• by Richard Whittington , Op-Ed Contributor, October 23, 2024

Democracies depend on an informed electorate.

Exaggeration and dissimulation are nothing new to politics, but Social media has helped accelerate and spread falsehoods. More than half of U.S. adults say they sometimes or often get their news from social media’s vast info-pool (and cesspool). Trust in media is at a low. artificial intelligence threatens to further confuse, misdirect, and distract vast numbers voters. 

What’s a democracy to do? Just as pressingly, what’s a “trusted” media organization that’s no longer quite so trusted to do? This is an urgent question: Given the explosion of AI, misinformation and disinformation may erode public trust past the point of no return.

The truth in real-time

The answers will involve AI. This is an informational arms race, and AI amounts to a laser weapon in what has been a knife fight. One approach that harnesses AI under the flag of factuality – and which can happen with technologies available today – is to integrate what you might call “truth subtitles.”

Whether they’re packaged as actual subtitles or an attractive chyron or a sidebar or simulcast to a mobile device or VR headset doesn’t really matter. What does matter is that as the primary source (or his or her news-making proxy) speaks, a large language model (LLM) taps into a hierarchy of sources – news organizations, government databases, academic papers – and does a rolling fact-check. This could be done during high-profile debates, video interviews (such as on Sunday political talk shows), speeches, press conferences, podcasts, whatever. 

Should a candidate say, for example, that recent inflation has been the worst in the history of the country, generative AI could immediately note that the 8% peak in 2022 was lower than the 13.5% peak in 1980. It might also add that the 16.9% average wage gain since 2021 nearly kept pace with the 19.6% overall increase in prices. And, perhaps, that the current inflation rate of 2.5% isn’t far north of the Federal Reserve’s 2% target and provide that to the viewer in real time.

GenAI that’s trained on reliable sourcing can provide that sort of detail much faster than a TV economics correspondent, and it’s available 24/7 and simultaneously across various feeds. As with all GenAI, its utility and trustworthiness will depend on maximal transparency in its sourcing and training as well as safeguards around ethical and strong data, security, and privacy protections. A complementary system could break fact checks into texts, social media posts, TikTok-style videos, create annotated transcripts, or otherwise repackage its fact checking for later consumption in forms the viewer chooses.

If someone builds it, will they come?

The technical aspects will be the easy part of truth subtitling. The hard part, as usual, has to do with people. It would be optional, so would they bother turning truth subtitles on? Could they get past their own cognitive biases and trust it? Could it be hacked? How would you source it and assess the quality and bias of sourcing used to train the LLM? (Thereareoptions). And, perhaps most importantly, who would pay for it, maintain it, and, ideally, get paid for it?

Media companies should certainly be motivated. Truth subtitling could drive viewership and social traffic – not to mention licensing revenues from smaller organizations who decide to plug it in. Given the obvious public interest, public-funded media such as the BBC might also take the initiative. Google or Meta, for whom such a system could be a possible moneymaker, could also lead development.

GenAI-powered truth subtitling would, I believe, ultimately reduce the need for truth subtitling in the first place. Politicians, leaders, and their proxies will learn to hew close to the truth or be immediately and relentlessly fact-checked in ways today’s media, hollowed out as it has become, just can’t. 

Truth subtitling alone can’t solve the problem of an uninformed electorate. But we can build it today, and our democracy may ultimately depend on someone doing so.