Technological change tends to happen slowly at first, and then all at once, like an S-curve. This certainly feels true for AI, where so much has happened just in the last year.

Today, the development of AI (what can be done) has already outrun legal and ethical considerations (what should be done).

For example, media companies could provide summarised and personalised news with no editorial oversight, but that would come with the risk of generating fake news and no way of catching it. In our experiments, one in 10 machine-generated summaries contain “hallucinations,” or stuff it just made up. Significant effort remains on both a technological and ethical level, and it’s hard work.

Thankfully, many things can be done that are neither technically nor ethically complicated, which we most certainly should just be doing. We’d like to share how we’re working to find these things at Schibsted News Media and how others can do it, too.

AI is pretty different from the technology we’re used to working with in media organisations, so the first and crucial step is simply to understand it better. At the end of the day, AI is just a tool. And as with any other tool, you’re not going to be able to use it to solve much of anything if you don’t know how to wield it.

In the long run, I think we need to build media organisations where working with data to fine-tune or even train machine learning models is about as common as writing code. There’s a lot to do to get there, but just using models trained by somebody else is a great way to get started.

For us, it was great just to take a day to find and play around with all the freely available AI models that exist nowadays. Just to mention a few, OpenAI’s models are really powerful and really easy to use, and Midjourney is even more so. If you want to get a little more hardcore, there are open-source models for pretty much anything on Hugging Face.

Our goal was simply to learn and understand what’s possible. Our hypothesis was that, when we do, AI will be just one more tool we can use to solve the problems we encounter on a daily basis.

What happened was even better: When we shared what we’d learned with the rest of the organisation on the very next day, there was a literal line of journalists waiting for us afterwards. They told us that what we’d learned about automatically transcribing interviews from speech to text using OpenAI Whisper could change their lives.

Three weeks later, we’d built a Web site and an app to make it easy to do. And after just two months, it has saved journalists at Schibsted more than 3,000 hours of painstaking work. All it took was knowing that journalists had to spend all those hours transcribing manually, and that AI is finally good enough to do it automatically.

That was awesome, and it got us thinking: What other problems don’t we know about yet that we can we solve with our newfound understanding of AI?

To find out, we dedicated another day. This time, we invited people from the newsroom, too. Instead of learning about solutions, our goal was to learn about problems we could apply them to.

It was pretty ambitious. But our thinking was that, even if we just get to know each other better across functions, that’ll be worth the day. We need to be working way more cross-functionally and together going forward.

If we also spread more knowledge about our journalistic processes and what’s now possible with AI, then it’s more likely we’ll find ways of doing it even better in the future. And if we even manage to find some ways that we can do it right now, then we should do this every day!

We found 37 more things that AI can help us with right now.

That’s 37 more things that we think would be useful and which we’ve prototyped and verified that the technology is already good enough to do. This is 37 more of those things we both can and should just do. We’ve gotten started on the next most exciting part, and we aim to hope to tell you more about it soon!

We hope sharing these simple tricks we’ve done to start using AI today could be useful to you in doing the same. If you’d like to brainstorm how exactly you might conduct some days like we did, please contact me and we can learn from each other.

Finally, we hope others also consider sharing what can and should be done with AI, and how it’s working for you, so we can help each other get it right. A lot rides on it!

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .