Trust good, crap bad

Generative AI makes it easier than ever to crank out commoditized content. Local newsrooms are understaffed, fighting to keep their heads above water, and often owned by private equity firms who worship volume and scale.

What could possibly go wrong?

I see two short-term futures for generative AI in local news: The first is one where even well-meaning, mission-driven local newsrooms see it as a cheap ticket to growth and efficiency. Driven by shrinking staffs and short-term revenue needs, some will be seduced by vendors peddling magical solutions to the wrong problems: We’ll help you publish more stories faster! We’ll game your SEO by flooding the Internet with commodity content!

Others will convince themselves that having fake authors write fake articles is a strategy that serves their mission. Once-trusted brands will be swept up, along with their credibility, in what Peter Kafka has artfully termed the Tsunami of Crap.

The second future, and the one I’d prefer to live in, is one where we spend real brain calories figuring out how to use these technologies to do better journalism and create more engaging experiences for our readers. In this future, we understand that trust and quality are core to our competitive advantage, and we use AI technologies to help us enhance them.

Sounds obvious, right? Trust good. Crap bad. But it’s also a hard problem, because these technologies are still young and poorly understood. Applying them effectively will require exploration, prototyping, speculative investment, and some amount of risk — all things time- and resource-strapped local newsrooms are not typically well equipped to supply.

Did you know large language models are also ace time-series forecasters? Neither did anyone else, until researchers thought to ask. Technologies like vector databases can help us build better search experiences. Tools like SEC Insights show us what might be possible if we apply techniques like retrieval-augmented generation to investigative reporting. Language models can perform some complex classification tasks as well as humans, which could help us better understand how our journalism resonates with readers. And so on.

The highest and best uses of generative AI in the newsroom will not come to us passively. We need to get our hands dirty — and share what we learn — if only to build mental models for where AI technologies can align with our businesses and our mission. Plus, this stuff is potentially world-changing. We have a journalistic obligation to understand it.

The New York Times has the right idea, hiring a small team to explore and flesh out potential applications of generative AI in its newsroom. Their efforts will be exciting and instructive, and I can’t wait to see what they come up with, but the Times is the Times. Its needs and the needs of local newsrooms are not the same.

In the coming year, my hope and expectation is that we will see local newsrooms follow suit, through a combination of partnership, convening, collaboration, and selective investment. We’ll certainly be working on it at our shop, and I’d love to hear from anyone else who is doing the same.

The alternative is the first future. We ride the crap tsunami at our peril.