This article was published on InsideHPC. We have reproduced it here.
Keen movie goers and frequent fliers will undoubtably have come across the latest offering from Mimi Leder, On the Basis of Sex. Starring Felicity Jones as one of the first female Harvard Law graduates, the movie tells a story of women’s rights in the 1970s that stands out for a number of reasons.
Not only does it have a female lead, a female director and an indisputable pass for the Bechdel test, but I also picked up on the role the US Department of Defense played with one of the earliest examples of mainframe computing being used for a mass search of documentation.
The story begins with Ruth Ginsburg’s determination to succeed at Harvard Law school, only for her hopes to be shattered when no one in 1960s New York would hire a female lawyer. Move the clock on 15 years, she’s a law professor and her teenage daughter inspires her to bring a landmark case before the US court of appeals.
The case she brings is a loophole in which an unmarried male caregiver is discriminated against by the tax law. The hope is that by amending the law to remove the discrimination against men, the floodgates will open and a precedent will have been set to overturn the hundreds of Federal laws that discriminate on the basis of sex.
The compute challenge
What could be performed by a high school student and a Raspberry Pi nowadays took several days to run in the 1970s. Upwards of 20,000 pages had to be searched to find every law that mentioned women, sex, gender and related key words. The hundreds of laws that were uncovered spanned everything from the piloting cargo planes to working in mines quietly. Though the film’s portrayal of the Pentagon computer has some basis in fiction, the process wasn’t.
Although by modern standards this job was small fry, the automated task of searching for the offending laws and many other such compute problems tackled in the last century paved the way for the HPC machines that affect the lives of everyone in the world today. They were a critical precursor for the natural language processing and machine learning techniques that we are only just beginning to appreciate.
The future we are building now involves many decisions made by machines. Like children learning about the world, the machine learning algorithms are only as smart as the information we give them. What many seem like a neutral task with no possibility of discrimination will many times turn out to be more complex.
To draw a contrived example, we’d all empathize with a driver coming down a hill towards a crowd with failing brakes who swerves to save a child, catching others in his wake instead. Will the self-driving cars also act in favor or against individuals on the basis of age? What about gender or race? How will we ensure that the training data reflects the ethical values that we would like the system to have? Will machine ethics evolve in the way human ethics have?
The butterfly’s wings
Ruth Jones was able to affect great change by spotting a small opportunity to use the status quo to show that change was possible and reasonable.
There’s a real argument that creating a more diverse HPC workforce will help us to create machines that make more diverse decisions. By bringing as many minds together as possible – men and women, of different ages, races and ethnicities – we can make sure all problems are considered on the basis of every possible human angle.