[ExI] META:MATH Spreadsheet for 'Don't Knows'

spike spike66 at att.net
Sat Jan 2 18:12:04 UTC 2016



-----Original Message-----
From: extropy-chat [mailto:extropy-chat-bounces at lists.extropy.org] On Behalf Of BillK
Sent: Saturday, January 02, 2016 7:46 AM
To: Extropy Chat <extropy-chat at lists.extropy.org>
Subject: [ExI] META:MATH Spreadsheet for 'Don't Knows'

Looks like this could be a good toy for playing with probability ranges in spreadsheets.

<https://medium.com/guesstimate-blog>

Introducing Guesstimate, a Spreadsheet for Things That Aren’t Certain.


BillK

_______________________________________________


Cool!  

BillK, you have been around here long enough to remember how I was predicting time intervals until the next record-breaking largest known prime number would be discovered.  Damien Broderick wrote about those predictions in the 2001 Forge hardbound edition of The Spike, pp86-87.  

I won a ton of "money" in Robin Hanson's Ideas Futures with those predictions, betting on intervals the math hipsters insisted could not be calculated.  After I won thrice, no one would be against me, so I couldn't make any more "money" on that concept, but no matter: the uncertainty interval was growing such that the predictions would have become almost useless anyway.  (Side note: Ideas Futures players, do you remember how Ideas Futures spun down?  (Interesting story behind that.  (Robin Hanson is still around.  (He was in town a couple years ago for Peter McCluskey's wedding.))))

I used the same techniques your friend's guesstimate sheet uses, with a few exceptions.  My distributions were not so much Gaussian, although I used that one at least once.  My model had a Poisson binomial in there for one of the terms, a Gaussian distribution to model cumulative probability, a polynomial growth model and so forth, a less-sophisticated version of what Anders posts here occasionally (thanks for that Anders, you offer your friends and followers a free education in stuff that matters.)  I just did an old-fashioned Monte Carlo with the multiple probability distributions and made my prediction.  I nailed it thrice, but I realized after the fact that I got lucky all three times: I had compensating errors.  Or if not errors, then I had compensating weaknesses in my models.

Anyway, it is quite the topic of discussion at the local Mersenne Prime geek parties we have around here every time GIMPS finds a new record prime.  The true mathematicians argued that the PDF superposition technique is not valid for this purpose (I now think they were right) but the engineers argued that predictions based on superposition of cumulative probability are made all the time (true enough, but that doesn't make it mathematically valid.)  Probability theory is full of paradox.  Engineers still abuse math.

spike

 





More information about the extropy-chat mailing list