More on barcodes, etc

From: Alan Dechert <dechert_at_gmail_dot_com>
Date: Sun Apr 17 2005 - 21:07:33 CDT

Arthur wrote,

> How many blind people say they prefer the current approach?
> How many sighted people say they are concerned about the
> barcode? When OVC's CTO speaks about technology along with
> an acclaimed and accomplished technologist, I think that
> retreating behind "go and demo it" is unfortunate.
> Certainly a demo with a barcode is easier to produce.
> It's not at all clear that the production version should use
> a barcode. The paper talks about OCR'ing the text, by the way.
>
Barcode issues surfaced in some off-list discussions we were having
recently. We've had extensive discussions about barcodes in the past,
and some of the issues remain unresolved. Arthur's email warrants a
detailed response, so here goes.

First I want to talk a little about official OVC policy -- where it
comes from. The OVC president/CEO (currently me, Alan Dechert)
represents official OVC policy. This is by necessity. Someone has to
articulate these policies publicly. Sometimes these policies appear
in print; sometimes these policies are conveyed in email to other
interested parties; sometimes these are conveyed personally at events
at which OVC is involved. It's my job to know what those policies
are. When issues arise in various fora, I can't say, "I don't know, I
have to ask the board." I have to know the issues and know where OVC
stands on these issues.

Generally speaking, OVC policy is set by consensus. Theoretically,
the OVC board of directors has the power to set policy. Some policies
are set that way but most policy decisions have been made outside of
board meetings. This informal consensus rule generally holds for the
board too since we've never decided anything with some close vote like
4-3, or 3-2, etc. Almost everything we decide at board meetings is
unanimous with some abstentions here and there. We tend to vote on
things only after we've had enough discussion to agree on what we want
to say.

A great deal of OVC policy was grandfathered-in -- the product of
discussions that took place before OVC was founded. Mostly, these
policy decisions were made in conjunction with our various academic
partners over the years. In April of 2001 Henry Brady asked, "so how
will people make money with this system?" We started to describe
(conjuring, modeling) how the organization might work.

A lot of things were decided APR 2003 - July 2003. Someone recently
asked if OVC endorses IRV. OVC is neutral on scoring methods because
one of the main discussants back then, Arnie Urken, insisted that we
remain neutral on scoring methods. Everyone else involved then agreed
and no one has ever made a case for changing that policy. That's how
the policy was made. During this period, Doug Jones played a major
role shaping the OVC model. My wife Lori and Jay Tefertiller (ISIS
Technology) worked out a lot of legal and organizational issues (we
decided to be a "consortium"). Arthur Keller played a major role too
beginning in July of 2003. Doug and I cobbled together the bylaws in
NOV - DEC 2003. Ed Cherlin, David Mertz and quite a few other
engineers contributed ideas

In 2003 (and earlier), we consistently heard DRE makers and their
"supporters" (dare I say paid shills?) insist that paper printouts
were unfair to reading impaired voters since these voters cannot
verify the vote on paper. This position was expressed in a videotaped
presentation by Marion Taylor of the League of Women Voters at the UC
Santa Cruz forum on electronic voting we organized (mainly Bob Kibrick
and Arthur Keller) in OCT 2003.

Long before the debates in 2003, I had come up with the idea of
putting a barcode on the long edges of the printout so a blind person
could utilize the barcode while keeping the text hidden from view
(Actually, Ryan Ronco, assistant ROV in Placer County CA suggested the
barcode to me in APR 2001. Ryan was mainly suggesting it as a way to
facilitate tabulation. I moved it to the edge as an accessibility
feature).

As I designed the ballot for our demo, I figured out what barcode
system I wanted to use. I decided on Code 128 because it could encode
all the information we needed to encode and scanners could be had dirt
cheap (I bought a bunch of scanners for about a dollar each). I
looked for GPL'd Code 128 software and found Jan Karrman. Jan joined
the project. Most of the technical issues with the barcode were
worked out between Jan and I. Sidebar: An example of some OVC policy
making .... David Mertz insisted that it might be possible for someone
to discern some information about the ballots by eyeballing the
barcodes. I never believed that. We came up with an obfuscation
routine that was easy to implement and cost nothing, so I went along
with it. If it was all up to me, I wouldn't have bothered with it.
The gain was small but the cost was negligible. In other words, OVC
policy is sometimes contrary to what I believe. If the consensus
opinion is contrary to what I believe, I'm glad to go along with it so
long as it doesn't conflict with some concepts I hold and consider
basic and important.

The scheme we demonstrated for reading-impaired ballot verification
was very important. We publicly shattered a very persistent myth.
Shawn Casey O'Brien, a disabled rights activist, showed up at our OCT
2003 UC Santa Cruz forum and gave a speech denouncing the paper trail
because it was useless to blind people. I demonstrated why he was
wrong about that, and ended the debate. You don't hear anyone making
that claim anymore (blind can't verify paper ballot). All opinion
leaders know it's false. No disabled rights activists have shown up
at any of our events since OCT of 2003 to make that claim. OVC can't
take all the credit for that since others have made the point too. I
do think it's fair to say we played a major role in dispelling this
myth. Since accessibility has been a major theme in voting
modernization, it was absolutely necessary to break this myth in order
to get the idea of a computer generated "paper trail" accepted for
voting.

Some credible technologists (and others less technical) have said that
we should do away with the barcode since it seems to be contrary to
the goal of transparency in election administration. I don't happen
to believe that, but I'm happy to go along with it since switching to
OCR does not impact the overall veracity of the OVC approach. I am
perfectly willing to say, now, that while our demo utilizes the
barcode, our production system may not. I see pros and cons. I see
tradeoffs. I see perceptual issues.

Before getting into some of the pros and cons, let's lay out some main
principles as a backdrop for the discussion.

First Principle: the voter is paramount.
--------------------------------------------
The voting system should be designed and built with this principle in
mind. Security, trustworthiness, cost, clarity, simplicity,
ease-of-use, accuracy, accessibility, auditability, inclusiveness,
convenience are all important issues in election administration. But
what's secure, trustworthy, inexpensive, clear, simple, easy-to-use,
accurate, accessible, auditable, inclusive, and convenient for
election administrators must be secondary to what's secure,
trustworthy, inexpensive, clear, simple, easy-to-use, accurate,
accessible, auditable, inclusive, and convenient for voters.

The voting system is for voters, and paid for by voters.

Second Principle: costs must be minimized
---------------------------------------------------
The four billion dollars being spent on voting modernization is a
fluke of history. Before 2002, states and counties have never had
much money for voting equipment, and should not expect a lot of money
to be available in the future for new voting equipment. Budgets are
constrained everywhere. Expensive voting systems means long lines and
voter disenfranchisement. Ultimately, open voting will take hold
through market forces -- open voting systems must be very inexpensive.

Also, if America is to maintain a leadership role in promoting
democracy through out the world, we have to show how a great voting
system can be built and maintained inexpensively. How many
jurisdictions in developing countries could contemplate deploying the
$5,000 AutoMark system? It's a ridiculous waste of resources.

Third Principle: election process must be fully auditable
----------------------------------------------------------------
Institutional protocols must be in place to enable full public audit
of all aspects of election administration. If there is anything a
voter wants to know about election administration, he or she should be
able to get answers in whatever level of detail desired without ever
hearing, "trust us" or some other indication that information about
the voting system is not available to them.

Fourth Principle: Election administration is not -- and can never be
-- a computerized process
---------------------------------------------------------------------------------
The process of administering elections mainly involves procedures
carried out by people. Computers can assist in the process, but can
never be expected to run elections. Computers can enhance the
security and accuracy of election administration, as well as keep
costs down.

There are other important principles, but these are the three main
ones I want to keep in mind right now. In light of the foregoing, I
have the following observations about the barcode v. OCR debate with
respect to the computer generated summary paper ballot;

1) Machine reading should be incorporated into the process of counting
votes. Computer reading can act as a check against human error, and
vice-versa. Security and accuracy will be best when both computer and
human counting are employed.

2) A barcoded ballot may be more tamper resistant than one with no
barcode. We expect that rules surrounding the OVC barcoded summary
paper ballot would include a rule that says something like, "No
pencil, pen, etc. marks of any kind are to be made on the summary
paper ballot. If you want to change anything on the ballot, you must
re-do the process and print a new ballot. Any hand markings on a cast
ballot will be ignored." The rule should be ironclad. Voters will be
encouraged to carefully review their ballot on-screen before printing.
 If there are scribbles on a barcode, it may still be readable or the
other duplicate barcode could be read instead. We could also place
another copy of the barcode in the text area if necessary, and/or
another version of a barcode (not easily recognizable as a barcode)
could be included that may be read by a special reader in case a
ballot has been defaced. An OCR only summary paper ballot may not
work well with such a rule (hand markings ignored). A cross-out may
cause the OCR reading to fail. If the mark-out is complete, or the
selection torn off, it may be impossible to determine what was printed
on the ballot.

3) OCR-only is not any more transparent, and may be even less
transparent than a barcoded ballot. For example, the architecture we
are employing involves using the machine reading to create a
reconstructed electronic ballot image (REBI) and compare that with the
electronic ballot image (EBI) from the voting machine for that
particular ballot. If the contents of the REBI and EBI match, the
machine reading was accurate. However, this proof alone is not
adequate. If malware is installed, it's conceivable that while the
REBI and EBI match, they don't match the printed text. The lie would
work differently for OCR as opposed to barcode, but could still give a
phony result in either case.

You vote for Smith, but Jones is recorded in the EBI. Smith is
printed on the ballot.

Let's say a tiny number "3" is placed in a strategic location on the
printout For the OCR system, it can read the ballot and report
"Smith" but also sees the tiny 3 and knows that it should record this
as a vote for Jones, the third candidate in the contest. The REBI and
EBI both say Jones. Jones gets the vote despite what is printed on
the ballot.

The barcode could be similarly encoded to lie to the voter saying,
correctly, that "Smith" is printed on the ballot while Jones is
getting the vote.

Given that OCR software is quite complex, it would be difficult to
spot malware that could cause printed ballots to reflect the voter's
choices but record the electronic vote otherwise.

So, regardless of whether we employ OCR or barcode, a manual check of
the electronic record against the printed text must be standard
procedure. A computerized voting system with a paper ballot is no
more secure than a computerized system without a paper ballot if all
the checking is done by computer. So long as audit procedures are in
place to check the printout against the electronic record, malware has
a very low chance of success on a system with the summary paper
ballot. The OCR/barcode debate doesn't matter in this regard.

4) OCR may be no more transparent or secure than using a barcode, but
it may appear more secure. Appearance is important so this should
count for something. Most likely, the average person understands no
more about how OCR software works than s/he understands about how
barcode software works. On the other hand, barcodes are ubiquitous --
trusted tools of commerce. It's not clear which the public would
prefer. Trials and surveys are needed.

5) Our experience has been that people that raise objections about the
barcode are usually objecting to machine counting altogether -- these
people are often nontechnical. They want hand counting instead. They
may be no happier with OCR. We also find that these objections are
not necessarily unmoveable. Most people get over the objection
without too much difficulty when it's explained how the system works
and that manual checks will be in place to verify that the readout
from the barcode matches the printed text.

6) OCR may be less accurate than barcodes. The barcode with good
error detection should be 100% accurate. That is, if you get a read,
the odds are astronomically high that it's right.

7) An OCR scanning station might be more difficult for blind voters.
If the blind voter has to remove the ballot from the folder to insert
it into some device, there is a risk that the voter will drop the
ballot or have some other difficulty -- especially considering that
blind voters sometimes have other disabilities. The only way to
really find out would be to build both OCR and barcode stations and
see which work better for reading impaired voters.

8) An interesting test I'd like to see: two different mock elections,
one utilizing OCR and one with barcode. Set up the verification
station for each. Will normal reading/sighted users use the
verification station? Some small percentage would use the
verification station for barcoded ballots. My guess is that for the
OCR system, they will use it much less if at all. In both cases, if
the mock election was repeated with the same subjects, I predict that
usage of the verification stations would be less the next time. It
may be that our verification station was for the demo only.
Especially if we go to OCR, I don't think it would be used enough to
justify setting up. This may be a good thing. If we have one system
set up for voters that need the accessible system, the verification
station may be integrated with that.

9) OCR print may compromise human readability of the ballot. If
you're not worried about OCR, you can optimize the text for
readability for humans. A fixed width font may work best for OCR but
will be less efficiently read by humans. It may not fit as easily on
the printout as a proportional font. If we adopt User Centered Design
principles and place the voter as the primary user, this may rule out
OCR.

10) Precinct level tabulation may be slower or more expensive with
OCR. I suspect that a station to count, say, 500 ballots, will be
more expensive and/or take longer than a station that reads the
barcodes with a hand-held scanner. Again, this is speculation and
real data would only come with trials and experiments.

Summary
--------------
In summary, I still think the barcoded ballot with manual verification
that the text matches the electronic ballot image, is the most
efficient way to go. OCR does not eliminate the need for the manual
check. Open source helps too but is not a cure-all either.
OCR may be more saleable, on the other hand. Some of the questions
may be approachable even without major funding. But real answers to
some of the questions above would require some real tests.

I have no problem with the assertion that an OVC production system may
not include a barcode. I would like to see some trials though.

Alan D.

_______________________________________________
OVC discuss mailing lists
Send requests to subscribe or unsubscribe to arthur@openvotingconsortium.org
==================================================================
= The content of this message, with the exception of any external
= quotations under fair use, are released to the Public Domain
==================================================================
Received on Sat Apr 30 23:17:09 2005

This archive was generated by hypermail 2.1.8 : Sat Apr 30 2005 - 23:17:22 CDT