From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Sun Jan 01 2006 - 15:31:11 MST
Damien Broderick wrote:
>
> But, as an hour or two reading the claims actually made by
> parapsychologists would demonstrate, the size of the deviations from
> chance typically encountered in traditional psi studies are of the order
> of one extra correct guess per hundred (where p is 1 in 5), if not
> markedly less.
>
> The Mega Millions lottery requires you to identify five numbers out of
> 56 and an additional one number out of 46. Quickly now, tell me your
> brilliant protocol for achieving this aim given a phenomenon that
> manifests itself as an excess of (at best) one extra correct guess in a
> hundred. It takes you 100,000 guesses to attain one standard deviation,
> which is hardly enough to instil confidence. True, you're only risking a
> dollar, so you might be prepared to stop after 100,000 guesses.
Damien, if you carefully read my original proposal - rather than John's
gloss - I spoke of doubling your money on each round. Matching the Mega
ball only, in the Mega Millions lottery, is good for a $3 payoff on a $1
ticket, if I recall correctly (I'm drafting this email in a brief absence
of Internet access). So you can risk half your money on each round, and
double your total funds on each win. Richard Loosemore claimed results on
the order of 1 in 200, which should let you win most times on a 1-in-46
problem.
(This turned out not to be correct. Matching the Mega ball is a $2 payout.
Still, betting half your money, with significance on the order of 1 in
200, should let you consistently increase your funds over time.)
> I've bored you all, I can tell. Killfile the gullible idiot. Such
> incredible horseshit. Don't look, don't look.
Here's another funny thing about parapsychology's robust, repeatable
results - aside from the fact that you know and I know that they will
vanish like soap bubbles popping if you try to use them to make money.
Contemporary parapsychologists consistently find equally significant
statistical evidence for negative psi as positive psi. Similarly, when
someone reports that subjects can significantly predict future events,
shortly thereafter someone using roughly the same protocol reports that
subjects can significantly manipulate past events.
Now if we are talking about physical phenomena, manipulating past events is
considerably harder to swallow than predicting the future.
And if we are talking about biology, antipsi is much less of an
evolutionary advantage than psi.
And if we are talking about fraud, manipulating the past is so ridiculous
that most fraudsters would try to fake something more plausible.
Bad statistics, on the other hand, are symmetrical between psi and antipsi
phenomena, between prediction and manipulation, and between past and future
events. If bad statistics produces a false appearance of positive-psi
precognition of future events, then bad statistics should just as easily
produce statistically significant evidence for a psi-miss manipulation of
past events.
Furthermore, on the hypothesis of bad statistics, you will find tiny effect
sizes that are nonetheless statistically significant. The larger the data,
the tinier the effect size that is nonetheless statistically significant.
You will never find anything clear, blatant, and obvious - when someone
claims large effect sizes it will always turn out to be unrepeatable or
fraud. And when you try to use the effects to make money, or do anything
else useful, they will go away.
You can, I suppose, try to carefully explain this away as something other
than bad statistics; carefully explain why antipsi is an advantage, and
carefully explain why quantum weirdness in the brain can manipulate events
that have already happened. But it is a bad sign when you have to
carefully explain things that just don't look the way you would expect them
a priori to look. It is like carefully explaining why God is omnipotent
and omniscient and omnibenevolent, and yet the physical world contains
hurricanes and tsunamis and starving babies. You can try to come up with
reasons why that's all part of God's plan. But plainly the world does not
look at all like we would a priori expect it to look, in the presence of a
powerful benevolent entity that could fix stuff, even just the really
really awful stuff like torture of children by the Inquisition. The world
does look exactly like a neutral product of natural law. Why does psi look
not at all like we would expect psi to look, and exactly like bad statistics?
And why aren't you running out to try and double your money in the lottery?
There must be a part of you somewhere that is using essentially the same
model as I am, for you can predict in advance which disappointing
experimental result you will have to excuse.
I am willing to believe that parapsychologists have followed all the
conventional rules of science. Until parapsychologists win the lottery, or
it's demonstrated that it was all fraud after all, this implies that the
conventional rules of science are too weak. It implies that orthodox
statistical methods are such as to sustain an entire field of
investigation, with papers regularly published on seemingly significant
data, in the total absence of any real effect. This is a disturbing
prospect to face up to. But it is not nearly as disturbing as taking at
face value that psions can manipulate the past.
-- Eliezer S. Yudkowsky http://intelligence.org/ Research Fellow, Singularity Institute for Artificial Intelligence
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:55 MDT