AI Bias Bypass ?: An Overview of Bias In Generative AI

ALEXA! Play "Confessions Part 2?by Usher"

I'm sure you can tell by my emails and newsfeed lately that I am knee-deep into this AI world right now, and I'm dragging you guys with me kicking and screaming!

But now I have to stop and take inventory of everything that I've learned over the past six months, and there are some hard truths about generative AI.?

I'm a low-key techie, and I like testing the boundaries of new tech to see where the vulnerabilities are. Sadly it didn't take long for me to find the trip wires in AI. It's the blatant?biases. **deep sigh**

I mean, I was sort of expecting them since most products, including tech, are designed?from the creator's perspective with some consideration for a small core group of users.

But I think I was, shall I say, "hopeful".??

...while Usher croons in the background, I'm going to drop some bias bombs in this message.

However, first I need to hit you with the "what we AIN'T?doing" list (insert hoodrat accent).

--?we ain't bashing the AI creators regarding their bias (I don’t have time for that)

-- we ain't advocating for diversity in AI (I'll leave that to the DEI folks)

-- we ain't boycotting platforms until they "do better" (That would hurt us more than them)

What we ARE doing is looking at the bias squarely in its digital face and finding the workaround. I call the alternative methodology the?AI Bias Bypass ?. My goal is to keep users from being derailed by bias and keep them generating outcomes that improve their lives.

NOW, LET'S ADDRESS WHAT THE BIASES ARE

I have uncovered FOUR prevalent biases in the generative image and video platforms.

1. Skin Tone Bias?- People of color struggle with generating quality images that reflect their skin tone. This is likely because developers are not aware the terms that each ethnicity uses to describe their own skin tone.?
2. Body Type Bias?- Body inclusion clearly wasn't a big consideration based on the default images that appear in image generations. For example, with the "face-swapping" generation, people with fat faces are often given a skinny body.
3. Hair Texture Bias?- Words to describe hair texture varies among cultures. And hair texture also affects hair length. So getting the right hair texture takes a little bit of work.
4. Age Bias?- Ethnic groups also age differently. People with less melanin show signs of aging sooner. Also, AI-generated images of kids seem not to match the requested age.

Over the next few weeks. I will dive deep into these four biases and share my tactics using my AI Bias Bypass (TM) methods.?I aim to have you generating crazy cool AI images and videos in no time, so stay tuned!

Catch you on the flip side!

Jai Stone