Facebook says that it will expand an online course in deep learning to more students to help improve the diversity of its AI division. After a successful pilot program at Georgia Tech, the company will roll out this graduate-level course in deep learning to more colleges across 2021. The focus will be on offering the system to universities that serve large numbers of Black and Latinx students. It’s hoped that, by improving the diversity of the people building these systems, some of the more odious biases will be weeded out.
This is part of a broader program to encourage people to enter the computer science field even if their undergraduate training is in another area. In 2019, the company put $4.2 million into Northeastern University’s Align program which helps encourage under-represented groups take graduate programs. It helps, too, that the global pandemic means that there’s a greater embrace of online learning, which itself can reach people that would otherwise be unable to participate.
As well as co-developing the Georgia Tech course, Facebook AI’s team lectured, mentored students and allowed a pathway to an interview for a position with the company. Students will be able to take part in various projects with real-world application, including work to improve medical imaging. They will also, of course, be able to examine work that Facebook is doing -- including algorithmically identifying hateful memes -- on its own platform.
Initiatives like this reflect the number of notable instances where AI / deep learning systems have been shown to have severe biases. This includes facial and speech-recognition systems that discriminate for employment opportunities or criminal profiling. In 2019, a report said that the profession was on a “tipping point” towards a disaster, where algorithms encode historic biases against certain groups.