As far as anyone claiming they can clean the voter rolls using artificial intelligence, please….
Artificial Intelligence is so wildly exaggerated and misunderstood, I wanted to make this post on what it can and cannot do.
So how do I know anything about Artificial Intelligence?
Well, I was there was when it was being created as well as worked with some of pioneering developers who’s work is now in text, image, video and voice recognition.
The number one things to understand about ANY forms of Artificial Intelligence and/or Machine Learning is it MUST BE FIRST TAUGHT TO RECOGNIZE SOMETHING.
At this time, no one has a very accurate way of identifying INDIVIDUAL “phantom voters” mostly because of the challenges around verifying identity.
People have tried to verify identities online, but you may be just comparing a digital fake to another digital fake because public voter rolls are used to fill most people search sites.
We can however reliably check for the following:
Address Verification = Check if address is valid and deliverable.
Registered in Two or More States = Compare Registrants state to state.
You can somewhat reliably check the following IF the state gives you the full birthdate of the registrant:
Move Records = Check if registrant gave a permanent forwarding address
Death Records = Check if registrant died.
However, it also has been our experience that the accuracy for death and move data starts to fall off in states who do not publish the registrant’s full birthdate.
One poor soul, was attacked by election officials publicly on video when he found out a young man who shared his father’s name (a Junior & Senior) was the one who moved out of state to attend college and the father stayed behind at the same address. He initially thought he found duplicates voting. A full birthdate check would had told him they were two different people.
Accuracy & reliability also falls off a cliff when you start trying to use property tax records. Not everyone owns the home they live in or is listed on the deed. Sometimes the deed will be in one family member’s name and that may be the parent of who is actually living there. We found it is about 25% reliable at checking the people on the voter rolls.
And for all the talk of finding patterns in voter IDs, this could be dismissed as a normal cognitive bias called Apophenia. This is why we see turtles or horses while looking at clouds or people scream because they believe they see Jesus on their toast or tortilla.
Apopheni is why we have Rorschach Tests where we smeared ink randomly on pieces of paper and ask a person “What do you see?”
The voter IDs are auto-incremented in most states, so they will already be patterned and not randomly generated. More importantly, how do you verify that the voter IDs that are being identified with this approach are indeed phantoms?
If you are pointing to registrants with the same name and birthdate, you have to be extremely careful.
Search any state for “Jose Martinez” and group them by shared or similar birthdates and you will quickly discover that’s not an accurate predictor of duplicated phantom registrations.
For the sake of argument, let’s assume there is an algorithm in the voter IDs. You still have to spend a lot of time and money to prove a SINGLE duplicate is a phantom. So regardless, even if there was an algorithm in the Voter ID, this is very tedious, unreliable, and expensive approach that may even ruin your credibility.
On the other hand, we can check 100,000 addresses per minute with the US Post Office with The People’s Audit. It typically only takes a few hours to do a full stack of analytics for large states such as Florida and Texas.
So bottom line… if you can’t accurately and reliably categorize the data and do it quickly and efficiently, you cannot feed huge quantities of data needed into a machine learning model to use artificial intelligence. Therefore, Garbage In = Garbage Out.
And for the the things we can reliably detect such as a mail vote with an undeliverable address, why do we need to feed it to a learning model, when we can already find and flag them using the US Post office?
Artificial Intelligence was designed to take vast quantities of data and learn how to recognize something. Much like teaching a child how to speak, it can see images and learn to spot them in the real world.
For example, one of the first real world experiments with artificial intelligence was teaching it to recognize pets so motion detection alarms would not go off if a cat or dog move about the house while the owners were out with the the alarm on.
The learning models for this were feed billions of images of pets of all types, breeds, sizes and then each one needed photos taken from multiple prospectives while the pets were in different positions. Just think of the amount of pictures a single dog would have sitting, laying, walking, jumping running and rolling over…even how their tail wagged.
Then they had had to be segmented so they could be recognized when sitting together. This took YEARS to develop.
All this so the Artificial Intelligence can be TAUGHT the motion being detected is a pet vs a human. And don’t get me started on how humans could fool the motion detectors into believing they are a pet.
I can’t stress enough how expensive and time consuming this was to develop, and to this day, still is not highly accurate. But it’s attractive to consumers looking to get a new security system and then they have to learn to deal with the occasional false alarm phone alerts…which humans can do far easier than Artificial Intelligence.
So what we do differently with The People’s Audit is monitor the BEHAVIOR of EACH AND EVERY REGISTRANT including verifying their address information each month. So we watch what changes each month and filter out explainable events. For example, a last name will sometimes change because of a wedding or a divorce and people occasionally move.
And after you document every change, and watch that over time, you start to see some suspicious activities. For example, seeing voters that seem to move for every election. Or registrants that are added a few months before an election, voting and then being deleted immediately after that election.
That’s pretty suspicious!
So in conclusion, I hope this post will help you separate fact from fiction when it comes to the use of Artificial Intelligence in cleaning voter rolls.
There has been quite a few people that tricked folks into sharing data and worse getting paid to never deliver a solution.
If anyone knows what they are doing, they should be able to show you a proof of concept at a minimum and demonstrate they are effective at accomplishing something at best.
The People’s Audit, with the help of the grassroots, was able to get 1.2 million invalid registrations removed from our voter rolls in Florida.
The Republican Party of Florida is trying to take credit by saying so many new Republicans either moved here or were recruited to change parties. The reality is, which you can see in the chart above, the Republicans only added 46,200 new voters in 4 years… all while there was historic enthusiasm for Trump.
What made all the difference was the removal of nearly a million Democrat and a quarter of a million No Party Affiliated invalid registrations. We also estimate that nearly a million invalid Republican registrants were removed during the same time. There was only a slight net gain for Republicans.
So cleaning the voter rolls is what turned Florida red for the first time in history!
And not just slightly red. The Republicans now lead Democrats by 1 million registrations in the state of Florida.
Things may change in the future and we are constantly looking for new ways evaluate the voter rolls, but in the mean time, The People’s Audit has demonstrate a proven model for cleaning voter rolls in a large state with a lot of turn over population.
If you feel moved and can donate, we are 100% Grass Roots Fed and appreciate all donations. If you are interested in bringing The People’s Audit to your state, use the form here to start a conservation about it.
Share this post