Using AI to fight misinformation
In his latest newsletter, Bruce Schneier springboarded off of articles in Washington Post and The Atlantic to write:
There’s a balance between the cost of the thing, and the cost to destroy the thing, and that balance is changing dramatically. This isn’t new, of course. Here’s an article from last year about the cost of drones versus the cost of top-of-the-line fighter jets. If $35K in drones (117 drones times an estimated $300 per drone) can destroy $7B in Russian bombers and other long-range aircraft, why would anyone build more of those planes? And we can have this discussion about ships, or tanks, or pretty much every other military vehicle. And then we can add in drone-coordinating technologies like swarming.
Fighter jets, ships, tanks, … and information. It's common knowledge in journalism that if it takes X amount of time to come up with misinformation and Y amount of time to debunk it, Y will always be greater than X. In other words, misinformation takes less time (and likely effort) to produce than legitimate information. Network modelling exercises have also found repeatedly that false information travels faster. Taken together, the cost asymmetry experts are beginning to perceive between a fighter jet and the means to destroy it has been around for a long time vis-à-vis information, and in fact the only reason the 'information side' hasn't lost the war, such as it is, is that there exists in the population a certain (but admittedly diminishing) level of awareness that it's possible to manipulate people into echo chambers as well as to look past the chamber wall to find a whole different reality.
Generative AI has of course added considerably to this problem but as a tool it isn't limited to producing noisy or bad information — that propensity comes from the humans in the loop. I think if we're to keep our heads above the water, it's important for journalists to recruit gen AI to the task of rebutting misinformation then and there rather than wait for journalists to manually pieces articles together. Articles of the latter variety are capable of important change when done right but they take time. When a former ISRO chairman says Sanskrit is a language suited for computer science, a coherent and complete rebuttal that's also clearly written will need at least two or three hours to come together. At least. This process can be accelerated by a journalist in the loop cobbling a rebuttal together with, say, ChatGPT o3 (the "advanced reasoning" model), making sure the sources are legitimate and reputable, and finally checking the text (or visual) for inappropriate language — all in minutes.
There are legitimate apprehensions about journalists using AI. For me, personally, using AI-generated text is a moral offence against the act of a person communicating with their community, with human and public interest at heart. There are injustices embedded in the training and operationalisation of generative AI models that no one, journalists or otherwise, should help perpetuate and that everyone should help address and resolve. At the same time, however, the corpus of annotated data that animates these models represents a substantial amount of human-made knowledge that we should be able to draw on — especially without having to be mediated by profit-minded technology companies — to negotiate the contemporary information landscape. Open-source bespoke models in particular could a long way by being free to use and having their information sources (e.g. "just thehindu.com") restricted by default.
