AI has a bias problem : Amazon Killed an AI Recruitment System for Discrimination Against Women

One of the main fields in Artificial intelligence is machine learning, involves training computer software to do new things, by feeding them a huge amount of data about the subject in hand. Often Data alone isn't ground truth sometimes it can be biased and it will simply force the AI system to do biased decisions _ Amazon seems to provide an excellent example of biased systems.

Amazon team has spent years working on a recruitment system. By looking at a collection of resumes and name this AI-powered system should be able to rank candidates. In order to achieve fed the machine learning algorithm with a decade's worth of resumes of people applying for jobs at Amazon.

It's known that the tech industry is male-dominated and most of the resumes Amazon fed to their systems came from men. As an outcome, the system trained in this selection of data, the system would favor men over women.

Artificial Intelligence has a Bias Problem

The Amazon system has learned to downgrade resumes with the word "women" or "female" and assigned lower scores. In the other hand words such as "executed" or "captured" suggested candidates to be ranked highly.

The development team of this system tried to make adjustments and stop the system from taking such features into account, but unfortunately, they said it was impossible to stop it from finding new ways to discriminate women. Apparently, the biggest issue was the data itself that.