Picture Imperfect: How AI Helped Debunk Myth of the Perfect Mom

Sed molestie tellus mattis diam rutrum auctor. Cras vitae elementum eros, non rutrum ligula. Nam ac diam nec augue tincidunt lacinia sed nec nisi. Nam pretium non diam ac porttitor.
by Emily Bryce

On social media, in magazines, on advertisements all around us, perfection is personified.

Everywhere we go, idealized images of models with flawless skin, snowy-white teeth and just-so hair surround us.

In a recent ad campaign for personal care product maker Baby Dove, MHP Communications asked U.K. creative agency Happy Finish to use AI to turn this model of perfection back on itself.

By making the unreal seem real, they sparked a conversation about one of the most emotional topics imaginable: mothers.

Aimee, the Parental Paradigm

Baby Dove sponsored a survey that found nine out of 10 first-time mothers said they feel pressure to be perfect due to images they see in social media and magazines.

But what defines the perfect mom? According to social media and magazines, “Aimee” is the parental paradigm. With her pristine complexion, flowing blonde hair and sea-blue eyes, Aimee gazed at commuters passing through London last April from billboards while holding a newborn next to the question “Is there a perfect mum?”

The reaction to the blond, conventionally beautiful model was furious — until Baby Dove revealed Aimee is not real.

In fact, she’s the amalgamation of nearly 2,000 advertising images collected from social media posts and the pages of glossy magazines. Happy Finish created her using generative adversarial networks, running on an NVIDIA GPU.

Described by AI pioneer Yann LeCun as “the most interesting idea in the last 10 years in machine learning,” GANs work by pitting two deep learning networks against each other.

In this case, a generative network tried to create the most convincing “mom” from the database of images. The other, discriminative network tried to detect its flaws. With one network creating and the other analyzing the creations, the two networks learn from each other as they strive to reach “perfection.”

The result was Aimee, an AI avatar that challenges us to rethink how the media represents women, revealing the flaws of perfection.

AI as Change Agent

Happy Finish created Aimee using some of the newest AI deep learning techniques. She also illustrates how AI can challenge our ideas and break down human biases, some of which will almost inevitably infect the very way people create their AI models and algorithms. (Learn more about how humans bias AI in the AI Podcast interview with Narrative Science’s Kris Hammond.)

AI can be a tool to help us understand how we interact with the world. And it can help us tackle some of the issues we face.

To find out more about GANs, check out our AI Podcast conversation with their creator, Ian Goodfellow.