Women in AI

[This Article was originally published in The Age newspaper on 11/10/2019]

When I was younger I watched my dad, one of the most intelligent people I know, apply for jobs under the name ‘Tom’.

My dad is a proud Punjabi whose name is most definitely not Tom.

Like many others, he anglicised his name in the hopes of getting an interview, because in the early 2000s having an ethnic name that gave the slightest indication you were brown manifested itself as a big bright red flag in the minds of job application readers.

Almost 20 years later it’s my turn to go through the process of applying for jobs. Unlike my dad, it’s likely that my resume won’t go through the hands of a human but rather will be assessed and shortlisted by a form of Artificial intelligence.

AI has rapidly transitioned from a concept of science fiction films to becoming a constant in our day to day lives. We see terms such as ‘Siri’ and ‘self-driving cars’ quickly work their way into our vernacular as AI enters and radically shapes our economy and world.

One of the most attractive features of AI is its potential to make decisions free from human bias, thereby contributing to a fairer society — meaning my resume shouldn’t be judged any differently if my name was Sarah instead of Sukhmani.

This year the Federal Government released a discussion paper regarding an ethical framework for Artificial Intelligence in Australia. One of the principles at the centre of the paper was that of ‘fairness’.

However, this ideal of fairness isn’t an inherent property of AI, but rather is determined on how AI is programmed. If we want AI to take into account the diversity of our societies, we need to take intentional actions to increase diversity in the AI industry.

In the near future thousands of jobs will be dedicated to research and advancement in AI, but only a small percentage of them will be held by women. Women’s representation in traditionally male dominated fields is often treated as a matter of symbolism, with allusions to ideas of meritocracy and terms such as ‘best person for the job’ inching their way to the forefront of the discussion by the loudest voices in the room. However gender representation in the development of AI transcends mere ‘symbolism’ — it has significant impacts on what our future will look like.

While it is true that AI has the potential to overcome the limitations of human bias, we are increasingly seeing cases of AI algorithms perpetuating and reinforcing existing stereotypes and oppressive structures.

Recently Amazon’s hiring process came under scrutiny as it was revealed their recruitment algorithm favoured resumes of male applicants at the expense of female applicants. AI facial recognition technology has consistently been shown to be tailored to white male faces, identifying them correctly with 99% accuracy while misidentifying dark skinned women 35% of the time.

These are just a couple of examples which illustrate that, given that AI algorithms are entirely shaped by humans, the capacity for AI to result in fairness is contingent on human actions to make these algorithms fair.

As AI enters into almost every industry possible, making decisions from things such as whether a home loan should be approved to whether a prisoner is likely to reoffend or not, the presence of bias in algorithms can be dangerous. There is an urgent need to address whether biases in the technology are leading to exclusionary and discriminatory practices and what we can do about it.

AI operates through machine learning. Simply, that means that a computer uses an algorithm to read and make decisions from a dataset. Bias in AI stems largely from the datasets used. If the data reflects historical inequalities, the decision making of AI will replicate those inequalities. If a dataset ignores or excludes portions of the population, such as women, AI will operate on the assumption that those portions of the population simply do not exist. If the large majority of those in charge of the AI industry happen to be male, there is a high chance that the products of the industry will treat ‘male’ as the norm.

The good news is that we’re not doomed to some digital dystopia which engrains and perpetuates the equalities we have today. Increasing representation in the AI industry is a necessary and effective step at mitigating the risk of creating tools and systems which serve to reinforce existing inequalities. We need more discussion on how to create inclusive coding practices.

AI is rapidly changing our world. Throughout history women have had their futures shaped for them — but this no longer has to be the case. The future is digital and the AI industry is malleable. There need to be tangible steps taken aimed at increasing representation within the field, because ideals of a fair society can only be achieved if there is diversity of lived experience amongst those creating the technology which will shape our future.