O Twitter is creating a new competition between researchers and hackers, with the objective of finding someone capable of correcting the supposed racial and gender bias in their image-cropping algorithm, according to information from Business Insider.
The competition, which will offer a $ 3,500 reward (around R$ 18 thousand reais), wants help to identify possible damages that the algorithm may have suffered. In addition to the winners, the runners-up will also be financially rewarded.
“Calling all bounty hunters! We’ve just released all the details of our bounty challenge which is open until August 6th”, shows information from the company’s website. “With this challenge, we aim to set a precedent on Twitter and the industry for proactive and collective identification of algorithmic damage,” they added.
The action comes after a group of researchers found that the algorithm favored more white people than black people and more women than men. Last year, doctoral student Colin Madland revealed the problem in a tweet, when he showed that the Zoom video calling platform erased a black man’s face when he used a green background.
A faculty member has been asking how to stop Zoom from removing his head when he uses a virtual background. We suggested the usual plain background, good lighting etc, but it didn’t work. I was in a meeting with him today when I realized why it was happening.
— Colin, scholar in residence since Mar 17/20 (@colinmadland) September 19, 2020
The contest winner will also be invited to present their work at the DEF CON AI Village workshop promoted by Twitter in August. “Successful applications will factor in both quantitative and qualitative methods in their approach,” the company said.
According to information from Insider, machine learning algorithms, like the one used by Twitter, depend on vast data sets. If these datasets are weighted in favor of a particular race, gender, or something else, the resulting algorithm can then reflect that bias.
Confirmed in study
After the reports on social networks, the platform’s engineering division carried out a study to verify what actually happened to the algorithm when exposed to photos with black people. Published in May this year, the analysis concluded which algorithm created a difference of 8% in favor of women and 4% in relation to white people. Among the possible reasons for the phenomenon, Twitter claims that it happens due to problems with picture and eye color backgrounds. But in the report, researchers reasoned that the failure is not justified.
“Machine learning-based image cropping is fundamentally flawed because it removes the user’s ability and restricts the expression of their identity and values, rather than imposing a normative look at which part of the image is considered the most interesting,” they wrote The researchers.
Sign the EXAM and access the most important news in real time.