Summarize Group Discussion · #SGD1-017

Labelling Altered Images

The discussion

Nadia: Every image created or heavily altered by software should carry a visible label by law. Last month a fabricated flood photo of our town was shared thousands of times before anyone questioned it.

Jonas: Visible labels sound simple, but almost every photo is altered now. If a phone's camera brightens faces automatically, does that need a label too? The category collapses as soon as you try to define it.

Thandiwe: I'd approach it from the other direction. Instead of labelling the fakes, verify the real ones with an invisible signature from the camera, so news outlets can prove an image is unedited.

Nadia: But that leaves ordinary people with no clue, Thandiwe. Most of us don't check signatures, we just scroll.

Jonas: Thandiwe's idea actually answers my definition problem, though. You don't need to decide what counts as altered; you only need to show which images are provably original.

Thandiwe: And platforms could display a small tick for verified images automatically, so Nadia's scrolling user sees it without doing anything.

Nadia: A tick for verified images shown by default is something I could support, as long as the absence of a tick is explained clearly. I'd still want labels for fully synthetic pictures, but I accept the line is blurrier than I thought.

Model summary

The speakers discussed how to deal with software-generated or altered images. Nadia wanted a legal requirement for visible labels on all altered images, citing a fake flood photo that spread widely in her town. Jonas objected that nearly every photo is now edited, so the category is impossible to define. Thandiwe proposed the reverse approach: authenticating genuine images with an invisible camera signature. Jonas agreed this solved his definition problem, and Thandiwe added that platforms could show a verification tick automatically. Nadia accepted a default tick for verified images, while still wanting labels for fully synthetic pictures.
Practise with AI scoringSummarize Group Discussion: how it is marked