Did Artificial Intelligence Alter Colorado's 2020 Election?
Database experts can show that Colorado votes were manipulated in 2020-21, using an algorithm that carries hallmarks of AI. Election official Tina Peters backed up the records & was targeted.
Colorado's top election officials and political leaders are awfully quiet these days after two computer science experts recently demonstrated evidence of vote manipulation in the election databases of Mesa County and Grand Junction in the state's 2020 and 2021 elections.
In the case of the Colorado databases, the outcome is akin to discovering that a business is keeping a second set of "books.” In the case of the voter database, they found another voter "adjudication" file, which is where ballots that cannot be read by the tabulation machines are parked so a human can decide the voter's intent.
A March 19 report by database expert Jeff O'Donnell, and Prof. Walter Daugherity, who teaches computer science at Texas A&M, documents how the database enabled “potentially unauthorized and illegal manipulation of tabulated vote data during the 2020 General Election and 2021 Grand Junction Municipal Election."
As a result of their six-month review of the forensic copies of the election database, the "totals for those elections are impossible to verify." They call into question "the results and integrity of Mesa County’s 2020 General Election and the 2021 Grand Junction Municipal Election."
Their work documents their discovery: a digital reloading of 20,346 ballot records into new election databases. A time stamp shows that thousands upon thousands of those ballots were copied within seconds.
Whoever or whatever triggered the new voter files screwed up. An election clerk noticed that that ballots she had already adjudicated were appearing in an adjudication file — again.
The clerk called Dominion's software support to explain the problem. Dominion reportedly told her it "could not replicate the issue."
And then, just like that, ballots that had already been processed stopped re-appearing in the adjudication files.
What's going on here? Dominion, whose U.S. headquarters are in Denver (other offices are offshore), did not respond to extensive queries about the findings.
The findings carry hallmarks of AI techniques, whereby a "trained" algorithm pursues outcomes such as parsing out votes and creating a new vote database, amid all the votes coming into the system. The discovery evokes "Machine Learning" concepts among core principles in the rise of Artificial Intelligence (AI) and Big Data.
Dominion’s installation of its "Trusted Build" update on its [election management network] began in May of 2021. The Colorado Secretary of State ordered all counties that use the software to install the update.
As O'Donnell and Daugherity note, as did many software experts at the time, the “Trusted Build” upgrade effectively over-wrote records of the 2020 and 2021 election.
Statutes are clear about voting records: they must be preserved. For example, 52 U.S. code § 2108 requires that voting systems used in elections for federal office produce a record with an audit capacity.
It also lays out penalties for "destroying, removing, or delaying delivery of election records."
52 U.S. Code § 10307 prohibits any person acting under color of law to “...willfully fail or refuse to tabulate, count, and report...” the vote of any person entitled to vote.
The 88-page report by the researchers lays out in detail how the "Trusted Build" update – a major change to the system's code and log files which traditionally should happen before, not after, elections – "destroyed all data on the EMS hard drive, including the batch and ballot records that evidenced the creation of new databases and reprocessing of ballot records."
The timing of the "Trusted Build" is also notable. It was pushed out around the same time that Arizona's Senate was launching its forensic review of Maricopa County's 2020 election.
Maricopa County also used Dominion software in its 2020 election. Dominion refused to cooperate with subpoenas of the election network logs and files as part of the state senate’s forensic audit.
Colorado's Secretary of State, Jena Griswold, did not respond to queries about why it was so important to force all the Colorado counties that use Dominion's voting software (62 out of 64), to over-write their election logs with the "Trusted Build" rollout after the 2020 election.
We now know that Big Tech's thumbs were all over the 2020 election to help the Biden campaign. We now know that social media platforms suppressed and censored the New York Post's reporting on Hunter Biden's laptop that raised serious questions about Biden family influence peddling with Chinese and Russian government officials.
We know much more about tech billionaire Mark Zuckerberg's $400 million that he donated to governments in 2020. Funds went to 92% Democrat-heavy districts under the excuse of "Covid relief." The Republican districts that did not receive grants were, apparently, not so "Covid relief" lucky.
Mystery ‘Trusted Build’ Rollout in Colorado Lingers
The researchers’ findings in Colorado’s Mesa and Grand Junction election records beg the question: Did Artificial Intelligence alter votes in the two Colorado county vote databases?
If so, the revelation raises major ethical questions, and would be one more bill of indictment against Big Tech's role in the 2020 election – from censoring information on its platforms about the Biden family's business dealings that go straight to the public trust, to Zuckerberg shoveling over $400 million into the election to influence and even take over the actual counting of ballots, which happened in Green Bay, Wisconsin’s 2020 election.
The election software vendor Dominion did not respond to extensive queries about whether machine learning techniques are at play in its code base and whether it would open it up for review.
The question has become even more relevant amid an ongoing criminal investigation of Arizona's 2020 election in Maricopa County following its forensic audit.
Arizona Attorney General Mark Brnovich has said in media interviews that attorneys representing county election officials have, for the first time, admitted they used AI-infused software to check absentee ballots for signatures. This raises ethical questions as well as legal ones. The state senate’s forensic review of the 2020 election shows that workers were feeding absentee ballots into scanners without checking the signatures late into the night on Nov. 3rd.
Now we’re told they used AI?
Out of 2.1 million ballots cast in Arizona’s Maricopa County in 2020, in what is considered the most extensive election audit in American history, two of the four audit firms found some 20,000 absentee ballots did not have proper signatures in a state that Joe Biden won by just over 10,000 votes.
AG Brnovich's letter also notes that over 200,000 absentee ballots were counted with zero chain of custody. Did they come from a ballot trafficking operation?
Brnovich's April 6th letter to Senate President Karen Fann says the AG's investigation is pursuing criminal charges in "various election crimes."
The investigation has "reached the conclusion that the 2020 election in Maricopa County revealed serious vulnerabilities that must be addressed."
Colorado’s office of attorney general did not respond to queries about the legality of the “Trusted Build” update.
The findings call into question whether Mesa County’s Clerk, Tina Peters, should be the one facing charges of tampering with Mesa County's election files; her job is to preserve election documents, which is what she did.
Had Peters not succeeded in making a forensic copy of the Mesa County hard drive, for which she was attacked, targeted, and is now facing criminal charges, none of this would have come to light.
Griswold, a highly partisan Democrat with no prior experience in state-level election oversight before she was elected Secretary of State, has confirmed the Mesa County database Peters backed up was a perfect forensic copy. This was among the files that O'Donnell reviewed for about six months, and Prof. Daugherity then reviewed to confirm O'Donnell's findings.
What else could have sparked the vote manipulation in the database?
Another theory is that it might have been a so-called "closed loop" controller. The reviewers, and other experts looking at the files, have observed that the database carries hallmarks of a "proportional–integral–derivative" (PID) controller algorithm of the sort used to help a pilot fly though fog, or to program cruise control in a car.
The scenario might work like this: If it looks like a selected candidate might win or lose by a chosen final percentage, the data inputs would be to keep counting votes – or inject ballots, say, from an extra database it had just created.
Until experts get a look at the base (source) code, the scenario is speculation. But the confirmed outcomes – two sets of vote files (like two sets of "books”), deserve an answer, as well as why state officials stood by while election records were wiped out across the state.
Machine learning (ML) is considered a subset of AI
Of all the major trends in software development today – across critical industries such as banking and finance, utilities, and defense systems, ML techniques are raising some of the loudest alarm bells among policy makers.
The simple reason is that ML techniques help software teach itself. Depending on the datasets to “train it,” the software could then what it was programmed to do, such as count paper ballots accurately. Or do something different.
A sub-chapter to the story of millions of American citizens getting involved with electoral integrity is that tech experts are stepping up to analyze voter databases and in many cases, finding practices that would get them fired in the private sector.
In many precincts in battleground states where the results are still in question, they have discovered management practices that are incompetent, if not sloppy by design.
In Arizona's Maricopa County, for example, county elections supervisors did not have administrative-level access to the voting database of the state's largest county. That means Dominion Voting Systems was essentially managing the election results. The company refused to hand over system-level passwords so a forensic auditor could review the voting logs and confirm best practices were deployed in the network.
Pennsylvania's voting database called SURE was quietly scrapped [in 2021] without much, if any, public notice. That appears to be as close an answer the public has received from the Commonwealth’s Secretary of State about an estimated 100,000-vote variance in the 2020 election results, where Biden won by 80,000 votes.
In Georgia, Dominion Voting Systems software is used statewide and relies on touchscreen voting without any paper ballot to compare the results to in a full audit, which a federal judge ruled violates its statutes. The state is also now investigating a ballot trafficking operation during 2020, and reportedly received some $45 million in Zuckerberg’s grants that helped turn government offices into “get out the vote” operations — mainly benefiting Democrat precincts.
The issues O'Donnell and Daugherity document in Colorado raise questions about whether algorithms were pre-loaded, or activated, to manipulate how votes were managed in the database.
"Without both cyber and database management system expertise, and unfettered access to database records and computer log files (many of which were destroyed by the actions of the Secretary of State) from the EMS server, the manipulation would be undetectable," their report says.
The report also says the absence of secure hash algorithm (.sha) files for each digital ballot image makes the authenticity of each digital ballot image, and the ballot-level record for those ballots, impossible to verify.
"Regardless of whether the voting system was connected to an external network or device, even momentarily, or whether a pre-installed software or algorithm was triggered by an external command or complex set of variable conditions, the execution of manipulating software or algorithm could plausibly be responsible for the results described in our findings."
As a result of the "unauthorized creation of new election databases during early voting in the 2020 General Election on October 21, 2020, followed by the digital reloading of 20,346 ballot records into the new election databases," there is no way to tell "the original voter intent recorded from the ballots," the report says.
Dominion's April 4th statement and updated April 9th statement ignores these findings. It instead links to statements from Griswold and a Republican elected official who were quoted as saying Colorado’s voting system is considered the “nation’s gold standard."
Prof. Daugherity says if any students in his database course configured the database findings they documented, they would have flunked his course.
Election-tech expert Prof. J. Alex Halderman of University of Michigan told Scientific American in 2018: “As paperless computer voting machines were being introduced, there were many computer scientists who—before anyone had even studied one of these machines directly—were saying, 'This just isn’t a good idea to have elections be conducted by, essentially, black box technology.'"
The Election Assistance Commission (EAC) has not confirmed whether it actually approved Dominion's base code as part of its "Logic and Accuracy" testing that goes into the certification it hands out to promote confidence in election integrity.
The Colorado report raises new questions about whether Big Tech also had its thumbs in the code base with ML and AI to manipulate results. A full examination of the code base would answer the question.
If the algorithm was written by a very angry, Trump-hating, progressive Big Tech type of programmer opposed to President Trump’s re-election, that would describe at least one top Dominion product executive. It also raises questions about Colorado’s "Trusted Build” rollout. Dominion has refused to explain why it should be trusted. #
Ed. note: the story was edited for grammar and style after publication.