Internetworkers: http://www.ibiblio.org/internetworkers/

[Jeremy Portzer <jeremyp AT pobox.com>]

Roger Austin wrote:
> Christian Stalberg wrote:
> > January 6, 2008  Can You Count on Voting Machines? By CLIVE THOMPSON
> <snipped>
> > specter of partisan hackers throwing an election. But the real problem 
> > may simply be inherent in the nature of computers: they can be precise 
> > but also capricious, prone to malfunctions we simply can’t anticipate.
>
>   This is not true. You can anticipate most errors, or at least you can
> estimate where most of the errors will come from during testing. Any
> capable analyst will understand the weaknesses of a system if they are
> paying attention and have the budget to address them.
>
> > This is because computer scientists understand, from hard experience, 
> > that complex software can’t function perfectly all the time. It’s the 
> > nature of the beast. Myriad things can go wrong. The software might have 
> > bugs — errors in the code made by tired or overworked programmers. Or 
> > voters could do something the machines don’t expect, like touching the 
> > screen in two places at once. “Computers crash and we don’t know why,” 
> > Felten told me. “That’s just a routine part of computers.”
>
>   This is a very stupid statement and the reporter should have challenged
> this Felten person about it. I guess it fit the reporters agenda so they
> didn't bother.
>   Touch screens and the software subsystems are more robust that that.
> Try it the next time you are at the Teeter checking out and see what
> happens. Those grocery store POC systems are much more complicated than
> any voting machine should be.
>   It is easy to blame the programmers.

[sorry for horribly delayed response]

Having worked at the 'Teeter, and been involved in the maintenance of 
those machines, I can tell you that sometimes the self-service machines 
DID crash and we didn't know why.  :-)

But seriously, they crashed because they were running Windows NT 4.0 - a 
robust operating system for its time, but hardly up to date, and also 
not really designed for embedded systems type work.  There are plenty of 
ways to harden things that weren't necessarily used on these machines. 
(This was about 3 years ago so things may be upgraded now.)

Also, the presentation layer was the only part that crashed.  The 
"back-end" servers and databases carefully guard the transactions - what 
really matters in a grocery store.  When the system boots back up, you 
can continue your order exactly where it left off when it crashed - you 
don't have to re-scan anything.  But I think this does underscore the 
point - computers are complex enough systems that you can't control 
everything - you can't find ALL bugs - and thus you must put in 
contingencies to deal with anything and everything.

I do think the voting machines have to be more complex, and more 
fault-tolerant than the grocery store systems.  As Tanner pointed out, 
they require a careful balance between anonymity and robustness to 
preserve data.  In a POS system, you can rely on a myriad different ways 
to preserve data, with tracking by transaction references, VIC cards, 
credit cards, etc. - anonymity is not important when you're a duly 
authorized person accessing the database.  Also, the costs of failure 
are not particularly high.  If a POS machine crashes you can just go 
ring up the order on another one.  It's a pain, but not a political 
isssue.  Even if you're not sure if your credit card has been charged, 
there are ways of verifying that later - you can check your statement 
and see if you got charged twice or not.  This kind of check-and-balance 
exists throughout the financial world and is much more challenging in 
voting machines.

I agree that fundamentally optical-scan machines seem to be the best of 
all worlds.

I'm living in Australia now where voting machines are banned, and have 
been for years.  All balloting is done by paper, with hand-counting. 
They do have various mathematical systems for allocating the counting so 
that some groups of ballots are counted twice for verification.  This 
helps catch any errors introduced by humans.  There have been arguments 
that machines would be more cost-effective but this actually turns out 
to be difficult to justify in the long run, given the high capital and 
maintenance costs for machines that are only used once every few years. 
 A much more versatile resource - the human being - has few capital 
costs, low training requirements, etc.

The actually tallying of the ballots is fully computerized, of course, 
made especially necessary by Australia's complex preference system.

--Jeremy Portzer