AI & Our Impossibly Horny Decade

**Introduction: The precipice we stand on**

This is an essay I've been trying to write for a while, but it's tough. Without giving too much away about myself, during the day, I'm a technology consultant to major companies. At night, I'm an avid enjoyer of all things erotic. I never really expected these two existences would merge, but lately it's seemed inevitable. Not only do I think the world of cutting-edge tech and adult entertainment are about to collide, I think it will change society forever.

If you've been online at all in the past few months, it's been impossible to ignore the deluge of art and images generated by Artificial Intelligence tools like DALL-E 2, MidJourney, and Stable Diffusion. These Generative AIs have a seemingly magical ability to take text prompts (descriptions of a desired image), and with enough delicate adjusting, strumming these tools like an instrument, the images emerge. Everything from hyper-realistic goofy mashups (say, "Mickey Mouse at the Louvre") to impossibly beautiful fantasy worlds.

There's a feeling in the air that we're on the edge of something huge. We're reaching a turning point where anything that our minds can dream up, can be visualized, in stunning, breathtaking detail.

But there's something else that's being unsaid. The Generative AI models that exist today are set up to only be able to produce Safe-For-Work content -- either because they're run on a respectable company's servers, and lewd requests can be denied; or by trying to scrub all erotic imagery from the training data that powers the models.

Neither of these can be held back forever -- Stable Diffusion exploded in popularity partially because it could be downloaded and run on private servers, where the NSFW filter was easily disabled. The Stable Diffusion team responded by trying to update a 2.0 model to exclude any NSFW imagery from the training data -- and therefore make it impossible to generate. But that just fueled the lewd AI enthusiasts more. As of writing this, a kickstarter campaign has raised nearly $50,000 to develop a custom model exclusively around pornographic imagery.

AI-generated pornography, as it turns out, is already here. Like a lot of the rest of generative AI, it's in an early state, often requiring lots of tuning and attempts to get exactly the image you want. But there are dozens of communities already perfecting the art of the Lewd AI prompt. Furry groups are tuning models to get especially good at emulating the anatomy of their cartoon animals. On other sites, specialized models are starting to emerge that can create an endless stream of photo-realistic porn on a specific topic: an endless stream of fake boobs; or a never-ending feed of variations on doggy-style sex shots.

A change is upon us. Up until now, there were two ways that NSFW visual content was generated: either by performers ready to bare it all to the public, or talented artists prepared to dedicate hours of their time to drawing out fantasies. But it is now inevitable that in the next few years, the ability to produce pornographic content -- photorealistic, drawn, or in any style imaginable -- will be a click away from anyone.

It's a cliche at this point that breakthrough technologies often find their first "killer use case" in adult entertainment (from home media to the internet). Most people need to come up with an artificial reason to try out Generative AI today. But anyone who's horny will have a reason to try out what's coming.

This essay is broken into two parts: what I think is about to happen to the world of adult entertainment; and why I think it's going to change the role of adult entertainment in society forever.

Strap in, because the next 10 years are going to be wild. And crazy horny.

**Part 1: What's coming - three predictions on the future of pornography**

(1) The amount of pornography on the internet is about to explode.

If you take away nothing else from this essay, you can think about this: Today, anyone can write an erotic story -- from a multi-part Literotica sexual epic, to a couple of horny fantasy Tweets. What happens when every story on this site and beyond can be visualized, in all its horny glory, with a few clicks?

The most extreme version of this -- "Paste a Literotica URL link below, then come back in an hour to see the photo-realistic full-length porno of it" -- probably feels far away. That's just ridiculous, fantasy technology, right?

Well here's two predictions you can log for the upcoming year (2023): (1) There will be a flashy press-release about "the first AI generated Porn Star!" that gets endlessly shared online within the next year , causing some sort of moral panic; (2) This will be followed shortly by OnlyFans making some public policy decision about banning "AI Performers" from the platform.

That'll be the signal that AI Porn tech is officially ready for consumer use. At that point, there's no reason not to expect the amount of AI-generated pornography to approach infinity, a never-ending feed. Some may be far away (5-10 years out), but some is going to be revving up almost immediately.

AI Porn will probably stay as photos for a little bit. The idea of "enter a sentence describing your fantasy, get an image out" is definitely close.

Generating hentai (or western-style drawn erotica) may arrive faster (i.e. AI can produce it at an acceptable level of quality) than photo-realism, due to the relatively simpler structure of drawn images (it's easier to get something that looks "right" than with photos). Hell, this week there's been a trend going around of this tool that turns any photo into the "manga-fied" version... and even this supposedly "SFW" tool takes innocent images and returns horny-as-hell hentai characters.

And the mainstream AI community is hard at work making it easier to get what you want out of SFW models, so there's no reason to expect that those same advances won't advance NSFW generation as well. For instance, I expect SFW generative research to try to solve problems like "have consistent representations of characters and locations between different renders" (this is relatively hard to do today, although tools are already coming out to get around these limitations). There will be huge interest in the SFW community to have (essentially) a "cast" of characters and their "sets" for mainstream storytelling.

But of course, this has huge applications for NSFW content as well. So maybe not "Literotica --> Porno Flick" right away, but if we can already generate images alright, is it easier to picture "Literotica --> Hentai/NSFW Comic?" Think about that -- any fantasy narrative, no matter how elaborate, depraved, or uniquely kinky, will be visualized beat-for-beat as an AI-generated comic, by anyone with an internet connection.

Video itself is an active research area -- companies like Meta have released models that can do GIF-length renders. What more applicable field than the thriving NSFW GIF community? It feels naive to not expect longer films on the horizon (for similar reasons as the SFW advances I talked about above -- because people want it for mainstream storytelling).

It's getting a little into "speculative" territory, but if and when this technology does come to fruition, and thousands (or maybe even millions) of AI Porn films are generated daily, they'll have no limits that current adult filmmaking must adhere to. They might be as high-quality as Hollywood cinema, or look like they're shot from a cellphone. They can have emotions as authentic as our best actors (apologies to the acting efforts of porn stars), and they could exceed the logistic constraints of filming pornography today (e.g. hundreds of chapters with casts of dozens of consistent characters). Not to mention defying the laws of physics, realism, or constraints (in space, orgy in the middle of Times Square, etc.)

In other words, expect the adult internet of the future to become dominated by content generated on a server, not filmed in reality or drawn by a human.

(2) Pornography is about to become get more personalized.

In my opinion, one of the beauties of human sexuality is its variety and depth. Everyone has their own preferences, turn-ons, kinks, and fetishes, from the major to tiny details. Part of the thrill of lewd content today is the thrill of the hunt -- searching out the story that perfectly captures your fantasy, or the dirty video that's just the perfect mix of romantic and filthy. When you find it, it's like an extra rush of dopamine (alongside arousal).

But there's another feature of Generative AI that could change all that: it doesn't just generate new content from scratch. It can also take existing content and modify it based on use input. Today, this is being used in SFW image generation as a kind of magical photoshop -- what if I take this photo of my mother, and ask it to give her a big floppy hat? Or erase that stranger in the background? What if instead of this background in our living room, it was a rock concert? And what if instead of my mom, it was a clown? etc.

You may see where I'm going with this: in the future, any single piece of pornography will be infinitely customizable.

This is already being explored in a basic way -- the AI-powered tool PornPen lets users add or remove tags to modify the images that others have generated. See a pic you think is sexy, but wish the model was a different age? Just switch from the "18" tag to the "40" tag, and a whole new pic gets generated. Or swap the bust size to your preference, or someone's mood from "happy" to "angry." Or change her clothes, hair, or setting. You get the idea.

In the future, expect that all porn will be this customizable. An erotic image, GIF, or streaming video won't just be a single thing, but a starting point and skeleton for infinite variations to perfectly fit the viewer's tastes. Or there may be a whole new field of "porn customizers" who tweak and alter content, before sharing out their supremely lust-worthy "directors versions" to the world.

And it goes beyond just switching the models to become the viewer's "perfect girl/guy." Whole sexual acts might be added or removed from a film -- the epidemic of cunnilingus-free couplings could finally be reversed. Maybe you can alter the plot line (these are no longer step-mother and step-son, what if it's student and teacher?). A heterosexual scene could become a homosexual scene (or a slider could make angles more or less homoerotic). A rough hookup becomes romantic. Cell-phone quality shots are dramatically-lit HD. A couple could become a threesome. A hotel could become a shopping mall.

That's not to mention another obvious implication -- who can be the start of any piece of pornography? Well obviously: you.

(3) You may not even need to seek out porn at all -- it will be made for you.

This is really where everything comes together. The previous section talked about the "thrill of the hunt" of looking for the perfect piece of smut to turn you on. But what if you didn't need to hunt? What if that thrill of discovery, or of discovering something new and uniquely arousing, just happened for you, constantly, for as long as you wanted?

Many adult sites are already building recommendation systems that customize "recommended" feeds of pornography based on what you've watched and liked in the past. Some claim these systems are using "AI" recommender systems, which may or may not be true. But AI Recommender systems would become uniquely potent if plugged into the previously covered generating, customizing capabilities.

In the future, your learned preferences won't just be used to serve recommendations of pornography, but actively shaping what's in images and videos as you consume them. Rather than choosing to view an individual image or video you may simply choose a couple of tags of interest and "start the stream." As new pornographic content is generated, you just give thumbs up or down reactions, or suggestions, and things shift around you.

With each minute, the system will learn more about how to make you aroused, and update the content accordingly. In this way, every porn viewer and AI-porn site will become an arousal feedback loop. All AI systems have goals. The goals of these systems will be simple: stimulate us. Just as AI can provide super-human strategies in games like Go and Starcraft, shouldn't we also expect that they'll be able to craft superhumanly erotic experiences with AI content as well?

You just strap in -- your peak lewd stimulation is fed directly to you. Where will it lead? You'll just be in for the ride as you find out.

**Part 2: Why it matters - the NSFW future**

This would truly spell out a brave new pornographic future. And while it's partially speculative, given the rapid advances in Generative AI just over the past year, I think it's silly to dismiss it as never being possible.

So far, I've only talked about "what" I think is going to happen, and that's just half the story. Even more important are the potentially massive downstream effects this could have. They range from mere curiosities to possible societal crises we will need to grapple with.

I'm putting them in no particular order (in fact some of the later ones could be the scariest).

1. Ultra-hacking of the reward cycle for unfathomably stimulating, rewarding, addicting pornography

This is largely laid out in the section on recommenders with a super-human customization ability. At an extreme, the infinite feed of hyper-tuned pornography could have noticeable societal consequences. Every generation clamors that increasingly stimulating porn will sexually immobilize generations, but the fearmongering isn't all bluster -- according to large surveys, young men are having less sex now than in previously measured generations. While these statistics don't label pornography as the cause of this drop-off, it's possible that the hyperstimulation of future pornographic experiences is so satisfying that it drives future generations away from pursuing real-life sexual experiences, relationships, or even reproduction (possibly if combined with other tech like VR and teledildonics -- letting people literally step into a world of infinite, hyper-tuned sexual pleasure).

2. Generation of illegal & disturbing content

This is probably the big one we need to figure out. If a model can generate images of sexual acts (already possible), and the model can generate images of minors (already possible), there will eventually be a model -- that can be run on someone's private computer (i.e. without the ability to be monitored) -- that can generate realistic, sexualized imagery of minors (not to mention other content people may find disturbing, content such as bestiality, or incredibly violent, gory imagery). And it's not hard to picture a community like 4chan with access to these tools, in an ongoing battle to 'one-up' the depravity of what came before. This is a huge area of concern with major societal consequences, far beyond what's laid out here. How could we prevent this from happening? It's completely unclear to me. In my opinion, this is probably the single top issue we must start preparing for now, rather than scrambling retroactively to fix what's been unleashed.

3. Targeted photographic sexual harassment

If DeepFakes were bad, we're just getting started. Imagine the ability for nearly anyone to quickly generate hyper-realistic imagery of anyone a graphic sexual scene, and either post it online or send it to the victim. How would you like it if someone emailed you a video that appeared to show you, or your loved ones, in an orgy, without needing anyone's consent? Even if the content is obviously faked, this is going to be psychologically scarring for many (how would you go through life knowing that if you pissed off trolls online, they might send your boss at work or family members a GIF that looks like you performing sexual acts -- and that image will be forever burned in their minds?) This has already started to happen (the number of obscene requests to Stable Diffusion including actresses like Emma Watson and Emma Stone was kind of unbelievable to me), and similar to Issue #2 above, we need to figure out how we're preventing it now before getting overwhelmed by the flood.

4. Economic Consequences to the Adult Industry

This is a potential sea-change. With an infinite universe of perfectly customizable women who can be digitally contorted to perform any sexual act, just the way you like it, will the demand for human-performed pornography plummet? Will it parallel human artists starting to feel discarded, with their work being the bedrock of AI systems while their potential future opportunities start fading away? Will this rock the adult industry even worse than the changes following the advent of streaming pornography (as outlined in Jon Ronson's fantastic "Butterfly Effect" podcast?). At the very least, might this accelerate a shift to the "OnlyFans-ification" of adult modelling, where part of the appeal is the ability to connect socially to adult creators? In other words, does human pornography become a luxury, or a commodity?

5. Increased Sexual Experimentation

Less of a risk and more of a postulation: could recommender systems make people more likely to explore & engage with new fetishes? For instance, could a recommender/customizer/generator system start subtly introducing a fetish to a user within content they're already known to like (e.g. someone who wants to explore a foot fetish could have feet slowly and subtly become more prominent in generated content that fits all their other preferences)? Using feedback from the user, these systems could guide the user towards the fetish in a "maximally acceptable" way.

6. Further Distortion of Sexual Norms

There's already outcries that the generation "raised on internet pornography" created expectations about sexual experiences based more on the fantasies of the adult industry than the sex people really have behind closed doors. If the main form of pornography consumed moves even further away from actual sexual acts, does it open the door for the next generation's expectations to warp their sexual reality even further? For instance, if the majority of videos of blowjobs that people are watching are optimized to be maximally stimulating -- but not necessarily grounded in how blowjobs work in real human anatomy -- could that actually change the way people expect to receive or give blowjobs in real life to match?

7. Sexual Manipulation - Subtle Version

The pornography created by adult entertainment companies has already started shaping the tastes of those who consume it. You'd expect that porn companies would respond to the tastes of the consumers (e.g. if more people fantasize about X, there will be more porn made for it). But that's not exactly true — adult entertainment corporations optimize for money. A mid-2010's deluge of step-family based pornography came about because companies learned it was a fetish people would pay to see content for. And it seeped into consciousness, then changing the type of porn that people were seeking. The future "personal porn agents" will respond to incentives in similar ways. Depending on how they're programmed, they could co-shape a user's sexual preferences, while optimizing for their goal metrics. In particular, if part of the fun of new sexual interests in their novelty, could agents be incentivized to subtly introduce novel fetishes in contexts they expect users will already find agreeable, then ramping up their prominence to reap the "novelty" rewards? Or could websites with porn agents steer users more towards content areas that they're monetizing (e.g. manipulate users into preferring camgirl content, while simultaneously promoting their camgirl sites?)