AI in journalism and the lessons of the early internet

There are lessons to be taken from newsrooms’ experiences of the early days of the internet - and following the arrival of ChatGPT a year ago, they’re finding a better balance between investing in AI and exercising caution. Where is the sweet spot between the two?

“No publisher that I’ve spoken with is in any doubt that they need to start using this new technology, and get on the AI train as fast as possible”, says media investor Saša Vučinić. “Their hesitations are limited only to how to do it safely, responsibly, and without violating anyone else's rights.”

It's been a year since Sam Altman, the recently ousted CEO of OpenAI, demonstrated the capabilities of the GPT 3.5 language model. He recently launched GPT 4.0 Turbo, featuring updates that have technologists worldwide eagerly developing new applications. In addition to the famous chatbot, media creators are also embracing AI by using tools to have articles read out loud by a robot, generate images that complement their stories, or translate entire pieces for those who couldn't access the original due to language barriers.

However, Erik van Heeswijk, CEO of smartocto, feels that the real value of artificial intelligence is yet to fully emerge. “Most tools launched in the past year have essentially been ChatGPT in a different guise. It's just a shell: the tool activates ChatGPT or another Large Language Model with a specific prompt and returns the output to the user.”

What’s AI’s real potential?

When you ask ChatGPT for better text, the output comes from a black box. What we need to think critically about is what ‘better’ actually means.

In the context of the newsroom, the answer lies with the evolution of content analytics. For example, Van Heeswijk’s Dutch-Serbian startup, smartocto, has evolved to perform tasks of a virtual assistant and go beyond calculating pageviews to be able to suggest headlines, identify where readers lose interest - as well as propose text improvements. As a result it can deliver strategic and practical advice to all roles within the newsroom.

AI in newsrooms can be transformative - but only when the tasks, goals and strategies of those newsrooms set the agenda.

When smartocto.ai launched a month ago, this point was made. “We possess a comprehensive understanding of our clients' goals and actively manage their data,” Van Heeswijk says. “Our algorithm plays a pivotal role in this process, enabling us to accurately estimate how clients can optimise their stories. It is from this informed position that artificial intelligence can truly exhibit its transformative potential.”

Where to invest?

The question remains whether media companies are ready to separate the wheat from the chaff. While there are plenty of experiments, success ultimately depends on securing investment, as Vučinić knows from experience. As managing partner at North Base Media, he fosters media development in Southeast Asia, India, the Middle-East and South America and also invests in Western tech companies, including smartocto.

“Every time the media industry undergoes a dramatic technology shift, publishers are tempted to start developing new technology in-house,” he observes. “Remember when every publisher tried to build a proprietary paywall? Then, a few years later, they were forced to admit that tech startups were better positioned to develop, maintain, and affordably build new technology faster. I see some signs that history may repeat itself once more.”

He believes the best approach now is to form partnerships with already proven technology startups, securing a role in shaping future products.

What are the lessons from the introduction of the internet?

Erik van Heeswijk also suggests something else: that publishers should first consider their own relationship to artificial intelligence. “Following the introduction of the internet, there were those in the media, especially newspapers, that ended up having to catch up on ten or fifteen years of developments because they thought it wouldn’t affect their business model. And others just poured millions of dollars into megalomaniac projects. Fortunately, we have these lessons in our backpack. The time to answer the ‘how’ question is right now.”

What might the answer look like? Van Heeswijk thinks back to the early days of the internet, and the lessons learned. “Define where you want AI to assist editorial teams and which tasks it might eventually take over. This also means freeing up people to focus on this, giving them a mandate and a budget. Ensure every department is represented. Journalism is too valuable to be left solely to technicians. This means setting up a meaningful innovation agenda. That means of course to experiment, trying out different systems and tools, but there has to be internal communication and focus on the outcome. If there is something that the early days of the internet has taught us, it is that you need to take these groundbreaking innovations very seriously, but you shouldn’t go overboard while the tech is still developing. It is a means, not a goal.”

And finally: keep an eye on copyright, privacy, and security. “Almost all publishers I talk to are adamant that they don’t want their data to train the model. We’ve considered this in the architecture of smartocto.ai. We can also switch to different Large Language Models if necessary. It’s unwise for anyone to rely entirely on, for example, OpenAI’s model. Be agile. Never bet on just one LLM.”

This story was first published on INMA.org