SSH and Model Collapse
How false assumptions guarantee socio-sexual failure
I’ve been thinking a lot about AI model collapse, which I’ve personally witnessed in my own work with music, writing, and image creation on a regular basis. For example, to maintain character consistency, it is absolutely vital to utilize a source image, as utilizing a second-order AI image will reliably result in a third-order image that is quite obviously not the first image.
This is particularly noticeable in video generation, when the character turns away, then turns back and has an observably different face. The problem is even more serious when the iterations are seeded with additional artificial content.
The more “poisoned” the data is with artificial content, the more quickly an AI model collapses as minority data is forgotten or lost. The majority of data becomes corrupted, and long-tail statistical data distributions are either ignored or replaced with nonsense.
Even when you keep the burden on the AI as light and easy as possible, and add only a small amount of artificial content limited to a specific domain, the models still generate inaccurate information. Consider the famous poster of Heather Thomas, iterated ten times with the only new factor being the color of the bikini, which was specified to be a different color each time. But not only is accuracy lost in the changed area, more and more visual corruption enters in throughout the entire image even though there was no need to fit to change at all.
Now, consider the way in which artificial content in the form of false information enters into the mind of a young male’s socio-sexual model of the intersexual world. Whether he is operating on the basis of false assumptions derived from film and television, bad advice from women in his family, or equalitarian propaganda instilled in him by the education system, all of it is going to lead to delusional premises and irrational conclusions.
This is where the SSH, being a taxonomy based on real-world observable behavior, is absolutely crucial, because it provides a lens for determining whether one’s individual socio-sexual model is more or less functional or not.
The "Copy of a copy" problem. It's theorized that this exact issue is why humans die of old age: our cells are copies of copies, and damage accumulates over a lifetime until the cell copies are too damaged to even perform their basic functions.
Originals used as a basis for iteration with AI are like "stem cells" - the basic blueprint, without errors, will produce "healthier" copies.
I do wonder, if AI could somehow be trained to reduce or even reverse copy error, what that might entail for biology research and study. We could very well have a real-life HemaTech. Hopefully without the vampire overlord puppetmasters behind the scenes, of course.
I’ve been recalibrating my socio-sexual model of the world since finding the hierarchy. I had false assumptions based on film, bad advice from my family, and equalitarian propaganda shoved down my throat at school and then reinforced at home. It’s been a helpful tool. I’ve torn down all the mental models given to me and have had to build my own. This is what a lot of zoomers have had to do.