It's not a secret that we're surrounded by digital information across social media every moment of every day, and it's often permeated by fake news and misinformation. With the election around the corner in the United States, misinformation's sheer volume and velocity are expected to rise to new heights12. The dangers of fake news and misinformation are quite obvious. Put simply, it is a threat to democratic processes around the world and can actively skew public opinion and even open the gateways to violence34. With the rise of AI, of course, there are numerous benefits, but one of the fears leaders have worldwide is that AI directly impacts the sheer volume of misinformation, making it easier than ever to generate lies and spread them across the globe in minutes5. This is where NLP (Natural Language Processing) is becoming extremely important and widely used14.

Before we jump into the specifics of NLP, it's important to understand how complex the game of AI-driven misinformation can be. It often cleverly mixes both truth and lies, which makes it extremely hard for AI-driven algorithms to catch1. The second factor that makes this problem hard to solve is that the news cycle never ends—it's constantly evolving, making it extremely difficult to keep up with the scourge of misinformation2. And then there are people—who see a fake story and then share it across social media as truth. Some stories are so salacious that they spread like wildfire, and before you know it, a lie becomes a part of society's fabric3. For example, a recent fake news story that accused Kamala Harris' campaign a part of a plot to kill a whistleblower that supposedly reported that ABC gave Harris' debate questions prior to the Harris vs. Trump debate. The fake news saw a prominent member of Congress share the story on social media, which furthered the completely untrue story. While the member of Congress came forward to admit the story wasn't true, and she wrongly shared it, there are many people still sharing the story4.

AI and NLP innovators are working day and night to solve the issue of political misinformation and fake news by enabling machines to further understand and interpret the content that is being created and shared across social media. Then, with AI-powered algorithms, scan and analyze the content being created by humans12. NLP works to enable this ability so that an AI platform can better understand the context of a piece of content, the sentiment behind it, and even the intent of the user1. This includes looking to identify patterns of misinformation, the use of inflammatory and sensationalist language, false claims, and even the use of keywords and phrases that are found across the spectrum of fake news. Social media and AI platforms are also working to create fact-checking processes to assist, which has led to quite a bit of discussion over what a fact is. What one person may consider a fact may not be what another does, which serves up another set of challenges that tech innovators, and society at large, must weave through3. NLP models are capable of extracting claims from posts and then comparing them against data sources that are considered reliable. This means that there must be a set of verified facts intact for the machines to pull from. For example, Google's Fact-Check Tools are already doing this, highlighting fact-checked claims in search results4. Furthermore, AI is increasingly being able to detect coordinated efforts to spread false information and fake news via posting behavior. NLP can root out bots and fake accounts by looking closely at the account's use of language, how quickly and often they respond, and the patterns they use to engage on a platform15. For example, Facebook is using AI to remove vast networks of accounts that Meta deems inauthentic and working to manipulate its users into believing a false narrative. Additionally, Meta has identified instances of Russia and China using the platform for its interests to influence its members6.

There are a growing number of AI-driven platforms that are doing quite a good job of identifying fake news, including Snopes, Factmata, and FAKEBOX. These platforms work to evaluate news articles and the claims that are being made daily12. In the case of FAKEBOX, it was developed at the University of Michigan and employs a mix of deep learning and NLP to analyze the steady flow of news articles being pushed into the world and assign credibility scores3. So far, it has done quite well, especially where political discourse and health-related misinformation are concerned4.

The first and most glaring limitation is ambiguity and context. The human language is very complex and nuanced—we're talking sarcasm, the idioms we use, satire, and so much more. Data scientists are well aware that misclassifying legitimate content as false can and will infringe on free speech.1 A perfect example surrounds the past election in the United States. Some people feel that they have done the research necessary to claim that the election was stolen from Trump, while others feel it is quite clear that the democratic process was impartial and fair. Who owns this truth in this scenario? Should someone have the freedom to claim that they have the cornerstone of truth and declare it on social media even though it defies recorded history?2 Another challenge surrounds the use of memes and coded language. How does AI keep up with it all? So far, the answer from leading AI gurus is that constant updates and retraining will always be a fact of life for every social media and AI platform. There are also biases to consider within AI models. Let's say Elon Musk and his affinity for Trump may lead him to skew his AI model to fit the narrative of Trump, versus keeping it fair and impartial. The same can be said of Mark Zuckerberg and Meta. How does the public keep these tech leaders from targeting certain groups and perspectives they deem wrong? And of course, there are always going to be malicious actors out there who create content that is specifically designed to avoid AI detection.3

As much as the world of AI would love to get away from humans having to play a huge role in moderation, it doesn't appear to be a good idea. It has been proposed that the best model is for AI to handle the initial filtering and identifying suspicious content. Then, a human being can look at the content deeply and use their judgment on whether a piece of content should be removed12. These fact-checkers can do what AI fails is incapable of—providing contextual insights to decide whether a piece of content can be 100% trusted3.

Political misinformation is evolving as we speak—so too does ongoing research and testing in AI and NLP. The prospects of these technologies combating the issues are promising, but we can't expect for this issue to be solved today, tomorrow, or in a year.12 A growing number of leaders and thinkers around the world are calling for ethical standards and oversight to be put into place as soon as possible. This means sustained collaboration between AI leaders, governments, and society at large to build a framework for protecting the integrity of information and the rights of those who use the technology.34 AI and NLP are indispensable in our time and age. If done right, they will help foster transparency and collective responsibility while scaling to effectively monitor and deal with the spread of fake information.56