Overview

VoteFlipper is more accurate than almost any political consultant or strategist and provides answers millions of times faster. Specifically, VoteFlipper focuses on two questions:

  1. Will someone vote in any given election?
  2. Who are they likely to vote for?

The answers to both questions help political causes figure out where to spend resources to attract like-minded folks. How likely someone is to vote in any election is important for making sure resources are not wasted, while predicting party affiliation is important to make sure messaging isn’t incorrect. With both of these questions answered, VoteFlipper answers one combined question:

  1. How “flippable” are they?

Understanding strength of party affiliation (“flippability”) is required for uncovering new opportunities. VoteFlipper measures the way voters vote, as well as their demographics, donation patterns, and geographic variables to determine their strength of party affiliation. This strength also changes with time. For example, if a voter sat out 2016’s general election despite being affiliated to a party, their party loyalty probably isn’t high.

VoteFlipper's Structure

VoteFlipper is a council of three artificial intelligence programs. Each program has developed a set of neural networks that progress from a general overview to detailed specifics- much like our physical brains do. This helps eliminate the “noise” of bias or statistical outliers. With noise eliminated, each voter structure gets calculated by an almost individualized neural network for maximum chances of a perfect prediction.

Meet the Council

VoteFlipper’s AI council consists of Protocols Balthazar, Caspar, and Melchior.

Protocol Balthazar is a ‘tree’ structure of neural networks designed to understand voters who have known voter variables; properties such as voting history and proper registration dates play heavily to Balthazar’s strengths. Balthazar produces True/False predictions on if someone will vote or not in any given election.

Protocol Caspar is a ‘tree’ structure of neural networks designed to understand voters who have unknown voter variables. Caspar works in the spaces that Balthazar doesn’t. Caspar values sociodemographic inputs highly, such as age or race. Like Balthazar, Caspar produces True/False predictions on if someone will vote or not in any given election.

Protocol Melchior is a ‘tree’ structure of neural networks and formulas designed to work with the output from Protocols Balthazar and Caspar. Melchior is the one who predicts party affiliation and then further refines its predictions to determine strength of party affiliation. Melchior is specifically suited to be applicable to any election due to its combination of Balthazar’s and Caspar’s inputs and a variety of optional human-controlled biases.

How does VoteFlipper Work?

VoteFlipper works by taking any anonymized voter’s data and running it through the council until the three AI’s come to an agreement. Spanning the political spectrum, VoteFlipper was trained on tens of millions of voters with hundreds of inputs per voter.

With the protocol separation and ‘tree’ structure, VoteFlipper can be seamlessly configured to be used in any primary or general election by any candidate or any cause. The current public demo of VoteFlipper runs all three Protocols with approximately 40 ‘tree’ branches total. It sits on a single, medium sized web server. The full version of VoteFlipper has several hundred ‘tree’ branches. VoteFlipper becomes more accurate with more branches and is able to offer additional insights into geography-based demographics, donations patterns, and party affiliation.

VoteFlipper Does what the Others Can’t

Before VoteFlipper, developments in AI have not reached the field of political analysis in any meaningful way. Current political software is like a calculator; it does math for the inputs it is given. VoteFlipper changes that by being akin to an advisor, making connections between the data that humans may overlook because of bias or error.

For instance, one of the major political parties uses a calculator that scores voting likelihood on a 0 to 1 scale. A score of 0 is “not likely to vote” and a score of 1 is “likely to vote.” This approach is broken because the people using this software have to guess what value someone’s score cutoff is before they attempt to spend resources on them. This means lots of lost opportunities.

Unlike other political software, Protocols Balthazar and Caspar both make a choice and stick by it. Balthazar is particularly strong and scores accuracy at almost 90%. Caspar isn’t far behind. Simply stated, when VoteFlipper makes a voting prediction, either True or False, the prediction is going to be correct about 9 out of 10 times.

VoteFlipper has one other distinct advantage compared to its competition; it’s lightning-fast. New instances of VoteFlipper can be created in a business day. Similarly, on one mid-grade workstation, VoteFlipper can calculate the entire USA’s voting population (~153 million) and provide per-voter results in just a few days.

A Brief History and Use

Human inefficiency in political consulting is often the reason campaigns lose. Thus, in 2016, VoteFlipper was started to discover if politics and AI could be merged properly. To do that, these questions needed answers:

  1. Could AI accurately predict elections and avoid handcrafted algorithms/polls?
  2. Could a system be built that supplements consultants or provides guidance to campaigns of any size?
  3. Could it be generalized to any election and be used by any candidate?

In order to accomplish these goals, VoteFlipper needed a lot of raw data, customized databases, a detailed understanding of human social connections, and a good trial environment.

Throughout 2017, the data was gathered and databases were created. However, without a real-world trial, VoteFlipper couldn’t get the properly-tuned social connections it needed.

Fortunately, in 2018, VoteFlipper was recruited into a Non-Party Affiliated race in a tourism heavy city (2.7 million visitors per year). This provided the perfect environment for testing as NPA races are notoriously difficult because candidates do not get the luxury of party identification written on a ballot. In addition, the transient nature of the pilot city’s population meant traditional voter tracking methods were harder to use and VoteFlipper needed to be able to understand people moving in and out.

During the primary, VoteFlipper’s alpha was a great success. Out of 17 candidates, 7 on the ballot, the campaign that used VoteFlipper took ~49% share with a non-incumbent candidate. However, there was plenty of room for improvement, as the alpha still relied on human inputs and tracking.

About two weeks after the primary, VoteFlipper’s beta was fully mechanized and rolled out. The beta gave the campaign new information about districts that the alpha overlooked during the primary. On election night, the beta proved to be a resounding success with transformational results.

For example, the 6th place voting district, the main area where VoteFlipper’s beta suggested the campaign go, skyrocketed to 1st place. Absolute change in districts was as high as 25% (ie, 50% vote share to 75%).

The third version of VoteFlipper, the one that the demo is based on, is bigger, smarter, and faster than its predecessors. Using what was learned from the real world and all of 2018’s results, VoteFlipper launched with three full protocols and is completely capable of handling any election thrown at it; federal or local, primary or general.

VoteFlipper's Future

While VoteFlipper might be fully launched now, no software is truly complete. The main goals for future VoteFlipper include:

  1. More versions of VoteFlipper that are specifically designed for segments of voters. For instance, VoteFlipper will have spinoffs such as “Young Voter” VoteFlipper.
  2. More integrated data. Future VoteFlipper versions will understand financial data and purchase history to increase advertising relevancy to VoteFlipper’s “flippable” voter predictions.
  3. Smart Email Lists. VoteFlipper will help campaigns by predicting when an email will have the least amount of annoyance and maximum attention span. This will avoid spam and decrease perceived “annoyance.”
  4. Smart Donation Requests. Similar to overload with email spam, interested parties are often spammed with donation requests. VoteFlipper will tell campaigns when an interested person is most likely to give a donation and send a request at the predicted time.

Thanks for reading. If you'd like more info, check out the faq. Also, check out the contact page if you are interested in working with VoteFlipper!