AI is such an ephemeral term, let me focus on current-day systems, like transformers.
I do enjoy generated explicit content no more or less than any other media - whether it is text, still images, audio, or video. It really depends on the content itself and how the individual piece turned out. I like to compare it to drawn material - there are some artists I like, others I do not. Same with machine learning based approaches, there are models trained on more data or less data, there is overfitting, underfitting, and more ore less amount of generalization that together with the director's input can produce appealing or non-appealing material.
But that is just the thing, it produces material, not content. One might argue that the amount of porn available today is enough and we do not need to create more, I did some estimations on that and based on the extent of my image collection alone and metrics I collect on it, I could watch unique images for eight hours a day, every day, for the next 100 years and never look at any image twice. And my collection is not that large in modern terms.
So volume is not an issue, what is a problem is selection. Among all the art styles, content, positioning, and execution surely it is difficult to search and find exactly what one is looking for, especially if it is difficult to articulate.
Enter generative networks! Take the insane amount of material available, and train a model on it. Out comes a glorified search engine for explicit material. Or rather, interpolations of material available. If I like to look at three people getting it on, dressed in blue and green lingerie while one of them is looking at the viewer - I can simply ask for it and something comes out - no more having to come up with a contrived way to find what one is looking for.
And who cares that no actual human created that piece of art exactly? How would you even know given the shear amount of material out there?
Then there is the ethical and capitalistic side of it. Some artists share their work for recognition and profit, which worked around and negated in the training process, not to speak of the obfuscation happening when generating material - even if the result is close to an actual piece, no-one would ever know who it was from. I think this is a serious problem that must be addressed. I do not really care about the capitalistic motivations though.
Then there are transformative models that do not as much rely on interpolating a given art space but taking one thing and putting it onto another. Things like deepfake come to mind, but there are also ways to use models to create art incrementally, think combining something that takes text and generates stick figure animations. Another model would flesh out background, yet another could turn stick figures into realistically looking people, perhaps from a set of consenting people that have been used in training these models. Add some music and out comes a video that has much more right to be called original content than any image-generating models. Also, credit can be given to the people involved and what is more: None of them would ever have to perform any erotic or sexual act to begin with.
This opens up a possibility I very much wish for. Removing exploitation from the porn industry! Yes this way would eliminate a source of income - steal jobs if you will - but in my mind that is a good thing when it comes to porn industry.
Another form of "AI" is the large language model - ask a question, get an answer. These are phenomenal at two things: Contextualizing well-established knowledge, and providing an answer at any cost. Much like image-generating models, these could be used to create stories, fiction. Looking at kink, that could mean actual stories, it could mean getting inspiration for scene planning. It could open up ways to find new things to explore. For established kinks LLMs can provide lists of points to further research. They are an excellent tertiary source provided ALL the facts are checked.
One thing in particular is creating a form of virtual dominant for the very specific use case of self-bondage. The problem there lies in conditions for release, which in my hypothesis need to be more complex than a timer or a single set condition, but need not the complexity of an established D/s relation with an actual human.
There may even be a way to use language models to replace actual subs and doms. Needs and wants are so diverse, it stands to reason that purely interpolated interactions are sufficient, after all fantasizing is a thing actual people do. It then comes down to individual suspension of disbelief and the amount of context the LLM is able to hold on to. This is perhaps fun for casual play, but any long-term interaction that would inevitably involve a form of growth is out of the question.
4. What do you think the risks are when involving AI in kink?
Foolishly relying on LLMs generating answers to safety-related questions. Misunderstanding, misconception, bodily harm, death.
When relying on them too much, especially exploiting them as substitutes for long-term interaction, can perhaps lead to losing touch with reality.
5. Would you mind if I just put a non-related question here to get a rounded 5 questions?
I am a bit disappointed you did not ask an LLM to write a fifth question
