Farshad Kheiri, Lead Data Scientist, and Hicham Mhanna, VP of Engineering, BCG Digital Ventures, discuss an exciting new proof-of-concept which leverages blockchain technology and machine learning to combat fake news, originally presented at our Startup Talks series in Manhattan Beach.
There’s nothing new about fake news. Yellow journalism and propaganda have been around for over a hundred years. But in the age of social media, the ability to spread false information has increased exponentially, giving legitimate news stories a run for their money.
Most recently, the fake news phenomenon was cited as a contributing factor to the outcome of the U.S. Presidential election. The result of the election took much of the nation by surprise. But it’s not so shocking when you consider the fact that: Over 50% of the most viral articles covering the election turned out to be fake stories; 62% of Americans now rely on social media over traditional news sites as their primary source of information; social media is an enabler of virality and therefore a breeding ground for fake news.
Though technology has helped spread fake news, it can also help mitigate its effects.
Prior to the 2016 election, websites like Snopes, Politifact and Factcheck manually examined the news to assess the validity of articles. The process was laborious. Relying on human screeners (and thus subject to human error), it provided limited coverage and scalability.
Google has been trying to revamp their page ranking system to order pages based on the facts they contain, not just their relevance to search terms and popularity. While this sounds like a great solution, creating the right algorithm is a complex process requiring a lot of time and effort, and Google is still far from building a practical solution.
Under pressure from public figures and various governments, social media and search engine giants have begun taking steps to tackle the problem. In the most prominent case, Germany instituted a 50 million euro fine to penalize social media sites that keep fake news on their platform. Tech leaders such as Tim Cook publicly called on Mark Zuckerberg to help crack down on fake news, as Facebook commonly gives fake news articles a platform of distribution.
Facebook was able to approach the problem with a pragmatic methodology, focused on detecting bots and suspicious users with in-house algorithms and third-party verifications. However, because Facebook still primarily relies on manual, third-party fact-checking services such as Snopes and Politifact, many fake news items can still go viral before they are flagged.
Even without considering timely detection and flagging, Facebook is facing major challenges mitigating the consumption of fake articles. In fact, their platform solutions have had an inverse effect, as articles that get flagged as fake can stimulate curiosity and actually attract more readers. What’s more, Facebook generates a significant portion of their revenue through clicks (regardless of the legitimacy of the shared link), creating a conflict of interest between their underlying business model and the detection of fake news.
It remains to be seen if those with vested interests, such as Facebook, are qualified to fight the fake news epidemic. But what we do know, is that as more people turn to social media as their primary source of information, it will only get worse. It’s more important now than ever to develop a foolproof way to prove what’s real, and what’s not.
To address these ‘trust gaps,’ DV has created Geppetto: a first-of-its-kind platform which uses cost-effective, tamper-proof methods to screen news articles in real-time, accurately identifying those which are valid and those which are fake.
That sounds great, but how does it work?
Copy the link of any article you can find online and paste it into Geppetto. Geppetto will then “read” the article to check for any falsehoods using a “veracity engine,” consisting of several natural language processing and machine learning models which score the legitimacy of the articles and the reliability of validators (voters) and publishers. The algorithm then spits out whether the article is “true” or “not true.”
Online content travels at rapid speed, allowing fake news to evolve at a shocking pace. As such, the performance of any model built on historical data can decay with equal velocity. To address this issue, we employ a continuous learning (CL) algorithm to update the model in a perpetual loop. The CL component requires a knowledge base to store historical data and re-evaluate the models.
Through this process, Geppetto’s analysis will becomes increasingly granular, constantly improving its ability to spot red flags and level of accuracy. Simply put, the more articles the system analyzes, the smarter it becomes.
Users can engage with Geppetto in three ways: (1) As a publisher launching content directly via the platform; (2) As a ‘voter’ who can validate/negate a piece of news; (3) As a consumer searching for a piece of news with a specific thesis (i.e., “I think this article might be fake, so I’ll use Geppetto to double check”).
We know machines are smart, but we also know they can only make so many logical conclusions. To address this problem, Geppetto uses a voting system enabled by blockchain technology to ensure all stored data is tamper-proof and decentralized.
First, Geppetto encourages users to vote on how trustworthy they believe an article to be. Geppetto then permanently attaches those answers to the hashed version of the article using blockchain technology. This way, no one can ‘trick’ the system into believing an article is true when it isn’t, and we are able to fact check the machine’s mind against the aggregate human mind. The platform then stores the newly published news and verified news by the voters.
While it remains to be seen how the fake news phenomenon will unfold, we are certain that blockchain and machine learning can not only be used to create technologies that are useful and exciting, but that will change the world for the better, such as Geppetto. This proof-of-concept is just one example of the many possible applications for this technology we see on the horizon.
To learn more about how technology can help change the wold visit our website.