How AI is Amplifying Biases

Graphic of a robot and a human shaking hands, agreeing to destroy the world together.

A recent article by The Verge details another instance where artificial intelligence technologies are replacing human work. Using automation and AI for background check removes the human element of application reviews that would enable apartment managers to empathize and sympathize with the various factors in an applicant’s criminal history.

The feature tells the story of a disabled man’s mother who is attempting to find an apartment for him, but is unable to secure housing because of a minor offense that is more than a decade old. Apartment applications have been added to the list of things being targeted by AI developers, but many worry that the tools will only help to discriminate against vulnerable groups.

More companies are turning to AI to streamline application processes, including for jobs. In October, Amazon nixed its AI recruiting tool, according to Reuters. The tool was biased against women — it taught itself that men were more preferable than women in software development positions and ranked resumes with women-only colleges lower than other applicants.

One of the biggest issues with automation and artificial intelligence is that it learns from human actions; therefore, any biases that have been shown in reality, may be mimicked by the tools. These biases have been reflected in other AI tools, specifically with facial recognition.

Amazon came under fire last year when its facial recognition tool Reckognition misidentified black members of Congress as criminals. The company later responded saying the the tests for their technologies were not on the highest settings and therefore had negative impacts on results.

Google has also had to address racial and gender bias in its automation tools. The company had to correct and error in its facial recognition technology that misidentified black people as monkeys and reworked its Google Translate algorithm to provide gender neutral responses for career titles.

IBM announced that it is developing a Diversity in Faces (DiF) dataset to help offset some of the biases identified in facial recognition technology.

In all, artificial intelligence and automation is increasingly having a huge impact on the daily lives of Americans. It is affecting the opportunities one has access to and without the necessary developments, many people could bare the brunt of technology filled with negative biases.

By: Arriana McLymore

Illustration By: Michael Korfhage