As the COVID-19 pandemic continues its “samahara tandava” over the human race, propagation of fake news has emerged as another major roadblock in fighting this pandemic, especially in social media. Facebook Inc, the social media giant, which in fact owns top three of the world’s social media platforms, namely Facebook, Instagram and WhatsApp, has been stepping its effort to curb the rise of fake news since the beginning of the pandemics.

Recently, Facebook researchers have modified their Artificial Intelligence (AI) systems to curb the onslaught of hate and misinformation being spread via the platform as the company’s fleet of 15,000 human content reviewers or moderation contractors are at home with paid leave.

The researchers have focused on the system upgrade to combat malicious interference by organised campaigns aiming to sow discord and spread pseudoscience. “We have seen a huge change in behaviour across the site because of COVID-19, a huge increase in misinformation that we consider dangerous,” said Facebook CTO Mike Schroepfer in a call with the press.

Even though the company has engaged with dozens of fact-checking organisations around the world, the misinformation mutates into memes, doctored images and videos, making it hard for the organisations to identify and pull down misinformation.

A normal computer vision algorithm would have difficulty judges images that are screen-grabs of original news, doctored photos, identical visually but have alternative wordings, memes that use sarcasm to drive home the point. The system would either rate these as completely different images due to those small changes or all the same due to overwhelming visual similarity. Training the system to note these minute difference is not only tough, but as information gets cloned really fast on Facebook, training seems daunting.

However, analysing gazillion images is a speciality of Facebook. It already has a readymade infrastructure that compares photos and searches for faces, etc. It just needs to undergo advanced training to be taught what to look for.

“What we want to be able to do is detect those things as being identical because they are, to a person, the same thing,” said Schroepfer. “Our previous systems were very accurate, but they were very fragile and brittle to even very small changes. If you change a small number of pixels, we were too nervous that it was different, and so we would mark it as different and not take it down. What we did here over the last two and a half years is build a neural net-based similarity detector that allowed us to better catch a wider variety of these variants again at very high accuracy.”

The result of Facebook’s effort is SimSearchNet, a system dedicated to finding and analysing near-duplicates of a given image by close inspection of their most salient features. SimSearchNet is currently inspecting every image uploaded to Instagram and Facebook — billions a day.

However, one area that has proven especially difficult for automated systems, is memes. The major challenge with memes is that “the meaning of these posts often results from an interplay between the image and the text,” wrties Devin Coldewey and Taylor Hatmaker on Tech Crunch. “Words that would be perfectly appropriate or ambiguous on their own have their meaning clarified by the image on which they appear. Not only that, but there’s an endless number of variations in images or phrasings that can subtly change (or not change) the resulting meaning,” they noted.

While, systems are able to understand language, and classify images, how those two things relate is not so simple a problem. The Facebook researchers note that there is “surprisingly little” research on the topic, so theirs is more an exploratory mission than a solution. The technique they arrived at had several steps. First, they had humans annotate a large collection of meme-type images as hateful or not, creating the Hateful Memes data set. Next, a machine learning system was trained on this data, but with a crucial difference from existing ones.

Facebook’s system combines the information from text and image earlier in the pipeline, in what it calls “early fusion,” to differentiate it from the traditional “late fusion” approach. This is more akin to how people do it — looking at all the components of a piece of media before evaluating its meaning or tone.

Although the system has gone under rigorous training, the overall accuracy is around 65-70%, not quite ready for deployment. To help advance the art, Facebook is running a “Hateful Memes Challenge” as part of the NeurIPS AI conference later this year; this is commonly done with difficult machine learning tasks, as new problems like this one are like catnip for researchers.

Image by Gerd Altmann from Pixabay