Graham's Blog

Creating a Dating Profile With AI

Dec 10th, 2022

I’ve been single for a few years now, and I haven’t had a whole lot of success in the dating game. I’m very introverted and socially anxious, so dating apps seemed like the way to go for me. Sadly, I’ve not had a lot of success, and since Covid it seems like the dating scene has only got worse. I’ve been feeling pretty down about it, but I decided to have a little fun and let an AI help me assemble a brand new profile.

My goal is to, for an entire week, use my AI generated dating profile. That means my bio and prompts would be written by an AI and all my photos would be selected by an AI. Finally, as long as it doesn’t end in disaster, all my opening messages will be generated by an AI–all further responses will be written by me, because otherwise what would be the point?

I decided to use OpenAI for the written portion, and Everypixel to select the pictures. For the dating app itself, the most important ingredient in this experiment, I decided to go with Hinge, mainly because I’ve never used it before.

I decided against using AI generated images because, even though I’m letting an AI create a dating profile for me, I want it to be somewhat genuine. Also I’m not looking to catfish anybody with this image I generated just to see how good AI could make me look:

an AI-generated portrait of me

Originally this was going to be one post I’d write over the course of a week, but as I was writing I decided I should split it into 2 parts: Part 1, this part, about setting up the dating profile, and part 2, coming at the end of the experiment, about the experience and how it went.

My first step was to sign up for Hinge. I went through the sign up process, and got to photos. Hinge requires 6 photos to create a profile, so it was Everypixel’s time to shine! I collected up 42 photos from the last few years—I wasn’t aiming for 42 but, as that’s the answer to life, the universe, and everything, I decided to take it as a good omen—and proceeded to feed them through this AI (which I have no doubt added to some collection of images to train machine learning algorithms. Your welcome for the coming robot apocalypse!).

My first photo didn’t go over very well, with a whopping 0.9% chance of being awesome. Oh well, 41 more pictures to go. After going through the first 5 photos, not a single one scored more than 8%. That’s when I noticed I could get more info about the result, and found this gem: “Surprised by the result? This service doesn’t measure the coolness or beauty of a person or any object in a photo.”

So really all I learned was that my selfies and family photos are garbage. Good to know.

I had to go back to the drawing board. I came across which admittedly seems more dubious than Everypixel, but sometimes we gotta make sacrifices. Anyways, it seemed to go pretty well, but there was one photo I had to disqualify because the AI kept zeroing in on someone else’s face, and 3 that it absolutely refused to process, most likely because it couldn’t zero in on my face.

As I ran through the photos, I sorted them into folders based on the rating they got, from 1 to 10. I don’t mean to brag, but I got a couple 7’s. The problem though was I had 12 6’s, so I couldn’t simply combine these 2 categories so I decided to go back to our old friend Everypixel to pick the top 4 best photos of the ones rated 6. Within a few minutes I had my top 4, ranging from a 1.5% chance to 50% chance to be awesome. Quite a range, and not extremely encouraging but that’s the bed I’ve made for myself.

Interesting note: what I consider the 2 best pictures of me with animals made it through the process and made it to the top 6.

Next up was selecting and answering prompts. While not remotely AI, I decided it best to use a random number generator to help me select which prompts the AI should answer.

This is where I ran into my first major roadblock. Through the magic of random numbers, for the first prompt I landed on, “Give me travel tips for…” The AI, predictably, gave me some admittedly useful travel tips, but I’m not sure that was in the spirit of the prompt. I decided it best to remove the random factor and hand-select 3 prompts I think the AI might be able to handle.

My first selection was, “The hallmark of a good relationship is…” Boy did it deliver!

The hallmark of a good relationship is trust, respect, and open communication. When two people are in a relationship, they should be able to have honest and meaningful conversations, with enough understanding, patience, and empathy to make sure that both perspectives can be heard. A good relationship should also be based on mutual respect, with each person being able to express themselves and their feelings without feeling judged or belittled. Lastly, having a strong foundation of trust and honesty is essential for a healthy and fulfilling relationship.

Sadly, it was a little long but with a little bit of trimming I got it to fit into the 220-character limit.

Next up I picked, “The key to my heart is…” To which the AI responded, weirdly in context with the previous prompt, with the following:

No single “key” can unlock a person’s heart. The best way forward is to build a relationship based on the qualities mentioned above.

Finally, I picked, “My best dad joke.” The AI did not disappoint.

Q: What did the sea say to the beach?

A: Nothing, it just waved!

When I was experimenting with OpenAI initially, I had it prepare a short bio. Unfortunately, it seems that Hinge doesn’t have a field for a bio, but here it is anyway, for posterity.

I’m an introverted geek who loves all things tech, sci-fi, and fantasy. I enjoy staying inside watching movies, playing video games, and reading comics. I’m looking for someone who shares my interests and is up for some exciting nerd-related activities. I’m a reliable and loyal person who loves to make people laugh. I’m also an animal lover and I like to spend time with cats and dogs. If you think we could be a great match, let’s get to know each other better!

So, how well will this work? Will AI turn my love life around? Or will this backfire spectacularly? Nobody knows! But I’ll be sure to follow up in part 2. 🙂

Life Tech