If you listen to talks and panels about AI at media conferences, you’re sure to meet an old acquaintance: the robot journalist. It’s the simple idea that the future of journalism is pure and efficient automation. As exciting as it might sound to some, it’s a rather bleak vision of the future, with bustling newsrooms replaced by rows and rows of branded server racks.

It’s the same story Google and IBM told after the historic showdowns between human players and computers in the games of chess and Go. Both games were often seen as the next big hurdle of artificial intelligence. It’s, unfortunately, the framing most journalists willingly accepted, because it’s a story we love to tell: Man vs. machine. Creator vs. creation.

But hidden beneath those sagas of algorithmic triumph over human intelligence, you can find a much more interesting story. During the training of AlphaGo, Google decided to pair up humans and machines and pitch those teams against each other. Both human players — now supported by an electronic partner — played much better, faster, and more precise than before. The same thing happened after Garry Kasparov lost his legendary game of chess against IBM’s Deep Blue in 1997. Kasparov subsequently invented a completely new way of playing chess, also by teaming up human players with computers. The result is called centaur chess and regarded as being faster, more precise, and more accessible to amateurs, than normal chess.

There’s a lesson to be learned here and it’s not about automation, but augmentation. Teaming up with AI, to become better, stronger, faster. To become a centaur.

So what would your centaur future look like as a journalist? Around the globe, we can already see dozens of examples. Let’s grab the low-hanging fruit: You don’t want to spend hours on end transcribing an interview? Yeah, let’s give that to a machine learning system (though you still might want to take some time to edit the final transcript). How about using satellite images to find the potentially illegal swapping of cargo between ships on the global oceans? You could spend days looking for yourself, but your machine-learning colleague is able to do that for you. Or how about finding illegal amber mines in the deep forests of Ukraine? Sure! Helping you to recognize which of the 500-plus representatives and senators is standing in front of you? We got that! Deep diving into millions of leaked documents from the Panama and Paradise Papers? Grab a coffee and let’s start digging up shell companies. Interested in knowing which topics politicians seem to care about the most? Speak no more!

This is the story I’d love to talk more about in 2019. What tools can we create to help journalists? Especially in an age where the flood of information is often overwhelming even for professionals.

Especially considering that the current incarnation of AI as machine learning has its own set of limitations, as scholars, like Gary Marcus, Zachary Lipton, Francis Chollet and others have noted in recent months. The problem is that ML-systems don’t really understand the data they’re processing. For example, a system trained on recognizing dogs won’t be able to tell you that the picture of a squirrel you fed it is not a dog. It might be certain it’s a German shepherd.

These systems are also unable to understand context and information outside of its training. They can’t reason or use abstract knowledge the way humans can. But the even bigger problem is the inability of machine learning to distinguish between correlation and causation of the patterns found in the data.

The idea of the robot journalist betrays a flawed view of the purpose of journalism. It’s the algorithmic equivalence of the “view from nowhere.” The idea that reporters have to “report the facts,” without helping readers contextualizing. The biggest flaw of the robot journalist is the idea that there’s such a thing as an objective worldview, able to be gauged through data and algorithms alone — instead of a messy and complex multi-polar world which has to be carefully explored and questioned. That’s something no machine learning system will be able to do in the near future, if ever. And of course there’s a huge difference between learning the patterns of a well-written article and actually writing one.

Don’t get me wrong — automatic text generation will have its place in the newsroom of the future, but it won’t be the single defining use of AI. We will instead see more cases of AI helping journalists, not making them redundant. So let’s stop talking about the robot journalist in 2019 and start talking about the other future for journalism, the one with four metaphorical hooves: the centaur-journalist.