From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Sat Dec 31 2005 - 10:26:45 MST
Richard Loosemore wrote:
> I agree, the most robust stuff is precognitive. That was the design of
> the one experiment of mine that came out with an extremely significant
> result.
>
> It used to drive me crazy, with my physics background. I did all the
> work to collect the "guesses" of my subjects, but what I knew was that
> the targets that they were trying to guess at would not be determined
> until after all the data was collected and the pieces of paper were in
> my sweaty palms.
>
> Then I generated the random seed from a previously specified set of
> global weather statistics, used it to dive into the RAND Corporation
> Book of One Million Random Digits (the world's most enthralling book) at
> my start point, then compared about 12,000 binary digits in the tables
> and in my subjects' responses.
>
> A lot of pain later (didn't have a computer in 1984, had to do it all by
> hand), out pops the predicted correlation between run score variance and
> self-reported motivation, at the 0.005 confidence level.
>
> To those of you not well versed in experimental design, that means I
> predicted the correlation ahead of time, and the first time I did the
> experiment, it turned up and if the effect was pure fluke, then it was a
> 1 in 200 chance.
>
> But what a bitchin' result! All those guesses matching up with some
> predetermined random numbers in a book, with the only degree of freedom
> being the seed point into the book!
>
> Darned if I can make sense of it.
>
> Richard Loosemore
If you can get results significant at the 1 in 200 chance level,
you can predict lottery numbers well enough to double your money on each
round. It's the same amount of Shannon information. All you need to do is
declare that you will generate the winning pseudo-random numbers from the
seed that is the winning Mega ball, or whatever, and then check in advance
to see which seed's pseudo-random numbers work best. Think of the
publicity! You could, using these standard experimental protocols that
produce robust and repeatable results, settle the parapsychological debate
once and for all! In a way that no one could possible deny! And make a
heck of a lot of money on the side! Just keep doubling your money on each
round until you've got a few million dollars; or poll enough subjects to
get enough precognitive information to win the lottery outright.
Yet somehow I amazingly predict, in advance of the data, that the very
*instant* you try to match lottery numbers instead of random numbers in a
book, the robust repeatible precognitive effect will vanish like a
beautiful but strictly imaginary snowflake falling into a bucket of real,
dirty warm water.
And on some level, you know it too, which is why you tried to match random
numbers in a book, instead of coming up with an encoding that would let you
match the same amount of Shannon information about future results to the
winning Mega ball.
I too am disturbed by the apparent fact that parapsychological results pass
every standard test of science - and yet the same results vanish, like tiny
sparkling droplets of dew evaporating in the searing light of the morning
sun, as soon as someone tries to use exactly the same precognition protocol
to double their money in the lottery. It means that the standard tests of
science are too damn weak, and for all we know, a lot of other
"statistically significant" science is also junk.
-- 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 - 04:00:54 MDT