Fact checking today is a reactive process in which journalists wait for a falsehood to begin spreading virally and then publish their final verdict long after the falsehood’s spread has tapered off and the damage done. Much of this delay stems from the amount of time and research it takes for fact checkers to investigate a claim and determine its veracity. What if we inverted this process and required every social media post to provide external attribution for its claims and used deep learning algorithms to compare the statements in the post to the original material it cites as its source? Could this “reverse fact-checking” largely curb the spread of digital falsehoods?
The greatest limitation of today’s fact-checking landscape is the time and effort it takes fact checkers to investigate a claim. Collecting evidence, reaching out to organizations and experts for commentary and summarizing the resulting information into a final verdict is an extremely time-consuming process that offers few opportunities for efficient scaling.
One approach that is currently being explored is the concept of automated fact-checking, where machines could extract the core claims and argument structure of a post and compare it against known fact checks and other material to determine whether the post contains claims that have been previously disputed.
Yet such post-facto automated fact-checking at scale and robust enough to be deployed in the real world is still in the future. Most importantly, the sheer creativity of human language and the diversity of topics discussed online makes it difficult to determine authoritative sources for every claim.
Moreover, even the best natural language understanding algorithms today still struggle immensely with the simplest tasks and are trivial to confuse.
What if the burden of verification was placed on users themselves?
Imagine if every social media post that made a claim other than expressing the author’s personal opinion was required to cite authoritative external evidence to support its claims. A post offering that the author prefers Sprite as their favorite beverage or that they prefer a particular political candidate would be allowed without verification since it is merely an expression of opinion.
In contrast, a statement of fact, such as that a politician made a particular statement or that a certain consumer product contains a certain unsafe chemical or that unemployment rates are at a certain level, would all be required to attribute each statement to a URL or other clear universal identifier in the same way an academic paper cites each of its claims to their original sources.
Similarly, first-person witnessed imagery would require additional documentation to verify it has not been digitally altered. A clip of a politician making a statement in a closed-door donor meeting would require the person to use a third-party app to digitally sign the image with additional verifying information, such as a wide shot of the crowd to prove the image was taken inside the venue or perhaps through a trusted third-party application that timestamps and verifies the image as coming from the general vicinity the user claims it does (while masking the precise GPS coordinates for safety and privacy). Rather than a user-specific signature that could be used by security forces to trace the image back to its source, this could involve a cellular provider-offered or WhatsApp-provided signature. While this would not prevent “camera box” attacks and similar more sophisticated firmware compromises, it would at least offer some reprieve from the run-of-the-mill modifications and deep fakes that proliferate today.
In turn, automated algorithms would check the cited references to verify that the claims made in the post accurately represent the sources they are cited to. Thus, a claim that unemployment has changed by a certain amount would be verified against the sources it cites to confirm that the number in the post is what is stated in the cited source.
Of course, such a reliance on “authoritative sources” raises its own thorny methodological questions of just what counts as such a source.
Perhaps a first step might be to bypass the question of what counts as an “authoritative source” and merely bring social media users closer to the sources that underlie the posts they consume. Thus, if a poster cites a claim to a particular website, the verification algorithm would only confirm that that site does indeed provide that statistic and display its domain name alongside the claim in the post. Consumers of those posts could then decide whether they consider that source to be reputable.
In fact, once such a system existed, one could imagine a generalized fact-checking markup for social media posts in which algorithms would annotate the basic claims made in each post and connect those claims to all of the sources making that claim and all of the sources countering that claim. The algorithm would not pass judgement on the reputation of each source, but rather merely show viewers of that post the spectrum of sources the claim appears within.
Putting this all together, the combination of requiring all social media posts to cite sources for all of their claims with an AI system that reads each post, converts its prose into codified claims and compares those against the cited sources and the rest of the web to place it into context would offer the intriguing possibility of “reverse fact-checking.” Translating the informal world of social media into the formalized world of academic literature in which every claim must be externally supported through a citation to the literature, could go a long way toward upending the fact checking process, placing the burden of verification and truth on posters rather than fact checkers.
Perhaps the solution to our growing problem of digital falsehoods lies in thinking outside the box and making social media more like other fields of knowledge like academia.
In the end, it is clear that new approaches are needed as the volume of digital falsehoods continues to grow unabated.
* Kalev Leetaru is a senior fellow at the George Washington University Center for Cyber & Homeland Security. This article was first published in Forbes magazine.