Is the Generative AI Photoshop Beta a dangerous harbinger of a new frontier for the Thought Police?

I made a very rare exception to my policy of not doing free beta testing for billion dollar companies because the new AI features in Photoshop are really just that cool, plus they let me keep “regular” Photoshop alongside it in case of trouble.

There are a number of interesting features, but the most potentially interesting and useful is the “Generative Fill” feature that essentially takes image prompts and layers them straight into whatever you’re working on. Like Firefly, it looks like they trained a Stable Diffusion Model on their stock library and probably some other assets and just built a UI for it inside their apps.

There’s more than enough press about this right now that you can easily find if you’re interested in seeing the feature in action so I’ll skip that part.

The thing that got my attention was the terms of use and restrictions put on the tool.

You have to agree to specific terms in order to use the tool at all, and they’re above and beyond the regular Photoshop TOU.

Chief among them: No pornography or even nudity, and no copyright violations.

I tried to get it to violate copyright with generative AI but didn’t have any luck. One thing I will say Adobe is doing well is ensuring copyrighted material simply doesn’t make it into their training models. They are among the few companies on earth with a digital asset library large enough on its own to pull that off.

If I tell it to simply generate “Pornography” it gives me a warning and refuses to complete the task.

Both pornography and copyright violation are really matters of opinion, not objective fact. No less than the AP backed down and settled a lawsuit in what would have been a landmark case regarding the transformation clause in Fair Use had it gone to ruling. Why? One assumes because they weren’t sure they could win. (Fairey et al v. The Associated Press)

I’m also still allowed to create my own pornographic images and edit them in Photoshop (for now, I guess).

The idea that the very tools used to create art can be hardwired to simply not allow certain content to even be created is pretty disturbing.

Orwell and Bradbury envisioned futures where “banned” content was destroyed, but not even they considered the idea that such content could be cut off even before it gets created.

Given the 1984-like policies being enacted by Republicans in places like Texas and Florida right now, this discovery is more than a little unsettling.

I hope Adobe’s approach is not the norm, or the next logical step is that we’ll start having words and phrases banned from Office 365 and Gmail.