The Economist - 04.12.2021

(EriveltonMoraes) #1
The Economist December 4th 2021 83
Science & technology

Randomnumbers

Flipping heck!


R


andomness is a valuable  commodity.
Computer models of complex systems
ranging  from  the  weather  to  the  stock­
market  are  voracious  consumers  of  ran­
dom  numbers.  Cryptography,  too,  relies
heavily on random numbers for the gener­
ation of unbreakable keys. Better, cheaper
ways  of  generating  and  handling  such
numbers  are  therefore  always  welcome.
And doing just that is the goal of a project
with the slightly tongue­in­cheek name of
coinflips, which allegedly stands for Co­
designed  Improved  Neural  Foundations
Leveraging Inherent Physics Stochasticity.
coinflipsoperates  under  the  aegis  of
Brad  Aimone,  a  theoretical  neuroscientist
at Sandia National Laboratories (originally
one of America’s nuclear­weapons labora­
tories, but which has now branched out in­
to  other  areas,  too).  Dr  Aimone’s  starting­
point  is  the  observation  that,  unlike  the
circuits of digital computers, which will, if
fed  a  given  input,  respond  with  a  precise
and  predictable  output,  the  link  between
input  to  and  output  from  a  nerve  cell  is
more  haphazard—or,  in  the  jargon,  “sto­

chastic”. He wants to imitate this stochas­
tic  behaviour  in  something  less  squishy
than a nerve cell. By doing so, he thinks he
might  be  able  to  tune  the  distribution  of
digits  that  a  random­number  generator
spits out, without affecting their underly­
ing randomness. 

Random doodlings
That  would  be  useful.  Existing  random­
number  generators  produce  uniform  dis­
tributions. (A “3”, say, is exactly as likely to
appear  as  a  “7”.)  But,  as  Dr  Aimone’s  col­
league  Darby  Smith  notes,  the  real  world
that  computer  modellers  are  trying  to
model  does  not  work  like  this.  For  exam­
ple, the temperature in London in Decem­

ber may vary between ­7°C and 17°C, but is
most  likely  to  be  in  the  range  3°C  to  8°C.
Similarly,  vessels  are  more  likely  to  be  in
trouble close to a busy shipping route than
in a remote backwater. Distorting uniform
distributions  of  random  numbers  to  take
account  of  these  realities  is  tedious  and
unsatisfactory.  As  Dr  Smith  observes,  it
would  be  more  efficient  if  the  random
numbers used corresponded to the natural
distribution in the first place.
There  is  also  an  abundance  problem.
Finding random phenomena in nature that
can  be  transformed  into  computer  bits  is
not easy. Often the source is computing it­
self—for  example,  by  gathering  the  last
digits  in  the  numbers  of  milliseconds  be­
tween  keystrokes  made  by  zillions  of  us­
ers. Otherwise, specialist, expensive hard­
ware needs to be used to do things such as
measuring heat flux through a silicon chip.
To  eke  out  these  scarce  supplies,  such
truly random numbers are often then em­
ployed  to  seed  programs  called  pseudo­
random­number  generators.  The  algo­
rithms behind those generate sequences of
numbers  that  have  the  statistical  proper­
ties  of  randomness.  But  this  is  not  the
same  as  the  real  thing.  As  John  von  Neu­
mann,  one  of  computing’s  pioneers,  ob­
served: “Anyone who considers arithmeti­
cal methods of producing random digits is,
of course, in a state of sin.” Moreover, if the
purpose  is  cryptography,  this  method  is
particularly risky. The opposition might be
able to work out the algorithm involved.

How to generate better, cheaper, more abundant random numbers. And why
that is a useful thing to do

→Alsointhissection
84 HowwaterarrivedonEarth
85 Perovskitephonescreens
85 Hybridhelicopters
86 Trypanosomes and honeybees
Free download pdf