The curious case of why A.I is so bad at drawing hands

The sophistication of Artificial Intelligence (AI) is beyond our comprehension. It can do anything and everything and people are worried that this is the start of the robot uprising where they will become our masters.

Right now, Generative AI is all the rave, it can be prompted to create practically any image you want.

A cat in a suit, holding a dog while in space, boom, there you go my future servant.

If you want a heartwarming picture of a child holding a loaf of bread and running from her mother, say no more.

Artificial Intelligence tools such as Dall-e, Firefly, Midjourney, Dream Studio (Stable Diffusion) and Generative AI by Getty Images can render an image that can pass as realistic, can make photos better and even generate a photo of yourself underwater or in space.

In 2022, an AI-generated photo won first prize for digital art at the Colorado State Fair, triggering an intense backlash from human artists.

But for all its technological knowledge and advancement, AI has one glaring and hilarious Kryptonite; drawing realistic hands.

While it has come a long way from last year when it was practically impossible to draw a hand or fingers, you still get instances when it generates fingers with extra digits, and an arm behind the head but the hand is on the thigh.

For this story, I prompted Microsoft Co-Pilot to generate an image of a black girl holding a phone and voila, the first image it spat out was nightmarish and quite hilarious.

At first, all is well, then two seconds later, hilarity ensues. Yes, the girl is on her phone but check where the hand is and where the arm is. Even horror movies haven't thought of this yet.

There is no reason why something as advanced as Generative AI should be struggling with this.

BuzzFeed emailed Midjourney, Stability AI, which makes Stable Diffusion; and OpenAI, which created DALL-E 2 to try and understand why Generative AI has a problem with hands.

“It’s generally understood that within AI datasets, human images display hands less visibly than they do faces,” a Stability AI spokesperson told BuzzFeed News.

“Hands also tend to be much smaller in the source images, as they are relatively rarely visible in large form.”

BuzzFeed also interviewed Amelia Winger-Bearskin, an associate professor of AI and the Arts at the University of Florida.

Generative Artificial Intelligence that’s trained on billions of images scraped from the internet, Winger-Bearskin explained, does not understand what a “hand” is, at least not in the way it connects anatomically to a human body.

When training AI on things like hands, many variations of hands are fed into the machine learning model to find patterns and make predictions.

From photos to art, hands can be shaking hands, holding bottles, pointing, clasped, or holding fish, so the model never really gets one true image of how a hand looks like and how it acts.

AI models have not yet been fully trained with sufficient data to focus on hands specifically.