1. The only viable solution:

“Find trustworthy image sources” via repeated trust tests.

No other solution will work (see box below).

Once viewers know to look for that little “TTG” label, they have a handy tool for assessing the image provider’s trustworthiness.

TTG also makes things easy for image providers who care about their credibility: a discreet “TTG” can be tucked into any image credit almost as subtly as a copyright © symbol.

Proposed solutions that fall short:

2. Visual detection is not a realistic solution to the AI deluge.

In tests offered to the public, visual detection rates higher than 90 percent are uncommon — meaning that for every 10 million images incorporating AI-generated content, a full 1 million would go undetected by even the best (human) detectors.

3. AI detectors often do not detect AI.

The detection rate is likely to go down even more as AI evolves.

4. Laws requiring disclosure of AI use would be unenforceable.

It’s ’way too late anyway: there already are billions of unlabeled AI-infused images out there.

5. Voluntary precautions by AI companies are a non-starter.

(There are as of yet no successful examples to point to.)