top of page
  • Jose Arrieta

On the Use of Algorithms for Generating Adult Content

I often ask myself what my grandchildren will say our generation's empathic failure was. By this, I mean the racism of our grandparent's generation or the sexual discrimination of our parents. It often feels like every 50 years or so, a new barrier is passed, and as Obama often said, the arch of history bends towards justice.

As a migrant, I selfishly believe that the taxation without representation many of us are subjected to might be one. A similar thing might be said of the lack of representation in the democratic process of children and parents of multi-child households. The fact that we have a better retirement than childcare system in Germany is not a real mystery.

Just lately, though, one came to mind. During the past months, the internet has been ablaze with large language models that can generate Turing-passable content, be it text, images, or video. The companies that give us access to these tools tend to provide strong restrictions so that people do not generate pornographic content with these tools. Indeed, many lives have been hurt by deepfakes and revenge porn.

However, countless humans are exploited in the pornography industry. I completely understand that controlling a genie once out of the bottle is hard. But, if something can happen, it will. I am also starting to believe that this genie might not be all bad. I mean, what if we could make good enough pornographic content that requires no physical labor, just the crafting of good prompts? If the disruption is large enough, the price for porn might decrease enough so that exploitation becomes not worth the risk. Maybe.

What follows is my first attempt at trying to understand this topic. I am sure it is very biased and not a new idea at all. But only in showing my bias can I learn to be more empathetic. Ultimately, I hope not to be a homophobic uncle or racist grandpa. If I am to succeed in this, I need to learn at some point where my empathic weaknesses are.

Creative Destruction

Pornography has been at the forefront of the audiovisual revolution that starts with the first Silverhalide cameras coming into existence. Few images were more coveted in the late 1800s than closeups of pieces of a human normally covered in layers of cloth.

The fascination continued for the intervening century. High-quality color print, color video, and even 3D films were first commercialized to enable more to see the naked flesh of the young and attractive. The revolution spilled over to information technologies. For decades the hard drives that house the internet were filled--mostly-- with pictures and videos of naked humans. Even today, the biggest videoconferencing and webinar platforms focus on displaying young women without clothes.

Algorithms Enter the Scene

In 2012, AlexNET, a mostly naive neural network, beat the best image detection algorithms at the ImageNet challenge. This victory changed the world. This victory changed the porn industry. As soon as 2016, face recognition tools emerged to check which people were hidden in the hard drives of pornsites. This caused a backlash as a vast portion of the material hosted in these HDDs was uploaded without the consent of the people exposed. Websites like opened systems to check if your body parts were leaked online. As backlash, PornHub, a titan in the industry, removed all "unverified content," cleansing their hard drives of countless hours of content.


So far, I write without condemnation. Pornography is not necessarily illegal and not necessarily immoral. Every entrepreneur can use ignorance a few times as a defense. That is before it turns into negligence.

That said, there is rampant exploitation within and around the pornographic industry. Countless lives are destroyed to satisfy the demand of mostly men willing to cross boundaries. Moral and legal boundaries are often crossed. Legal boundaries as the sharing of content without authorization. Or moral boundaries as the building of production companies in places filled with economic and societal turmoil.

A small subset within this industry tries to be fair and just. Ethical porn is a current topic of conversation that tends towards direct payment by the streamer to the physical laborers themselves. Yet, not many pay for porn, and as always, fairness and justice has a price tag attached. In terms of a disruption theory, fair and just porn is a higher quality product, not a minimal viable product.

The Rise of Deepfakes

In the past weeks, the internet has been bombarded with texts, images, and videos generated from Large Language Models such as GPT3, Stable Diffusion, and DALLE-2. These systems have access to--for intensive purposes-- the totality of human knowledge. They use this knowledge to create Turing-passable outputs. The texts are often boring, but the images are beautiful, at least when artistic care is given to taming one's prompts.

The gatekeepers have kept the generation of pornographic content outside the boundaries of acceptable prompts. This has worked well, outside minor controversies. Gatekeepers do this because of deepfakes. Also known as revenge porn, this implies the creation of pornographic content that resembles a real human, often used to shame them. But the main problem, if I understand correctly, is linking faked content to a specific human. Let's assume we can go around this problem.

A Last Exploitation

If we take all legally uploaded images and videos available on the internet, one can generate pornographic content. In fact, Unstable Diffusion is trying to do exactly that. I claim this is a morally ambiguous problem.

On the one side, we have the Luddite problem brought on by any creative disruption. Just like graphic designers and painters, the people who make a living performing and filming pornographic content will be hurt. The fate of the creatively destroyed is an ancient problem faced with disregard by most economic theory. So I won't engage with it.

There is a clear point in regarding the copyrights of the "texts" (e.g., letters, pixels, and videos) used to train the models. The bodies depicted in this content are human and form the crucible upon which diffusion algorithms create their outputs: i.e., the content. As before, this is a credit assignment problem and can be resolved under some form of innovation in the copyright system. Similar to how Tiktok remunerates the people who made the music used in viral videos: a fickle kind of truce.

What could be gained is the creation of novel content that pleases the user base and does not require humans. It will, first and foremost, become a form of innovation. It will remove humans' need to bear their bodies and perform highly taxing physical work to earn a wage. As such, it will decrease the cost of production.

Drain the Swamp

As our societies prosper, it becomes harder to find people willing to spend their time making pornographic content. In other words, as societies prosper, the price of pornographic content increases. High prices incentivize new firms to provide content. The marginal firm often comes from marginalized communities that face turmoil and exploitation. It is no mystery why there is less content coming from Western than Eastern Europe.

If algorithmically generate content is good enough to satisfice a large part of the internet, the price of pornographic content will go down. With the lower price will go much physical labor and potentially a lot of exploitation. That is, as long as people watch porn for the moving pixels and not for watching evidence of a person having sex.

As a society, we might need to support the current workers who lose their livelihoods and provide a form for remunerating the people whose physical labor helps train future content. But in doing so, we might transition to a world in which our grandchildren might tell us: "Hey, Grandpa, how could you all watch porn? Did you not care for the people's rights?"

PS: The image above comes from the Humanae project from Angelica Dass. I love this project. It actually led to my first-ever post.

1 view0 comments

Recent Posts

See All
Post: Blog2_Post
bottom of page