Continuing my series of AI generated character art using bing.com/create, I am pleased to introduce one of my favorite characters from my upcoming novel. If I had a backlog of published work, I'm fairly confident she would still be one of my favorites. You've already met Sarah and Robert, and now it's time to meet...
Alfia
Alfia is Sarah's friend and fellow classmate in the computer science A.I. program at New Galveston University. She's determined, sarcastic, commanding – and she’ll do whatever it takes to win. Even though she projects front of smug confidence, she's riddled with insecurities, afraid she'll never live up to her father's reputation... or expectations. I could go on and on about Alfia, but you'll just have to read her chapters to find out more.
Moving onto Bing's image creation, this is where things started to come unhinged, as hinted at in Part 1 of this series.
Unlike the other prompts where I just took whatever Bing gave me from the first prompt, this one took a lot of work. I wanted to see if Bing could branch out from a base physical description of a character, so I included a description of my character doing something. In this case, I wanted to show Alfia playing cards and eating Chinese take-out.
Right off the bat, Bing blocked my prompts for violating an internal set of standards, probably to make sure people aren't generating hateful images. Fair enough. But what had I done wrong?
I described Alfia in a similar fashion to my prompt for Robert. I gave Bing a simple action (playing cards). Eventually, I figured out that asking for Chinese takeout was what had offended Bing's standards. A bit over zealous, but I could see how an AI that doesn't actually understand what I'm asking for could conclude that I was asking for something inappropriate. Fair enough.
Shockingly, after removing the offending phrase, I was rewarded with – super objectified stereotypes for women of color?!? Good thing there wasn’t any Chinese takeout in the images!
Face palm. At this point I realized our expectations of AI can quickly fall short as an AI's output is just a derivative of the inputs. Bing didn't really create those images. It just served up what we already create. Which was definitely an eye opening experience. I hope that seeing what generative AI creates will help hold a mirror up to our own society and cause use to think a bit more about what we choose to create.
Existential crisis aside, I finally got a well-dressed result for a prompt. To be honest, I'm not too disappointed at the final result -- Alfia looks like a girl who could school anyone at cards, Atari, and... whatever's in her hands. I got nothing for that.
Prompt:
modern indian woman deviously playing games, pixel art
Result:
Verdict: 8/10, but only after losing faith in mankind.
Okay, so the hands are a bit off and the eyes are looking in different directions, but at least there's a CRT screened TV involved. I'm actually really surprised that 'Chinese takeout' was blocked, but 'Indian woman' was accepted. I actually tried several descriptions based on her general physical features, but that seemed to push Bing's images in the wrong direction, as previously mentioned.
Well, we're almost to the end of our 'Meet the Cast' post series. I hope you've enjoyed the trip so far, and I look forward to sharing the final post about the book's main character... which is now complete!
Comments