Neural network training (Patreon)

Admittedly it was incredibly petty and I should have just moved on once Andrew and I had our fallout, and honestly I could have stopped at any point. It’s not like Sarah and I were actually a pair, but it is still shitty for a friend, a good friend, to make a move like that. With so much effort spent I was however really anxious to see my plan through to the end.

I had spent a lot of work curating the different image sets needed: fashion models, athletes, and porn stars of various kinds, predominantly male. I then opened a “research project” web site where people could describe the images as detailed as possible. I got a big response by advertising for help on porn sites. It was dirt cheap to advertise there, and it allowed me to get both straight and gay men, and some in between, to contribute.

Then followed text analysis, feature extraction, regression analysis, and finally building and training an AI that could differentiate between straight and gay just by analyzing what people look at in images and how they describe them. It really was a shame that I couldn’t use any of this in my actual thesis at the university. The really big question was however if I could use this data on Andrew.

I had told him that I wouldn’t delete the unfortunate images of him unless he helped me with my thesis by subjecting himself to “alien anal probing”. He probably didn’t take that quite serious, but I also offered to give him an equally embarrassing photo of me from the same party, as we both knew I wouldn’t actually delete the images of him. Whatever his reasoning, he showed up on time despite what had happened between us.

He was close to walking out though when I showed him the butt plug with vibrator and electro-stimulation that I wanted him to shove up his ass. It was still in an unopened packaging, a bottle of lube provided, and I assured him that it really was part of the study. After a lot of hesitation his bad conscience made him agree to it, and I made every effort to be as professional about it as I could. He got the room to himself to insert it and to attach a stimulus-ring around the base of his cock and balls. When he called me back in, he was sitting by my test setup still wearing his black track pants, but with wires coming out the back. I got him.

I connected the wires to the control box that connected to the computer. Next I put a headband with electrodes on his head and connected that to the box as well. Finally I put a headset on him and told him to turn towards the computer screen for the test to start.

The program starts out with a set of diverse and innocent images showing a wide selection of men and women in pictures that all could have been in normal magazines. Most of them however showing aspects that could be taken as slightly sexual, such as divers, swimmers, fashion models, construction workers, and so on. During the first stage the subject is prompted to describe the images with no feedback given, while eye movement is recorded through the webcam eye tracker. This creates a baseline classification.

The program then switches over to training mode. For every image it shows it will prompt the subject to describe it, and respond with positive stimulation for all descriptions that are in line with how a gay person would describe it. If the descriptions are too generic it will ask follow up questions like “describe the face”, “describe the clothes”, “describe the arms”. If the right keywords are still not provided it will move into even more direct questions like “is he handsome?”, “is he athletic?”, “are the abs good looking?”. These are all words taken from the gay descriptions of the image, to provide a vocabulary the subject can use in future descriptions.

The positive feedback at this stage is low levels of electro-stimulation to the prostate. The more unprompted the desired descriptions are, the more intense the positive feedback is. At the same time, transcranial direct current stimulation is used through the headband to amplify and reinforce the behavior.

In this phase the program starts out with fashion images, initially with an equal gender mix. Descriptions on the female images give no rewards and are slowly phased out from the mix as the subject gives less and less enthusiastic descriptions. At the same rate images of male athletes are phased in and electro-stimulation of the cock is added. Once the descriptions pass a threshold in “gayness” the program will phase in pornography in increasing levels of graphic nature and the prostate vibrator will be used as well.

Finally the training program will begin phasing in males in more traditional masculine roles, police officers, firefighters, police offices, factory workers, inmates. The high levels of reward feedback are maintained for descriptions and eye-tracking that correlate to homosexual behavior.

As a last step the program reruns the initial set of images to see what the difference in descriptions is, and more importantly the eye-tracking.

It took Andrew three hours and 20 minutes to complete the program. His baseline went from 11% gay to 92% gay, which is a fantastic change. I need to run the baseline test again in a few days to see how permanent the change is, and Andrew agreed to come back for a follow-up. Regardless of outcome, I don’t think he’ll date Sarah again. Or any girl.

Read my commentary.