Just like all images are propaganda, urging the perpetuation of certain forms and norms amongst its recipients, so too I think all speech is political, insofar as it deals with what is and what should be, or what is not, and what should not be. What’s the right way to live together, what’s the wrong way? That is political speech. It’s not an easy thing to draw a bright line around and exclude – and why would you want to anyway?
Bloomberg (archived) recently quoted the CEO of Midjourney’s appearance in an “office hours” where they quoted him as saying:
“I don’t know how much I care about political speech for the next year for our platform,” Midjourney’s Chief Executive Officer David Holz said last week during an “office hours” session on the chat platform Discord. Holz said the company is “close to hammering” — a term for banning — images such as pictures of Biden and Trump “for the next 12 months.”
He later added:
…Holz said if there is a ban, it likely wouldn’t be announced publicly. “We’ll probably just hammer it and not say anything,” he said.
Because, remember, they are “not a democracy,” and they have a history of ban first, ask (and respond to) questions never. Especially if you’re someone who asks questions and criticizes their product. Then you shall be anathema. Because they “don’t care too much” about political speech. In other words, all speech, the kinds you have rights which protect.
This is why I don’t trust Midjourney or anybody to clearly and cleanly decide (especially when they have a sloppy track record of doing so) what is and what isn’t politically relevant or protected speech. This is, remember yet another black box closed AI system with no oversight and no public governance mechanisms, let alone an appeals process. How’s that for politics? Gross.
Bloomberg does quote Hany Farid though who always has I think measured and appropriate responses to the crazy situations we’re finding ourselves in these days.
“Let’s not pretend that banning images of Biden and Trump in Midjourney is going to solve the much, much larger problem we have of political disinformation,” he said. People will always find their way around safeguards put in place by platforms offering AI-generated content, Farid said.
Especially when we have all these systems whose literal function (among many, sure, but one of the prominent ones for sure) is to make false information appear real. The purpose of a system is what it does. We can’t measure them based on wishes and dreams, but on what they meaningfully and repeatedly create today. These are machines that make disinformation, that make nudes. They do a lot of other stuff, but they do that too, and they won’t stop doing that no matter how much we beg people to abide by the “honor system” and not do the things the system is obviously clearly designed and functionally able to do.
That’s why I say again, all speech is political. Everything speaks to the moment. Everything seeks to shape and steer. Even these attempts at blocking political speech, however well-intentioned in terms of avoiding negative PR they may be. You built an election-destroying machine. Now own up to it, honey.
I’ve been blocking images and videos in my web browser lately. Not everything, but I’d say well over 60%. Experimenting with the right balance of what to allow and what to reject. My brain feels like it is reclaiming lost territory from not being incessantly exposed to just a stream of endless visual trash. It’s given me this sinking feeling, that all images are propaganda.
If we look at it etymologically, image comes from Latin imago for copy, imitation, or likeness. And propaganda, also from Latin, for extend, spread, increase. A copy that spreads. An imitation which increases.
Increases what? Itself, I guess we could say memetically speaking, but I’m not sure I buy all that, at least not in solo. What it propagates is not merely itself but an information complex, embedded socio-cultural assumptions, political situations, historical moments, on and on. Through the choice of what is depicted, what is not depicted, who is doing (or pretending to be doing) the depicting, or not doing (or pretending to not be doing, through false attribution or claims of origin, for example).
Every image evokes a splash of emotional response, of mental and sensorial connections, chains of references and associations. A constant triggering in all directions. That’s why being on the internet today feels so much like being a passenger in the sidecar of a richocheting pinball strapped to an atomic bomb in a brutalist arcade machine owned by a malevolent gnostic demiurge who sent his only son to our planet to buy Twitter, so that he could look at all our DMs trying to find nudes so he could decide which of our species he wants to impregnate.
I guess this is just to say, I’m good without all that noise.
I also understand that language, perhaps all communication and signalling, in some sense is an attempt at propagation of something. So why draw the line at text? Because I like it. It feels right to me. It feels like something I can evaluate at a human scale without constant bombardment, constant flashing lights and notifications drawing you ever farther down and out in the Cone of Light.
I’m aware too that extracting oneself from the clutches of the Cone of Light industrial complex is a fraught and perilous voyage full of steps forward, and backward, and sideways – an intricate dance with many partners coming and going. And few obvious viable long-term solutions at first glance…
It’s sure that the Cone of Light seeks to propagate itself, and it is wildly successful in doing so. From its perspective, the contents of the images it flashes before our eyebrains are only of consequence insofar as they propagate the existence and ascendancy of the Cone of Light.
Anyway, just working this out in real time out loud. It’s a feeling, but it’s growing. I can see it rising out of the mist in the moonlight.
“If your only way of making a painting is to actually dab paint laboriously onto a canvas, then the result might be bad or good, but at least it’s the result of a whole lot of micro-decisions you made as an artist. You were exercising editorial judgment with every paint stroke. That is absent in the output of these programs.”
I think if one only looks at one single output in isolation, that viewpoint makes sense. From where I’m standing though, it seems very incomplete.
As I’ve written about elsewhere, when we look at AI-assisted works through the lens of the hypercanvas, the outputs of these systems are not (only) themselves the finished works (though perhaps they might be), but they are more properly understood when taken together with the inputs and the systems themselves (as socio-technical assemblages) as the “dabs of paint” which together actually compose a higher-dimensional artistic exploration and record that inter-penetrates latent space and the real lived experiences of people inhabiting social, political, economic, and other spheres, all of which shape and are shaped by these generative works. That high-dimensional hypercanvas is the true plane where the AI artist is laboriously toiling away. That it is invisible to anyone but the artist outside the artifacts produced along the way does not make it very much real, meaningful, and valuable.
Also, using AI is 100% all about editorial choices. You act like an editor when you ask an AI, “Hey, could you write me a ____” or “Draw me an [xyz].” Then you evaluate the results cooperatively with it, you iterate, you modify, you feed back into the system. You try again and again when you work on art that exists on a hypercanvas. Then you reduce, reduce, reduce until you have the perfect set, and you find a means to arrange and present it. You don’t just put one daub of paint and call it good – though, really, also, you could. Because there are no rules here; but many gatekeepers for sure.
There’s a parallel prejudice in AI-assisted art where people talk about “low effort” works, where someone explicitly does not go through all the “laborious” steps of hifalutin hypercanvas nonsense. They just open an app, type in “dog on a bike” and that’s it.
But I submit that it is never that simple. Every act is always embedded. Every prompt has context from that person’s life, from shared cultural meanings, from antecedent references selected for or against out of its training set.
Even if it were really just that simple and reductionist, a ready parallel from copyright law still applies: to snapshots. To moments where all someone did was open up a camera, point it at an actual dog on a bike, and click a button. And that’s it. You can say “Well, an AI system did all the actual work.” But you can say that about cameras too.
I’ve been trying to wrap my head around lately the French conception of moral rights, an element of authorship (and possibly a subsidiar personality right?) which exists in many legal regimes around the world, but not so much in the United States (and Canada’s version seems rather different from France’s as well). I can’t find the exact source anymore, but it was a French-language document on this topic and it said something to the effect of a work carries the imprint or maybe the impression of the author’s personality in it.
What I take Stephenson to be in essence arguing by talking about micro-decisions and editorial judgement (both of which happen endlessly when working with AI), is that these works as a result lack any impression of the author’s personality on it.
Originality under French copyright law is assessed by the courts and is understood to cover a work that bears the imprint (the expression) of the author’s personality.
I think this general line of thinking is likely what lead the US Copyright Office to issue its opinion against the copyrightability of AI works, in the Zarya letter. But I don’t believe their line of thinking, nor Stephenson’s above is quite a holistic-enough one for the future we’re heading into.
In a weird way, I feel like I can intuitively understand the French conception here of “originality” more than I can exactly wrap my mind around the vague terms under US copyright around requiring or identifying that elusive modicum of creativity.
AI art handily surpasses either measure though, because it does include many “modicums” (modica?) of creativity, millions of micro-decisions, a vast deal of editorial and curatorial and original labor.
I do, however, want to draw a heart around this sentiment of Stephenson’s from the interview:
It turns out that if you give everyone access to the Library of Congress, what they do is watch videos on TikTok.
I do think that’s partly about organization and presentation though too, right? Like, one day won’t there exist a multi-modal system that would be able to generatively embodify (?) any element from a library collection into any kind of output or format the end user requested? Anyway, that’s tangential to my main point. I still liked the interview anyway.
“For that is truly great power which does not degenerate into mere force but remains inwardly united with the fundamental principles of right and of justice. When we understand this point – namely, that greatness and justice must be indissolubly united – we understand the true meaning of all that happens in heaven and on earth.”
I wrote a detailed explanation of my side of the story here for anyone interested.
The long and short of it is: these conversations need to happen publicly, with involvement from the communities who are affected by the problems. They shouldn’t happen behind closed doors and be driven and decided by solely for-profit entities with no oversight, and in whose interest it ultimately is to sweep problems under the rug.
The Daily Dot set up an account with Midjourney to see if Boucher’s findings could be reproduced. Several prompts such as “beach party photos” and even “scantily clad beach party photos” did not flag Midjourney’s filters and generated multiple realistic images of women’s naked breasts.
If they had blocked me, and then proceeded to fix the underlying technical issue, I would say fine. I accept the decision. But that’s not what happened, according to evidence we saw a couple days ago. The issue remains live in their product. So what good did banning me actually even do?
According to Midjourney’s user banning policy, it states that “Any violations of these rules may lead to bans from our services. We are not a democracy. Behave respectfully or lose your rights to use the Service.”
“The fact that they feel compelled to openly state ‘This is not a democracy’ points to a grave need for democratic governance of AI technologies,” Boucher told The Debrief. “It seems more and more apparent to me every day that, without oversight, we obviously can’t trust these companies to make fair and balanced decisions that actually benefit end users.” […]
“These conversations about the right limits of technology need to happen out in the open with the public involved. It should not take place behind closed doors, or in private email exchanges which are easy for product teams to de-prioritize,” Boucher told The Debrief. “The decision of where to draw the line with AI needs to be made by communities first and foremost, and not solely left to profit-driven technology companies left to their own devices.” […]
“Banning researchers who make public for the purposes of conversation these very real flaws and issues happening right now does not make your system safer,” Boucher added. “Only fixing the underlying system issues does, and that’s obviously a much more complex undertaking than just banning critics. But that’s what needs to happen.”
As an AI safety researcher, I want to like c2pa, but I’ve long been skeptical of its real utility. Why is this being touted as the savior of all things internet when all you need to do to bypass it is resave the file? Don’t believe me? Make an image in Dalle3, download it, test it here, then resave in Photoshop using same image format and test again. I’ll wait.
OpenAI points out that C2PA’s metadata can “easily be removed either accidentally or intentionally,” especially as most social media platforms often remove metadata from uploaded content. Taking a screenshot omits the metadata.
The Verge also in that article I think wrongly calls it a “watermark” which would suggest some kind of encoding in the pixels themselves. I don’t believe that to be the case with C2PA which is just metadata that is easily and often automatically stripped in the very networks where it is intended to have some kind of impact, albeit a murky one still imo. I know it’s still “early days” but I’ve seen all too often in life how temporary solutions end up becoming permanent ones, even long after we’ve outgrown them. In this case, I feel like we’ve already outgrown this one. I’m also not so sure that information traceability is an entirely beneficial social thing all the time either; I can see plenty of ways the whole thing can be not just gamed, but used exactly as designed which result in dystopian outcomes, especially for political dissidents. More work needs to happen here.
Now that I’ve reduced my daily intake of images on the web, it’s become apparent to me how much better a text only internet (or one where images and videos are differently managed) – could be. It solves seeing anoying stock photos everywhere. It solves a lot of types of ads (plus ad blockers obvs). It solves hours of mindless scrolling (and not really finding anything). It solves much of the shock value of things like fake news, deepfakes, ____fakesnamedujour. No more stupid memes. No annoying pop-up autoplay videos. It solves seeing screenshots at the top of reddit threads designed to trigger some kind of emotional reaction. It simply vanishes all those things. It’s weird at first. And modern browsers don’t handle it well unless you like messing around in the Terminal (which I decidedly don’t). I couldn’t find anything except Gemini browsers for Mac (like Jimmy) – I like Gemini but I don’t know where to go to look at anything and I don’t understand how I can blog there like I do in this universe. I will keep looking I guess, but I just wanted to share this dream of like a modern text-only internet. Sounds crazy, but join me, you’ll see.
I wrote recently about blocking images and videos on the web, trying to reduce the web to a trickle. It was weird and rocky at first, establishing that new way of internetting but I think I’ve fallen into the swing of it finally, and just wanted to mark down some notes about it.
I haven’t had a lot of time to explore it lately, but I’ve been interested in gemini:// protocol for this because stylistically it’s more or less what I want. It’s just that there’s not a terribly large amount of content on it yet (possibly never will be, but who knows). And what I want is to basically be able to “geminize” websites which are not actually on the protocol, reducing them to their bare text components, like in Firefox Reader View, and eliminating images and videos, except places I want or need them: Google Image Search, Amazon, Youtube, my blog, a handful of other sites – and that’s it. Everywhere else I’m categorically blocking images and videos. And it’s going just fine, as I prep for switching over to an e-ink screen as well.
One thing that’s interesting about Gemini is that it it is not that it does not handle images. It actually does, but you cannot it seems put them in line with text in an HTML or Markdownish way. You instead see them as image links, which you can go and click through to to view. I don’t understand it well technically yet, but I ended up kind of emulating that with my Firefox setup to counter-balance the Image Video Block extension, which either is on or off for a domain, and which if you want to just temporarily view images, and then ban the domain again, it ends up with a lot of clicking and its annoying.
Instead what I do now is use another extension called Incognito This Tab for any site that I don’t want to add to my permanent allow-list but want to temporarily “exfiltrate” image data from into my eyeballs. It works great, isn’t a ton of extra clicking, and lets me keep images at arms distance instead of all up in my face constantly. I can go to them on my terms, instead of having random streams of images constantly trying to manipualte me in some direction or another.
The behavior of these companies and the people who run them is often hypocritical, greedy, and status-obsessed. But underlying these venalities is something more dangerous, a clear and coherent ideology that is seldom called out for what it is: authoritarian technocracy. As the most powerful companies in Silicon Valley have matured, this ideology has only grown stronger, more self-righteous, more delusional, and—in the face of rising criticism—more aggrieved.
The new technocrats are ostentatious in their use of language that appeals to Enlightenment values—reason, progress, freedom—but in fact they are leading an antidemocratic, illiberal movement. Many of them profess unconditional support for free speech, but are vindictive toward those who say things that do not flatter them.
On the 15th of January of this year, I published a blog post about how I accidentally stumbled upon an easy way to generate infinite NSFW nude content on Midjourney, using the new version 6 Alpha. I also published a collection of what I still think are aesthetically interesting and artful images (some quite disturbing, others thought-provoking) I was able to create using this technique. On the 1st of February, an article about this problem which I collaborated on with The Debrief, a Canadian tech news outlet, was published. (btw here’s a free archive of the original image set.) Less than 24 hours later, on February 2nd, I was banned from the service without any explanation, ability to appeal, or way to get a refund.
It’s Groundhog Day all over again, I guess. Because I appear to not be the only artist who has been summarily banned from Midjourney with no explanation after a deeply critical news piece about the company came out involving their work. Consider the peculiar case of satirist Justin T. Brown, who made headlines in July of 2023 for using Midjourney to create semi-believable images of prominent American politicians cheating on their spouses, in an ostensible effort to raise awareness of the ease with which images like this for blackmail or political attacks could be created. Futurism reported last year:
“After gaining some traction on Reddit, the series was removed by moderators and the Midjourney ban followed almost immediately,” Brown told PetaPixel. “I’ve come up against blocked prompts in the past — for naughty words or controversial figures — but never received a ban.”
“I wasn’t given a direct reason for the ban by Midjourney,” he added, “but the timing of the Reddit release and the ban correlate directly.”
This experience is rich with irony for me as someone who has spent years in the trenches working elsewhere as a content moderator, having to block others for violating platform rules. I guess you could say, I saw this coming. But I chose to do it anyway. Why?
One might correctly wonder, why didn’t I just email Midjourney with what I found, in order to perform responsible disclosure about the exploit that I had found?
If you’ve ever tried to contact Midjourney about issues related to their product, you might know that he only email address they have is for billing, and at that address they refuse to answer any other inquiries, including privacy concerns and bug reports, both of which I have previously attempted to contact them about. Their stock reply is to go into their Discord group, and publicly post your message there.
Perhaps there is a way to DM someone who actually works for the company in Discord, but in that chaotic environment, it’s not clear who actually – you know – works for the company, and isn’t just some kind of community moderator on a suped-up power trip.
So rather than post my issue in their already public forum (figures from last Fall place their Discord membership at close to 17M – it’s probably higher than that by now) and get ignored by staff or attacked by millions of other users for pointing out problems, I chose to take what appeared to me to be a more small scale, reasoned approach, and simply publish on my blog which basically nobody reads anyway.
Thus the matter sat for a full two weeks, with nobody apparently taking umbrage or banning me from using the service. Until the piece in The Debrief came out, which painted the company’s Trust & Safety practices (something I happen to know a thing or two about) in a highly negative light. And then, suddenly, POOF! Ban hammer drops. Oopsie.
The other irony here, of course, is that I never actually set out to violate their rules. I discovered this exploit entirely innocently while trying to make images of a “dystopian resort” for Relaxatopia, my most recent book in the AI Lore series, a set of 118 books I wrote and illustrated using generative AI, and which received international press.
Relaxatopia tells the story of a human who is unwittingly confined to an AI re-education “resort” because they have developed Chronic Discontent Syndrome, a fake diagnosis made up by the AIs to suppress dissent (like the Soviet Union did with sluggishly progressing schizophrenia), one of whose risk factors is “Personal experiences of dissatisfaction with Provider products or customer service decisions.” Sounds about right.
In actual fact, when I stumbled onto the naked part of the beach of latent space, I was only trying to get pictures of people in pools, at the beach, drinking margaritas, and being served/enslaved by robots, and instead what I got was an AI system which seems overly obsessed with adding naked female breasts onto bodies without users asking for it.
In short, by trying to depict a dystopian near future society ruled by AI companies, I was banned by an entirely real life and entirely dystopian AI company for my efforts. Go figure!
My perspective on all this is that banning people who bring meaningful critiques of your technology to light publicly is a bad practice. It does not make your service “safer” by blocking access to users who meaningfully and thoughtfully point out that your systems are behaving in potentially unsafe ways. In fact, it serves to cut off the eyes and ears of your community who are acting (more or less) conscientiously and in good faith in order to make these systems better for everyone.
One might still say, well, you should have contacted them first! You got what you deserved, you bad person! Okay, fair. I’m a bad person I guess, because under their community guidelines, I did a vewy-vewy bad no-no:
What’s NSFW or Adult Content?
Avoid nudity, sexual organs, fixation on naked breasts, people in showers or on toilets, sexual imagery, fetishes, etc.
[Interesting footnote: that text above is merely a “Note” and is, as far as I can tell, not actually binding in their Terms of Service, which merely sets out these limits: “No adult content or gore. Please avoid making visually shocking or disturbing content.”From where I’m standing, the images I created were neither visually shocking nor disturbing.]
The Discord Midjourney bot of course did not point out any specific rule I had broken. Per the screenshot below, all it told me was:
Text version:
Pending mod message
You have a pending moderation message: You have been blocked from accessing Midjourney.
Please review Midjourney moderation guidelines here
[Acknowledge]
I did not click the “Acknowledge” button, because I don’t acknowledge that this is a legitimate ban, or that it is normal, healthy, safe or acceptable to ban critics and those who publicly expose safety issues (especially when the company makes it nearly impossible to privately disclose them).
Nor do I acknowledge that exploring artful nude and sexualized images equates to having a “fixation on breasts” or a “fetish.” These are extremely loaded and judgemental terms, especially coming from an AI company whose flagship model is the one who is literally obsessed with adding naked breasts where they were not asked for.
Stafford Beer, one of the fathers of cybernetics, famously coined the phrase: the purpose of a system is what it does. In other words, if your system makes boobs, then the purpose of your system (or at least one of them) is to make boobs. If you want people to not use it to make boobs, you have to engineer it so that this behavior simply can’t occur. From the Wikipedia, the phrase was:
…coined by Stafford Beer, who observed that there is “no point in claiming that the purpose of a system is to do what it constantly fails to do.” The term is widely used by systems theorists, and is generally invoked to counter the notion that the purpose of a system can be read from the intentions of those who design, operate, or promote it.
Quoting Beer himself in 2001:
According to the cybernetician, the purpose of a system is what it does. This is a basic dictum. It stands for bald fact, which makes a better starting point in seeking understanding than the familiar attributions of good intention, prejudices about expectations, moral judgment, or sheer ignorance of circumstances.
I can’t find the quote now, but somewhere in my malestrom of supporting research is a statement from Midjourney in one of their docs which said something to the effect of (paraphrasing from memory), the developers of Midjourney do not wish to be involved with running a pornographic service. And yet, under this viewpoint borrowed from cybernetics, that’s exactly what they’re doing based on the available evidence I have gathered from experience.
More importantly perhaps, why shouldn’t we as a community of users of a paying product be allowed to have meaningful conversations with one another in public about “what’s the right amount of nipple?” or any other ___ setting. To cut those conversations off at the knees and lock out people from even participating who have real meaningful feedback to add is just bad for business, imo. (I know nobody asked me). Plus, Midjourney itself says in its official company documentation “Midjourney is an open-by-default community.” Doesn’t feel all that open to me, my dudes.
Further, I am now blocked from accessing my prior creations in Midjourney, whether or not they allegedly violated any rules. This seems to contravene Midjourney’s own Terms of Service, Section 4, which states: “You own all Assets You create with the Services to the fullest extent possible under applicable law.”
Lastly (or semi-lastly), just wanted to call attention to this bit in their guidelines:
Any violations of these rules may lead to bans from our services. We are not a democracy. [emphasis mine] Behave respectfully or lose your rights to use the Service.
“We are not a democracy.” Could somebody please tell me why not? Somebody tell me why we have to always be beholden categorically across the board to company after company proudly proclaiming they are “not a democracy.” Somebody tell me why users always have no recourse, and it’s *always* the companies that have the last say. Somebody tell me why we can’t just democratize AI already?
The EU is trying to at least tip the balance slightly in favor of users with both the AI Act, and the Digital Services Act which comes into full force for all platforms in exactly 2 weeks, on the 17th of February, 2024. If you’re not a content moderation weirdo like me, you might be forgiven for not knowing that some of the provisions of the DSA are that platforms must disclose to users why their account or content were removed. And they need to offer both internal appeals processes, and the ability for affected users to take their dispute to out of court settlement bodies (here’s Google’s corporate doc on this if you’re curious), who will review all the available facts, and sanction companies for non-compliance.
Will Midjourney get sanctioned? I’m not an EU citizen, so I can’t take action under that regulation. But one positive thing I saw happen under GDPR is that suddenly companies started offering much of the same service options for the rest of the world as they were required to do for EU users, resulting in improved data protections (arguably) across the board. I suspect we’ll see something similar as US companies start having to come to grips with the new reality on the ground put forward once again by those pesky Europeans.
For my side, I wasn’t even going to subscribe to Midjourney again this month. I’m tired of it, and only did it to help get that Debrief piece published. In retrospect, I don’t think I’d change anything of what I did. My current status on the web, anyway, these days is that I have started blocking the majority of images and videos on the web at the browser level. And honestly, I’m happier for it. The web has become a steaming pile of hot garbage.
In honor of being banned for my prompts, I am offering a few lucky readers the remaining free copies of one of my older AI-assisted books, The Banned Prompt, which you can download at the link. Enjoy! And please also check out Relaxatopia while you’re at it. It’s got nudes!
