Tech in Action: Using AI to defy biases

To bring food insecurity closer to home, Feeding America turned 1,000 images into a photo of one person.

Artificial intelligence and image recognition are two areas of emerging technology that are particularly vulnerable to bias; a programmer or data set can influence the decisions an AI system makes,while privacy advocates continue to be concerned about how image recognition could be used to discriminate and surveil, especially people of colour.

But a new campaign is suggesting that, in the right hands, that technology could also be used to counteract humans’ own biases and preconceived notions.

Not having access to food is typically thought of as a problem faced by the homeless, or as an issue in developing countries. But as Feeding America (a U.S. non-profit network of over 200 food banks) points out, 37 million Americans face food insecurity and it can impact anyone, regardless of where they come from or where they currently live.

In a new advocacy campaign meant to show just how close to home food insecurity can be, Feeding America worked with the Ad Council and Leo Burnett to compile images of more than 1,000 people in the U.S. that were demographically representative of the people who struggle to have regular access to food. An artificial intelligence network then used the images to create a single image of a person that was representative of all the photos. To take the photo out of the 2D world and make it more real, it was then overlayed onto a real person using visual effects technology.

The end result was a single spokesperson that could be used to represent everyone who faces food insecurity and speak about the issue, though the visual effects are also used to bring individual stories out to the forefront.

The AI-generated image will be used in a campaign that includes ads in TV, radio, print, outdoor and digital. The story about the development of the campaign and the issue of food insecurity is also appearing on Feeding America’s website, along with stories of real people and families.