How AI Exposed 76er’s GM for Fake Twitter Accounts
Last Tuesday night the President and GM of the Philadelphia 76er’s, Bryan Colangelo, was exposed by an online sports publication for having multiple “burner” accounts on Twitter. Allegedly using five different aliases, Colangelo channeled his inner keyboard warrior to anonymously criticize members and players of the organization while also leaking strategic information, including players' medical reports. Using analytical tools powered by artificial intelligence (AI), an anonymous source was able to find correlations between the accounts' linguistic styles, tweeting patterns and contextual interests. With this information, they were not only able to determine that all five accounts were being run by the same individual but they were also able to identify Colangelo as their admin user. At IMRSV, we employ similar social media monitoring software in the area of threat detection. With the Colangelo story temporarily bridging the worlds of tech and sports, it presents a great opportunity to discuss how our AI and machine learning technology can operate in this space.
In 2018, it shouldn’t take a genius to realize that everything online leaves a digital footprint. The generation that grew up documenting their childhood on social media is now old enough to be entering the spotlight in various industries. Young entertainment stars, sports prodigies, and up and coming executives must be painfully aware of all the embarrassing and inappropriate content they’ve posted online during their teenage years. While this trend appears to be especially damaging in professional sports, the case of fifty-two-year-old Colangelo, is breaking the mold in more ways than one.
In recent years, we have seen how young players have been severely affected by the controversial digital content professional sports teams have found on their personal social media accounts. Sliding down the board on draft day or getting disciplined by their respective organizations can cost highly touted prospects millions of dollars. For any sports fan, the likes of Josh Allen and Laremy Tunsil come to mind immediately. However, what makes Colangelo’s scenario so compelling in the tech world is the fact that his secrets were uncovered from unknown accounts, not his known personal platforms. AI was able to identify similarities between the five accounts without explicit directional instructions. Any human being can manually scan content when told where to look for it, it’s when information exists in limbo that solutions become more technical.
While this type of technology may be better served combatting more important issues, Colangelo’s blunder is a perfect example of how machine lead solutions can be used to solve problems that were otherwise unidentifiable. At IMRSV we employ similar technology that can be used to monitor social media for patterns of threatening behaviour. Our software uses NLP algorithms to manipulate and understand human language and speech. It can accurately contextualize hundred page documents or entire verbal conversations. Additionally, it uses data modelling and inference techniques to achieve a variety of tasks including sentiment analysis, text classification, information extraction, translation, topic modelling and summarization.
Colangelo’s current situation reminds us how the internet can be used anonymously for malicious intent. It can be very easy to sit behind a computer screen and post seemingly inconsequential content on the web. It's when this online behaviour turns violent or aggressive that our detection technology becomes vital. Looking specifically at security and defence, there is a constant struggle to identify the unkown-unkowns in order to protect against unidentified threats. This issue is shown with increasing frequency, as the revealing digital footprints of criminals are often only exposed after their illegal act has been committed. In the real world, our applications can be used to actively monitor massive open source databases (ex: Twitter) to intercept threatening content in real time before a crime is committed.
Beyond detection of individual content, we’ve been working on predictive tools to anticipate large scale events before they happen. Our preliminary research has included studies related to Twitter patterns before and after major military events. Terrorist organizations, such as ISIS, pollute the internet with propaganda to spread their political and ideological agendas. By accurately identifying this type of content, we’re slowly learning to recognize how spikes in correlated online material may be used to indicate an actual threat.
The reason why this type of technology can thrive in volatile environments, such as Twitter, is due to its adaptive capabilities. With language being a live entity, it’s dynamic and can change very fast. Elements such as acronyms or slang are ever evolving and essential features of any social media environment. AI can evaluate, learn and adapt to these changes which is essential for any robust solution. It's this versatility and independent learning capabilities that allow it to perform a variety of impressive functions.
Ultimately, the need for social media monitoring tools for personal, commercial, and state purposes is more prominent than ever. Especially as we push further into an age of digital supremacy, it’s becoming increasingly difficult to separate valuable information from all the “noise” that exists on the web. While our solutions at IMRSV are more tailored towards security, Colangelo’s Twittergate represents a controversy that both the tech world and the sports world can enjoy collectively.
Written by: Simon Hicks