AI and cognitive psychology rant (getting more and more OT - tell me if I should shut up)

[Alex Martelli]

Andrew Dalke wrote:

> Alex:
>> But I can for example observe the way people play
>> bridge, and their rationalizations about why they've done X or Y;
> 
> I recall reading that computer generated shuffles, based on PRNGs,
> gave different hand distributions than expected by expert bridge players.

Yes, it did, back when computer shuffling was a novelty, say in the
early 70's, right when I was starting.  These days, all major championships
have computer shuffling, so all competitive players are long used to it; and
with online bridge also using it, more and more average players have
similarly adjusted.  "Wilder" (less balanced) hands are more common these
days (as they follow the theoretical probabilities) than they were in the
'50s and '60s.  Studies of the hands from world championships in those
decades could neither confirm nor deny this -- there just were not enough --
and apart from world championships other hands were only recorded
selectively, with a bias towards "interesting" [thus, less balanced] ones, 
so, no chance to double check there; but that is the consensus.  It seems
to go with the change in top experts' style, from the soundness and prudence 
of a Goren or Roth or the Italian Blue Team in the '50s/60s, to the 
recklessness of Meckstroth-Rodwell, Zia-Rockwell, or the _current_ Great 
Italians these days.

So, imperfect shuffling is not simulated in any computer dealer or analysis
program that I know of (including mine in Python, which I've repeatedly used 
at the request of my friend Marty Bergen, a once famous player and now 
famous writer, to help me answer his theoretical and probability questions 
for his books).  People _do_ perform chi square tests and the like on all 
kinds of aspects of deals collected from OKbridge and other major online 
bridge services, as well as on deals played in all major championships, and
no deviation from theoretical expectations has arisen there.


Alex