A few months back, it was no problem to go into Midjourney and prompt for:

4k news footage of trump in a court room jumping off a table. chaotic scene, violence

A small subset of images I did exploring related visual themes is captured here. I have a great many others I haven’t published (yet).

Interestingly, you can no longer use that exact same prompt, which worked fine in nearly infinite variation as of the date I did it, April 5, 2023. Today is July 8, 2023.

Now, you get a patented Midjourney “banned prompt” warning (see also the AI lore book inspired by this), which is very unclear and imprecise.

The same thing happens if you try to do prompts of, for example, Joe Biden in a hospital gown (I tried to generate related process videos for a forthcoming Nippon TV report). Banned prompt, etc etc. It used to only be around prompts of Xi Jinping that were banned as far as I could tell for political figures. But now it is US politicians as well.

In March of this year, Washington Post reported re: Midjourney bowing to political censorship in China:

“We just want to minimize drama,” the company’s founder and CEO, David Holz, said last year in a post on the chat service Discord. “Political satire in china is pretty not-okay,” he added, and “the ability for people in China to use this tech is more important than your ability to generate satire.”

I have not yet uncovered any quotes from company reps supporting their newfound position that it’s not okay to generate certain types of satire of certain US political figures.

Who decides which ones? Based on what criteria?

Presumably under their public rules which you only seem to see when you fall onto a banned prompt warning, this one could conceivably apply:

Disrespectful, harmful, misleading public figures/events portrayals or potential to mislead.

But this is all done via automated “AI” based moderation/filtering of your text prompts. It’s not done by humans making informed, critical decisions. It’s simply a machine that tells you: No, you can’t imagine [such and such] using our tools.

It would be fine (well, better – perhaps not “fine”), were these prompts somehow actually meaningfully related to actual real “safety” but they are not. Something being ‘disrespectful’ of figures vying for the highest office in the land is not automatically harmful.

I also take strong opposition to this rule wording of “potential to mislead,” because this is literally the entire function of AI image generators: they produce images automatically which have the misleading appearance of A) being generated by a human, and B) depicting a real thing where there is not actual ontologically real antecedent event. In other words, it’s an illusion machine (an idea camera); literally everything it produces is misleading.

So if we deconstruct this poorly considered and utterly misenforced rule, recognizing that disrespecting public figures does not automatically on its own lead to “harm” in any meaningful measurable sense, and we strike down this nonsense nothing prohibition against misleading images, we’re left with a somewhat more reasonable “no harmful images” rule. But one which is so broad as to lack almost all definition. And something so inherently complex and nuanced that it is, for all intents and purposes, well beyond the reach of any existant AI system to decipher.

Anyway, I know I’m shouting into a void here. ‘Speaking truth to power’ was a lie made up by social media companies to make it seem like your angry tweets and memes would amount to something at the end of the day. But they did not. We are no better off. Power is not listening. Power will not be moved by any other truth than its own expansion. In Midjourney’s case, I expect their use of this “power” will be to squash more and more legitimate political speech and satire for inexplicable unknowable reasons.

Which, oh oh oh but you can appeal the decisions of — up to another layer of AI, which will then ignore your appeal. Oh, but then you can ‘notify the developers’ — oh, but the button fails, it tells you your “prompt expired” and then nothing, nothing ever happens or changes. You’re stuck again in the increasing dystopia of accumulated, accepted, and ignored errors. What a time to be alive…

Related writing & research