Dill on E-Voting ProblemDownload PDF
ELECTRONIC VOTING: AN OVERVIEW OF THE PROBLEM
David L. Dill
Professor of Computer Science, Stanford University
Founder of the Verified Voting Foundation and VerifiedVoting.org
The winners of an election are usually satisfied with the outcome, but it is often more
challenging to persuade the losers (and their supporters) that they lost. To that end, it is
not sufficient that election results be accurate. The public must also know the results are
accurate, which can only be achieved if conduct of the election is sufficiently transparent
that candidates, the press, and the general public can satisfy themselves that no errors or
cheating have occurred.
Unfortunately, the advent of paperless electronic voting (e-voting) is moving us away
from election transparency. E-voting technology is extremely opaque. No one can
scrutinize some of the most critical processes of the election, such as collection of ballots
and counting of votes, because those processes will be conducted invisibly in electronic
circuits. Voters have no means to confirm that that the machines have recorded their
votes correctly, nor will they have any assurance of that their votes won't be changed
In the presidential election of 2004, almost 30% of American voters reportedly used evoting
machines, and this number is increasing because of the deadlines set by the Help
America Vote Act, (HAVA) and the funding made available for new equipment by that
The basic problems of e-voting can be understood without an in-depth knowledge of
computer technology. A helpful analogy was proposed by computer security researchers
Drew Dean and Dan Wallach: Suppose voters dictated their votes, privately and
anonymously, to human scribes, and that the voters were prevented from inspecting the
work of the scribes. Few would accept such a system, on simple common-sense
grounds. Obviously, the scribes could accidentally or intentionally mis-record the votes
with no consequences. Without accountability, a system is simply not trustworthy,
whether or not computers are involved.
Are computers different in some important way from human scribes? Computers can
execute programs accurately and with great speed, but they are designed and
programmed by people who are no more reliable than our hypothetical scribes. Indeed,
the construction of completely accurate and reliable hardware and software is one of the
great unsolved problems of computer technology -- a problem that is actually growing
worse with the burgeoning complexity of computer systems.
Computer systems can also be subverted intentionally. Most people are familiar with the
"hacking" of systems by outsiders, often through the internet. Experience in computer
security has shown that resisting such attacks is extremely difficult. The attackers are
often very creative and determined, making them formidable adversaries. However, the
greater threat to most systems is from insiders. Software can be modified maliciously by
people with legitimate access before it is installed on thousands of individual voting
machines. Indeed, much computer crime is perpetrated by insiders, because it is easier
for them to commit crimes, and they are less likely to be caught. There is no reason we
should be more trusting of insiders in the election industry than in other industries, such
as gambling, where sophisticated insider fraud has occurred in spite of extraordinary
measures taken to prevent it.
Many lay people assume that malicious software can somehow be detected by inspection
or testing, but, perhaps surprisingly, there is no reliable way to do this. Computer
systems are the most complex artifacts known; finding cleverly hidden malicious code is
much harder than finding a needle in a haystack. (For some benign and fun examples of
how easy it is to hide things in software, search for "Microsoft Easter Eggs" on the
In the public debate, it may seem that there is some disagreement among technologists
about the dangers of paperless e-voting, because the same two computer scientists
opposing paper ballots speak at almost every forum. However, the overwhelming
consensus of technical opinion is that e-voting is dangerous, and that voters need to be
able to verify that their votes were properly recorded. The "Resolution on Electronic
Voting", which I wrote in January 2003, has been endorsed many of the top researchers
in computer science, including the authors of several of the most widely read texts on
computer security. Also, the Association of Computing Machinery (ACM), the largest
professional organization of computer technologists, has taken the position that "... voting
systems should enable each voter to inspect a physical (e.g., paper) record to verify that
his or her vote has been accurately cast and to serve as an independent check on the result
produced and stored by the system." A poll was conducted on this question, and fully
95% of the respondent ACM members agreed with the statement.
Auditing and Paper Ballots
Systems are usually made trustworthy through independent checks, called "auditing."
Secret ballots make voting uniquely difficult to audit. In other areas, such as finance and
e-commerce, audit trails necessarily include the identities of the parties involved in
transactions, but voting systems go to great lengths to destroy this information by design,
as required by our system of secret ballots.
To understand how voting systems can be made auditable, let's return to the scribe
analogy. One solution would be to eliminate the problematic scribe and have the voter
fill out the ballot and deposit it in a ballot box. Or we could make the scribe accountable
for his work by allowing the voter to inspect the ballot and deposit it in a secure ballot
box (or watch the scribe do so). Analogous solutions will work for voting. We could
simply use paper ballots marked by hand and counted by optical scanners; indeed, many
U.S. voting jurisdictions use these systems, and have for decades, and the systems are
highly accurate. Or "voter verifiable printers" could be attached to touch-screen
machines to produce tangible ballots that voters could inspect.
Instead of attempting to solve the unsolvable problem of insuring the integrity of
computer hardware and software, these measures place the responsibility on the voter to
check that his or her ballot is filled out properly. Imperfections in the technology,
whether from unreliable computers or unreliable pens, can be tolerated because each
voter can check that his vote was handled properly. With these paper ballots, it is
possible to do a meaningful manual count, for election auditing or in a recount, because
the records being counted will be known to reflect the voters' intent.
This generic scheme has been described by various cumbersome phrases, such as "voterverifiable
paper audit trail" or "voter-verified paper ballots." Unfortunately, the awkward
"voter-verified" modifier is necessary. A casual reading of HAVA has led many to
conclude that it already requires voter-verified paper ballots, since it requires a "manual
audit capacity." Unfortunately, this language is being interpreted to allow printing paper
ballots from electronic memory after the close of the polls, for use in a manual recount.
Of course, this interpretation renders the "manual audit capacity" nearly useless, because
these paper records may not reflect the intent of the voters, who were unable to check the
electronic records on which they are based. Hence, it is necessary to ensure that the
voters are able to inspect the paper ballots before they are cast.
Paper ballots are not a magical guarantee of accurate and fraud-free elections. Indeed,
there is a long history of errors and election fraud with paper ballots, but those problems
stem from inadequate procedures, inadequate checks and balances, and inadequate
auditing, not from the use of paper. Improving the trustworthiness of our elections will
require attention to many other issues. If machines are used to count the ballots, they
must be doubled-checked sufficiently using manual counts to detect and deter systematic
fraud. The physical security of the paper ballots must be maintained from the time when
they are marked by the voter until the last recount is complete. Above all, all aspects of
the election must be open to public scrutiny, and the public must actually scrutinize the
conduct of elections.
The conduct of elections in many places falls well short of these ideals. But the solution
to that problem should be to improve those procedures, not to eliminate the evidence that
could be used to detect errors or fraud. Using paperless electronic voting has been
likened by Kim Alexander, President of the California Voter Foundation to "dealing with
fraud by eliminating the accounting department." An ongoing nationwide effort to
improve election practices is needed very badly.
These arguments against paperless e-voting are often dismissed on the grounds that "no
election technology is perfect." While this is an undeniable truth, problems vary with
different technology. Paperless e-voting is more dangerous than paper ballot systems
because it opens the door to wholesale errors. A single bug, or malicious software
installed by a single individual, could be distributed to thousands of machines around the
country, which could then undetectably change a very large number of votes. And,
contrary to the frequent assertions of vendors and some local election officials, there are
no "checks and balances" that can reliably prevent or even detect these problems without
How did we get there?
The trend towards paperless e-voting has been driven by the laudable goals of
enfranchising more voters and increasing the accuracy and integrity of the voting system.
However, a crucial mistake was made, which was to make policy about computer
technology without being informed about the limitations and hazards of that technology.
Policymakers, without independent knowledge or advice about computer security, were
assured by vendors and other proponents of the technology that it was safe, and did not
Many claims are made of the superiority of e-voting, for example: the machines are more
accurate, and allow users to correct mistakes; they are accessible to people with
disabilities who cannot use paper or mechanical voting machines without assistance; they
are easier for poll workers to use; and they save the cost of paper ballots.
Even taken at face value, these advantages would not justify sacrificing the transparency
of our elections, but many of these claims turn out to be illusory when examined more
closely. Studies have indeed shown that the best e-voting equipment is more accurate
than the worst technologies, such as pre-scored punch cards, but most of the same studies
show that precinct-based optical scan systems are actually more accurate than e-voting
machines. (When using a precinct-based optical scan system, the voter fills out a paper
ballot by hand and then places it in the optical scanner, which counts and stores the
ballot. If there are too many votes for an office or a stray mark that prevents the ballot
from being read properly, the scanner rejects it so the voter can correct the problem
before casting a vote).
New equipment has recently become available to make it possible for voters with
disabilities such as blindness to use optical scan ballots while voting privately; for
example, there is a machine that allows voters to read and mark an optical scan ballot
using a touch-screen or audio interface. Furthermore, while e-voting machines are
accessible in theory, it is unknown how many voters with disabilities have been able to
use them successfully in practice. The Silicon Valley Council of the Blind surveyed their
members after a recent election only to discover that very few were able to use the new
machines that had just been purchased in Santa Clara County, California.
The claim that e-voting is easier for poll workers to deal with is implausible, and seems
not to have been confirmed by experience. Dealing with a workplace full of computers is
rarely easy in this day and age. Counties acquiring new e-voting equipment have had to
implement extra measures to make sure there are technically capable poll workers (a
difficult task) and to have technicians on call to deal with machine problems. Indeed, in
recent elections many of the observed failures of e-voting equipment are blamed (often
unjustly) on the inability of workers at the polling places to properly setup and operate
Finally, e-voting machines cost at least three times as much as optical scan systems to
purchase. Even ardent proponents of e-voting admit that this cost difference cannot be
recouped in less than 15 years, which is greater than the lifetime of most computerized
equipment. There is also some evidence that on-going costs for support and maintenance
of e-voting equipment are higher than were estimated in many jurisdictions.
Where do we go from here?
A trustworthy election system depends on three factors: technology, procedures, and
observation. Changes are needed in each of these areas. In many cases, election laws
will need to be amended.
As was argued above, we need technology that allows each voter to verify that his or her
vote was correctly captured. At this time, the only technology that can realistically meet
this need is paper, because most voters can verify the contents without computer
mediation (which is inherently untrustworthy), because it can be written indelibly, and
because the procedures for protecting paper are understandable by ordinary poll workers
There are now several proposals for voter verification based on advanced cryptography.
These systems are intriguing, but there are still many challenges to be met before they
can be responsibly deployed in governmental elections. First, they are not widely
understood even by computer science researchers, and have not yet been subjected to the
in-depth scrutiny by independent experts that is necessary to be reasonably sure that a
system is a secure. Indeed, some experts have commented that these schemes are much
more complex than secure computer and communications equipment that has been
certified for U.S. military and intelligence applications. Second, the operational and
logistical details for using these schemes in real elections have not been worked out.
Finally, and most importantly, these systems are completely non-transparent to the
average voter, who cannot begin to understand how they work or why they should be
The second component of a trustworthy election is the use of appropriate procedures. If
paper ballots are used, they have to be protected, and the processes for storing,
transporting, handling, and counting them have to transparent. Ideally, members of the
public and non-governmental organizations as well as political party representatives
should be able to observe all of the steps of an election, including machine testing,
polling place operations, counting of votes, auditing and recounting.
One of the most important reforms that could be adopted is routine auditing of elections
by choosing a small random sample of the ballots, and manually counting them. Careful
auditing should occur regardless of whether an election was close or whether there were
apparent problems, results of the audit should be made public, and problems detected by
the audit should be investigated. By adopting random manual audits universally, we can
distinguish the idea of objectively checking an election, to reassure voters of its integrity,
from recounts requested by candidates, which are often perceived as tactics for reversing
an unwelcome election outcome. Audits are also a mechanism for election quality
control. Routine audits will often catch problems in the conduct of elections that are not
close, so they can be corrected before they cast the outcome of a closer election in doubt.
The final factor in trustworthy elections is independent observation. In too many states,
election laws and practices do not allow independent observers to be present during
crucial parts of the process, such as the testing of equipment. In others, only certified
representatives of candidates or political parties may observe. This is fundamentally
wrong. Elections exist first for the people, not for candidates and parties, and the people,
the press, and representatives of non-governmental organizations must be allowed to
observe. Finally, the public has an obligation to participate in elections, not only as
candidates, but as poll workers and witnesses to the process. This participation should be
encouraged by election officials as well as by independent organizations such as ours.
Fortunately, some of these reforms are already underway. Many people in the U.S. are
belatedly recognizing that too many states and counties rushed into using e-voting in an
overreaction to the problems in the 2000 presidential election. The issue is now drawing
a great deal of attention. Many states have changed their plans to convert to e-voting, and
now insist on paper ballots. At the Federal level, there are multiple bills in the House and
Senate that would require voter-verified paper ballots.
Behind these developments is a large and effective grass-roots movement. The "paper
trail movement" is unusual; it does not follow conventional partisan and ideological
divides, and it has the participation of prominent computer science researchers, who have
great expertise in the relevant areas of technology, and who rarely speak out on other
policy issues. Last summer, more than 350,000 U.S. citizens submitted petitions
demanding voter-verified paper ballots. This movement will continue to gain momentum,
in part because increased attention to elections will expose more and more problems with
the use of e-voting.
The recent controversy about electronic voting has focused attention on the conduct of
elections, which had been neglected by the public and policymakers for far too long.
Although this attention is uncomfortable for many in the elections community, it is
healthy. Ultimately it will result in stronger foundations for our democracy.