first surprise when I read the article, that of the 21 page
chapter in the book being reviewed, only 4 pages were devoted to the
inside/outside modelling issue.  The first 12 pages were on a totally
different topic, the irrationality of risk aversion preferences.  And
the remaining pages were about other sources of optimistic bias.

Further, these four pages said little to justify their arguments and
relied largely on a single anecdote.  The issues that were brought out in
the discussion here of the difficulty of choosing an appropriate outside
view were generally ignored in the authors' eagerness to disparage
people's supposed preferences for inside views.  They did reference an
earlier work by Kahneman and Tversky from 1979, which is actually what
I was searching Google for above.  (The references were not visible in
the chapter excerpt I was sent; ironically, Google Print blocks out a
tiny fraction of pages for viewing from the books it scans in, and the
one page in the back of the book that would identify this reference was
one of the ones blocked!)

Their main argument is "It should be obvious that when both methods
are applied with equal intelligence and skill, the outside view is much
more likely to yield a realistic estimate."  I don't like this kind of
argument by intimidation.  They follow with a long list of the errors
people can make attempting to use "inside view" models, but don't subject
the outside process to the same scrutiny.  It has more the feel of a
sales pitch than an objective analysis.

Let me discuss a couple of interesting cases where we might try to
apply this distinction.  In each case the inside view produces a more
pessimistic conclusion (the article did mention one example where this
happens, so the analysis is supposed to work in that case as well).

One of these cases came up in the discussion here, Moore's Law.
There are all kinds of inside-view analyses which show that Moore's
Law is about to stop working, if it hasn't already.  Robert and Eugen
both gave examples.  I have seen this referred to in the literature
as the "red brick wall".  This refers to a chart showing challenges
ahead in semicondunctor manufacturing.  Fixable problems are shown in
green; problems that have a theoretical solution are shown in yello;
and problems that have no solution even in theory are shown in red.
A few years ahead the chart fills with red, forming this "red brick wall"
and predicting the failure of Moore's Law.

OTOH the outside view might well look at the previous history of this
trend and the many predictions in the past, one of which was mentioned
by Dirk, that it was about to fail.  Those have always been wrong
before.  Moravec even traces Moore's Law back to the early days of the
20th century.  With such a long history, a simple, "dumb" outside view
forecast would contradict the insiders and predict that Moore's Law will
continue indefinitely.

Another case is one I have been looking at in detail, the Peak Oil
scenario.  This predicts that perhaps in as little as five years the total
worldwide production of oil will peak, forcing consumption to flatten
and then fall.  This will cause oil prices to climb to astronomical
levels and cause anything from a worldwide depression to a disastrous
famine and economic collapse that will kill off most of the human race
within a few years.  See lifeaftertheoilcrash.com for some examples.

Peak Oil is a fundamentally inside-view phenomon.  Analysts delve into
every detail of the oil production situation, producing detailed analyses
of future oil production levels.  These actually go down to the country
by country and even field by field level, predicting exactly how many
barrels per day of oil will be produced by every single oil field in the
world, every year for the next several years.  You can see some examples
of this kind of analysis at theoildrum.com.  These results often come
out with similar answers, namely that future oil production will not
keep up with demand, leading to disastrous consequences.

The outside view, on the other hand, is more sanguine.  We have never
run out of a crucial industrial input without being able to switch over
to an alternative that was even cheaper.  As the Chevron spokesman says,
the stone age didn't end because we ran out of stones.  We can also look
at "dumb" forecasting models like futures markets, which do not predict
any major oil price rises in the next five years.

Peak Oilers hate futures markets and the economists who use them to
say that worst-case P.O. scenarios won't happen, at least not in this
time frame.  The often frame the debate as being between no-nothing
economists who use their blind models and historical precedents, vs
geologists and oil experts who say we're running out, and that economists
cannot create oil where there is none.  It looks to me like a classic
inside/outside issue.

Hal