Adobe patents AI tech that scores 'diversity' of images based on race, age, gender

 Adobe is seeking to patent artificial intelligence technology that audits and scores the “diversity” of images based on race, age, and gender, according to a recently filed patent application, the Daily Upside reported.

The computer software company’s patent application, “System and Methods for Diversity Auditing,” which was issued last week, uses face scanning technology to provide a “diversity score” based on the “sensitive attributes” of an image. 

“Diversity auditing is a subset of data auditing that refers to various processes for assessing whether a given set of data includes a diverse set of characteristics,” the patent application stated. “For example, a set of images can be audited to determine if the images depict people of diverse races, ages, and genders.”

The application noted that current similar systems require “manual identification,” which is not scalable for auditing larger sets of images.

“Accordingly, at least one embodiment of the present disclosure includes a machine learning model that identifies a plurality of images, detects a face in each of the plurality of images, classifies the face in each of the plurality of images based on sensitive attribute, generates a distribution of the sensitive attribute in the plurality of images based on the classification, and computes a diversity score for the plurality of images based on the distribution,” it explained. 

Additionally, the system “augments the set of images to increase diversity” to ensure diversity goals are met. If the score for a set of images is too low, it could inform the user. The AI program would no longer require a user to manually audit or tag the images.

The application noted that the image scanner software could be used to audit images on a company’s website to determine the level of diversity. Those results could then be compared to census or employment data.

Last year, the investors in the AI image scanner technology published a paper, “Generating and Controlling Diversity in Image Search,” that claimed stock image repositories and search engines reflect “generations of systemic biases.”

The paper cites the example that a Google Image search for “plumber” disproportionately returns photographs of “young white men.”

The authors argue that reordering existing images will do little to change the biased results. Instead, the inventors suggest using AI technology to create additional “photo-realistic high-resolution images.”

“The pursuit of a utopian world demands providing content users with an opportunity to present any profession with diverse racial and gender characteristics,” the inventors stated.