It’s 2017, and everyone iswrestling with fake news. Given the wide-ranging nature of artificialintelligence (AI) use cases, it is only natural to ask whether machine learningcan help us combat fake news. It should be fairly simple to build ourselves afact-checker, right? After all, software is able to flag spam email orpornographic images with a high degree of accuracy. It turns out that machinelearning techniques are useful, but they have their limitations—and we are faraway from a fully automated fact-checking tool.

Machine learningtechniques—whether they are supervised, unsupervised, or deep-learningmethods—are pattern detectors. You provide a large set of labeled input andoutput data and train the algorithms to grasp the underlying patterns. Thisworks excellently for narrowly defined tasks in which the corpus of knowledgeis well-defined.

A reference dataset for detectingfalsehoods does not exist. Let us say you intend to use the entire web as yourcorpus for automated fact-checks. If you take such an approach, you can onlycheck whether such a claim has been previously made or not. But this does nottell you much about the veracity. Furthermore, any such system won’t be able tohandle breaking news.

Alternatively, instead of usingthe entire web, you can create your own dataset. Such a data source has toconsist of both real and fake news items to train the fact-checking algorithm.But in reality, the truth is often contested, particularly in the politicalrealm. Any biases, or ideology, inherent to the sources involved gettransferred to this dataset.

Another issue pertains tofact-checking at the news-article level. Let’s say each article in turnconsists of more granular claims. You have to extract and evaluate each ofthese claims individually. But as students of dialectic know, it is possiblefor some of the individual claims to be incorrect, but the overall conclusioncan be true.

Another significant challenge isthe ability to transfer our intuitive sense of context to the algorithms. Mosthumans understand satire, humor, exaggeration, and other rhetorical devices andhave a shared understanding of reality—that The New Yorker’s Borowitz Report orThe Onion’s articles are false, but not false in the sense of fake news. It’snot possible for us to codify all of this context and our communicationnuances. In short, when we try to develop an overarching fact-checker, wequickly realize that our current machine learning tools are excellentpattern-matchers and pattern-detectors, but not full-fledged reasoningmachines.

So is there any way that we canemploy machine learning in the fight against fake news? It’s not my contentionthat automated fact-checkers are useless. But given the particular nature offake news, we should strive for domain-specific tools or narrowly targetedtools—for instance, those that identify whether the headline matches thearticle content or flag seemingly dubious articles for further inspection byhuman fact-checkers. Ultimately, a fight against fake news is not a problemthat can be solved by technology alone. We have to bring in machine learningtools, but human and institutional resources are more important.

Facebook’s response to calls forcurbing fake news dissemination via its platform illustrates a multi-prongedapproach. Users now have the ability to flag content as fake news, and based ona threshold level of flagging, such articles are sent for review byprofessional, non-partisan fact-checkers. If expert human reviewers deem anarticle to be false, it is labeled as such, and readers are alerted. Facebook’smachine learning algorithms down-rank dubious articles, and they don’t appearprominently in the newsfeed, thus reducing their reach. Here, you can see thata human-plus-machine approach is being leveraged, rather than the brute forceof the machines themselves.

Fact-checking (so far) involvesuniquely human faculties of reasoning and critical thinking. Automatedfact-checkers are useful as a first line of defense, identifying dubiouscontent for further inspection by humans and increasing their efficiency. Thus,in the fact-checking domain, humans lead, and machine learning algorithms playthe supporting role.