Trust is the hardest metric of all to steer

AI content: to label it, or not? New research has some interesting conclusions.

by Rob Corbidge

Published: 15:54, 07 December 2023

Rob Corbidge is Head of Content Intelligence at Glide Publishing Platform, applying the latest knowledge about advances and ideas in the publishing industry to our own product and helping clients get the most from their content.

Trust is tricky. Tesla's new Cybertruck features a drive-by-wire system, a world first in a production car. There is no mechanical linkage between the steering wheel and the steering system. 

Why it has taken so long to introduce such an "innovation" into cars, when such technology has served in such failure-critical systems as still-in-service combat aircraft since 1976 and commercial aircraft since 1984?

It's partly a question of cost up to this point and also partly a question of trust. 

We're concerned with trust. Any incident where there was implication of a drive-by-wire failure could have potentially destroyed the sales floor prospects for an entire vehicle run and besmirched the brand for many years. We will see if others follow where Tesla has led as consumers transition from driving cars to in fact directing processors that drive the cars. It's of course likely Tesla have introduced the technology with a bigger picture view to their own plans for self-driving cars and other vehicles.

Trust is a massive thing in publishing of course. There's a newsdesk mantra I still use, coming from a Dutch phrase: "Trust arrives on foot but leaves on horseback". It's true. Get it right and you're just getting it right. Get it wrong and all the times you have been right come into question.

A new piece of research from University of Minnesota Professor Benjamin Toff and Oxford University's Felix Simon explores audience trust in AI generated news. The genesis of the research was simple: Toff said he was asked about trust in AI-generated news and "I didn’t know the answer, so we did an experiment!"

For those interested, the paper "Or they could just not use it?”: The Paradox of AI Disclosure for Audience Trust in News is here.

The vital question seems to hinge around whether to actually tell an audience they are consuming AI-generated content and the potential pitfalls if a) you do, and b) you don't. 

As the paper puts it: ""While a growing number of publishers have begun responding to these concerns by adding labels to AI-generated content, there is no shared consensus about what such disclosure should look like nor agreement over what level of AI-involvement should trigger labelling."

The research saw a selected sample group shown a variety of AI-generated news content. Some of the group were just shown the AI-generated material with no disclosure that it was such. Others were shown the same content with disclosure, and others had disclosure along with a list of sources the AI-generated content was drawn from.

Perhaps not surprisingly, it was this second group who rated the news content as least trustworthy. Yet, for accuracy and fairness of the content, both the first and second groups scored the content about the same.

Interestingly, a sub group of those tested had what researchers in the field call Procedural News Knowledge or PNK, defined as "familiarity with what legitimate news production and reporting entails". A good number of GPP subscribers will almost certainly be included in such a group. This PNK group recorded lower levels of trust in the disclosed AI content than any other. 

However, I'd say the PNK group would without a doubt express lower levels of trust in any content. If you know how it's built, you know how to knock it down.

We are at an early stage for the uses of AI-generated content. It's important to assume people aren't stupid. In time, if such content proves to be of use, then people will become accustomed to it and trust will form. The human brain adapts quickly. Think of your own response to the volume of AI-generated images you likely see each week. Most of us have already absorbed the visual aspects that denote such images without thinking about it, and immediately categorise them as such.

Regarding such opinions around trustworthiness, as the researchers point out, any research into news journalism already carries a significant burden in that general trust in news, we are repeatedly told, is at an all-time low. Tell that to the 17th-century political pamphleteers of the English Civil War.