IBM to Fight Bias With Improved Artificial Intelligence

By Tracey Goins

  • IBM aims to develop fair and accurate facial recognition technology

  • IBM misidentified gender in up to 35% of darker-skinned females

Tech giant IBM (IBM -1.16%) wants to better train facial recognition systems in order to increase diversity and lower bias. Since facial recognition software is on the forefront of identification, and can be used to verify the identity of a person through photo or in real time, the data set needs to be much larger. To fix this IBM released a trove of data containing one million images of faces taken from a Flickr dataset with 100 million photos and videos.

Artificial Intelligence Face.png

Why This Matters: There is a looming problem with the algorithms. Datasets used to create facial recognition software lack diversity, causing a surge in misidentification, particularly among people of color. Among other tech giants, IBM was found to have misidentified gender in up to 7% of lighter-skinned females, up to 12% of darker-skinned males, and up to 35% of darker-skinned females.

IBM was found to have misidentified gender in up to 7% of lighter-skinned females

Law enforcement agencies are using facial recognition more frequently in routine policing. This unfairly affects African Americans who have been historically targeted for arrests. Criminal databases include a disproportionate number of African Americans, due in part to racially biased police practices.

The use of face recognition technology has a disparate impact on people of color, disenfranchising people who already face inequality. Even Amazon’s (AMZN -1.58%) controversial Rekognition was recently found to have misidentified 28 members of Congress as people who have been arrested for crimes, raising new concerns for racial profiling and potential law enforcement abuse.

Situational Awareness: Beyond law enforcement agencies, facial recognition is being used at airports, border crossing and even major sporting events. IBM is slated to release a new dataset, Diversity in Faces, with hopes it will help developers tackle gender and skin type biases. Artificial Intelligence systems inherently learn what they are being “taught”. A diverse data set increases accuracy and lessens bias in facial recognition. While critics call for justification for the use of the technology and in some cases an all-out ban, artificial intelligence continues to be a billion dollar industry.

CBx Vibe:Every Breath You Take” The Police