Twitter has announced the results of an open competition to find algorithmic bias in its
The competition confirmed the conclusions drawn earlier.The work, which won first place, showed that Twitter's cropping algorithm favored persons "slim, young, fair or warm skin tone and smooth skin texture, as well as stereotypically feminine features." The second and third place entries showed that the system discriminates against people with white or gray hair, indicating age discrimination, and favors English over Arabic in images.
First place in the competition and the main prize of 3500US dollars were received by Bogdan Kulinich, a graduate of the Kiev Mohyla Academy and a PhD student at EPFL, a research university in Switzerland. Kulinich used an AI program called StyleGAN2 to create a large number of realistic faces, which he varied in skin color, feminine and masculine features, and slimness. He then incorporated these options into Twitter's photo cropping algorithm to determine which one was preferable.
As Kulinich notes in the accompanying note,these algorithmic biases reinforce prejudice in society, literally cutting out “those who do not match the preferences of the algorithm in terms of body weight, age, skin color”.
Twitter's open approach contrasts with reactionother tech companies facing similar challenges. For example, when researchers led by MIT's Joy Buolamwini discovered racial and gender bias in Amazon's facial recognition algorithms, the company launched a massive campaign to discredit participants, calling their work “misleading” and “false”. After months of controversy over the findings, Amazon eventually gave up, temporarily banning the use of these algorithms.
Illustrations: theverge, Bogdan Kulinich</ p>