dimanche 25 janvier 2015

How to Predict the Unpredictable

There is a spoiler at the end of the prologue to this, which is that humans are predictable when they try to be random. Though, this reviewer was more intrigued in the opener to learn of Claude Shannon's 'Ultimate Machine', which she swiftly you-tubed. She's a fan of that already. Anyway, there are a couple of useful truths in this book (which is alternatively titled "Rock Breaks Scissors") Both of them are deceptively simple, and both have generally wider implications and usefulnesses than . . . well, wider than merely intriguing the nerd quotient who would tend to show interest (this includes your reviewer).



The first truth is that however hard people try, they find it very difficult to simulate randomness. In a cute rotation of Nassim Taleb's angle, never mind being fooled by randomness, it is not very easy to attempt to fool *with* it.



The author mostly credits this discovery to Louis Goodfellow, a psychologist who successfully debunked a nationwide telepathy stunt conducted via a radio show in America in 1937 by Joseph Rhine. But it was the Zenith Radio experiment that accidentally unearthed it. The ESP tests produced impressive mass-audience guesses about random selections, and popularised "Zener cards" during the ensuing virality. Goodfellow figured out that even if the transmission was a random sequence, audiences' attempts to guess it were not, and clustered around favourites. He published in a psychology journal a year later. The 'eureka' hidden therein, which was lost for a while in a sceptic/paranormal fight-fest, is far better than a mere falsification: in demonstrating that a touted display of mindreading was fake, Goodfellow actually discovered an authentic form of mind reading.



It's hard for the human mind to generate anything truly random, because random is different from random-looking. If asked to mimic a typical sequence of coin-toss results (for example), most will under represent sequences of consecutive heads or tails. Because these don't look shuffled enough, and folks think the law of large numbers also applies to small ones (which it doesn't, hence the need for 'large'). When asked to think of a random number between 1 and 9, people won't choose the extremes, or the middle, and they prefer odd over even. The result is that 7 is the reply systematically more than it should be. In games of rock-paper-scissors, players don't repeat the same choice three times in a row, and if they lose a throw they most likely switch. Alphonse Chapianis turned study of all this into 'ergo metrics'.



Here's the point: randomness is unpredictable. So anything that fails to be random is predictable, at least to some extent. And this can give those in the know an edge. Unless assisted by software, answers in a multi-choice test will not be randomly distributed nor randomly conceived. The lottery numbers will be, but popular choices for them won't, so there are some selections (for rollover jackpots, which make odds less stratospheric) that are better than others. The type of serve used by skilled tennis players isn't properly randomised either (penalty kicks in football are more likely to be). So if you randomise your serve, using an external cue, your prospects improve. Finally, it is pretty hard these days to compose a digital password or PIN that machines can't discover. Not impossible though.



Having an edge means being able to outperform randomness. That isn't much help if the hurdle to win is way higher than that, as it is in multi-choice tests, and in betting prices that include a spread, and in having to actually be better than your sporting opponent in the first place. But there is lower hanging fruit. A useful occurence is Frank Benford's law of naturally occuring first digits (they are inversely popular to their magnitude, and this is a derived property of a preponderance of exponential distributions in nature and society). Checking a series of numbers (such as Bernard Madoff's investment returns) against Benford's law raises red flags of fakery that are hard to spot otherwise. Idiosyncratic last digits, Chapanis-style, are a check on embezzlement too.



The book's second eureka is a corollary to the first. Given a severe difficulty in producing random series, people also don't recognise it (and doubt it) when it happens. Particularly the incidence of consecutive repetition in randomness which is universally underestimated. This bias is responsible for belief in 'winning streaks' (serial correlation) in sport, gambling, and markets. In fact consecutive wins don't happen more often than chance except in the presence of selective incentives (extra rewards), but the belief is a universal illusion.



Some follow on implications of this--what amounts to a biased crowd some of the time--are explained. In short it becomes a second-derivative parameter of out-guessing: out-predicting the public's collective guess. The latter is a foundational property of financial market prices. And it is possible to beat randomness here by betting against the winning streak bias. But this reviewer thought the lengthy last chapter on beating stock markets was too facile and no use. Since that is the arena she 'doesn't know least' about, she should perhaps be sceptical about the other chapters offering potential advantage as well.



But a richer seam of 'out-outguessing' seems to be in the realm of staying on the good side of differential pricing strategies, which are both increasingly widespread and more thoroughly researched thanks to 'big data'. A 2012 New York Times article reported how some of this was successfully gamed by blogger Emily Vanek. It turns out that exhibiting actions of a perpetual fence-sitter--switching brands, being marginal and equivocal (such as abandoning online shopping baskets after entering contact info), appearing unopposed but unconvinced, and importantly, deleting cookies religiously--will get you more attention, and more inducements--from those after your cash. Potential customers are always treated nicer than existing ones.



This does all feel like something of an effort though, which is of course why it can have an expected return better than nothing. And somewhere lurking in this enthralling text is the revelation that evolutionary fitness is normally better served by jumping to supposedly irrational presumptions. So that's presumably why we're not random. Randomness can survive on its own, and will probably be here longer.





via International Skeptics Forum http://ift.tt/1BbKeBH

Aucun commentaire:

Enregistrer un commentaire