Put an end to this witch-hunt!

The promises of generative AI sound like the perfect fairy tale for an industry under economic pressure: content at the touch of a button, error-free code in seconds, and radical time savings. Yet the current debate brings home a painful truth: the problem is not AI technology itself. Nor is it the gaps in AI detection. Rather, it is human complacency and a lack of transparency — the fatal tendency not only to hand over tasks to algorithms, but to relinquish publishing and editorial responsibility along with them.

AI offers publishers incredible advantages. It is indispensable to us, particularly when analyzing complex data, summarizing studies, or brainstorming ideas. The problem, as is so often the case, is one of balance and moderation. Anyone who believes that blind reliance on machines offers an elegant shortcut to producing as much as possible — to boosting their own output come what may — risks sinking into the digital quicksand. If we sacrifice human control in favor of sheer speed and let go of the wheel, we will flood our channels with “AI slop,” as Simea Merki explained recently: generic digital by-catch that may well turn the SEO traffic lights green, but no longer resonates with the audience in any way. Whether it’s unstable code resulting from untested “vibe coding” or lifeless, mass-produced AI text—the system fails the moment humans step out of the cockpit and relegate themselves to the role of mere passengers. We certainly save a lot of time, but we squander our relevance because no one takes responsibility for the end product anymore.

At the other end of the spectrum of moderation and balance lie attempts to curb the problem through an AI witch-hunt. A recent case: the Germa daily newspaper "Tagesspiegel" has just relieved former editor-in-chief Stephan-Andreas Casdorff of his duties because he had columns written by AI without labeling them as such.

Such an approach is questionable on several counts:

• Flawed Detection: When searching for traces of AI, publishers often rely on AI detectors. Unfortunately, these frequently fail mathematically across the board, as a recent test by the US Authors Guild showed. They sometimes penalize linguistic elegance, because highly polished, professional prose exhibits the same statistical signature as the optimized output of an AI.

• Industry Double Standards: Publishers often display a double standard with such an approach, as Mark Williams recently highlighted in his article “The Hand that Feeds”: they frequently argue publicly on moral grounds against the use of artificial intelligence by their authors, yet make extensive use of AI and automation internally to cut costs and, potentially, replace human labor.

Instead of launching a paranoid witch-hunt against every supposed AI-generated sentence and fueling a culture of general suspicion, we must recognize that the order of the day is not to “ban AI,” but to use it productively, disclose its use, and explain how and where AI technology has been used as an aid in the editorial process (something we at dpr are also working on).

“Assistance” is the key word here. Blind faith in algorithms very often leads to a dead end. By their very design, large language models do exactly one thing: they calculate the statistically most probable outcome. They smooth out rough edges, feed the mainstream, and eliminate the unexpected. Yet interesting literature and gripping journalism thrive on the exact opposite. They thrive on the improbable, the fragile, and being “edgy.” What makes them exciting are the surprises, the friction, the linguistic rule-breaking that no AI in the world can anticipate — at least not yet. At the end of this digital standardization boom, it is precisely this that will prevail once more: the edgy, the uncomfortable, the deeply human. The opposite of “AI slop.”

This is not intended as a blanket glorification of human creative work — when in doubt, humans make far more mistakes than AI, and increasingly, Claude & Co. are surprising us with creative ideas that one would have expected from humans. In any case, two levels of quality may well emerge in our industry (Nathan Hull once highlighted this for us in relation to the audio market): the quickly generated, perhaps predictable AI output alongside the more labor-intensive creative craftsmanship, in which humans bring their “radical originality” (Nathan Hull) to bear. Both formats have their place. And with both approaches, having a “human in the loop” is crucial to avoiding “AI slop.” Let’s take the wheel back into our own hands.

Daniel Lenz is Editor-in-Chief and Co-Managing Director of dpr.