How Twitter Can Satisfy Elon Musk's Request for Fake Account Clarity

How much social media content is created by real people, as opposed to bots or duplicate accounts? This is a problem that has persisted throughout the internet since its inception, but one that has gotten much worse in recent years with – people who run vast quantities of accounts as a means of providing paid interactions within a network.

This issue has gained more traction in recent weeks, with Elon Musk currently raising concerns as he attempts to buy Twitter. His initial bid of $44 billion was partially based on Twitter's claim that less than 5% of monetized daily active users (mDAUs) are fake, but has since placed his bid on hold as the Twitter CEO, Parag Agrawal, has failed to show proof of this estimation.

The Battle Against Bots and click-farms

Elon Musk’s concerns regarding fake activity are not new. While Twitter has measures in place to locate and shut down bot accounts, such as using CAPTCHAs, click-farm-runners can complete multiple of these with ease as they are not bots but rather humans running 100s of accounts. Twitter also monitors IP addresses and blocks accounts that are attached to the same ones, but click-farmers are able to obfuscate and change their addresses, so this only slows them down. Stoping click-farms altogether is not easy – one approach could be for Twitter to implement a KYC (know your customer) system where people must provide sensitive documents to get verified, but click-farms can simply buy accounts off users who already have completed KYC.

Currently, Elon Musk’s worries about the prevalence of fake accounts have placed Twitter at a standstill, as its executives are yet to produce the number or percentage of bots occupying their network. This has led people to question whether they know how to suitably provide such proof, ideally via the scientific method – as Elon Musk has been asking for in interviews. Identifying real people on the internet and gathering the results in a scientific way is admittedly tough, but it is far from impossible.

Andrew Edi, the co-founder of Idena, a proof-of-person blockchain, has created a model for measuring the number of bots or duplicate accounts on Twitter. Better still, this model does not use any identifying information (such as photos, ID, videos, or other documents), making this a non-invasive method that preserves the privacy of each user. Let’s delve into how this model works.

Finding a Ratio of Live and Real Users

To gain an estimate of real Twitter users, Edi proposes that Twitter run a set of trials. On random days, at random times, a random selection of Twitter users who are actively using the platform will be counted. This can easily be done by checking whether an account is posting, scrolling on their feed, using the search bar, or performing essentially any other activity at the selected time.

Once these users have been chosen, Twitter will then present them with a test, all at the same time, with a time limit for solving. This will be a simple CAPTCHA-style test that is designed to be easy for humans, and practically impossible for machines to pass. But perhaps most importantly, this test cannot be completed by one person running more than one account at a time, making it resistant to click-farms. It is called a Filter for Live Intelligent People (or FLIP) test.

If a user passes this test, then they will be considered a genuine and real person. This will allow us to make a statistical estimate of the ratio of real/fake users, providing a clearer picture of what Twitter’s ecosystem really looks like.

What is a FLIP Test?

At the heart of this method is the FLIP test, which is a simple picture-based task. Users are given two sets of images and are asked to choose which image tells a story. These images contain no text or numbers, making them perfect for people who speak any language and can be completed regardless of a person’s educational background.

The reason FLIPs are able to stop click-farms is that, while they are easy for humans, they require some level of concentration, and because these tests would be deployed all at the same time, it means that click-farmers cannot simply complete them on one device and then go to the next as it will take too long. Although FLIPs are not hard, they do take a little bit of time to evaluate the images being shown, which is important for weeding out duplicate accounts.

What Would We Learn?

This proposed model would help Twitter to evaluate fake activity. It would give an indication of how many real people are actively using Twitter, as opposed to bots, duplicate accounts, and clickfarms. It would not test how many inauthentic accounts have ever been created on Twitter, but rather how many are currently in use, which would give a strong indication of the ratio of real-to-fake users on the platform.

Whether Elon Musk or Twitter’s CEO would be happy with the results they find is a different story altogether. The topic of how many phony users occupy the Twitter space is a contentious one, but until a proper test is run, nobody truly knows just how bad (or good) the situation is. This model is a step towards finding clarity on Twitter, and if it works then it can be easily applied to practically any other social media platform.

Read the full method for measuring real users here.