Women in the AI era

The upcoming celebration of International Women’s Day is every year an opportunity to reflect on our progress to shape a more equal future by empowering women and girls and by offering them access to opportunities for education, for jobs, for growth.  Despite the progress made over the past years, European Commission data show that there is a growing gap between men and women’s participation in the digital sector in education, career and entrepreneurship.
And this is happening while there is growing demand for technology specialists and digital profiles. If we also consider the impact of the pandemic, which acts as an accelerator for digital transformation across all industries, then the gender gap in the technology industry is definitely reducing women’s opportunities for future jobs. But, there’s more to that.

A couple of years ago, I was participating in a panel discussion where we were called to address the usual topics: equal representation, equal pay, equal opportunity.
While the conversation was really heated and passionate, the moderator addressed me with the following question:

“Do you think Artificial Intelligence when used as a tool for business decision making can offer a second chance to eliminate biases and discrimination against women?”

Nice question, don’t you think?

Women in tech by the numbers

Let’s first look at the facts. European Commission is regularly monitoring the gender representation in the digital sector. According to the Commission’s 2020 Women in Digital (WiD) Scoreboard, only 18% of ICT specialists in Europe are women and the gender gap is present in all 12 indicators measured. Back in March 2018, EU issued another very interesting report named “Women in the Digital Age” which includes great insights that explain why people like me, a woman working in the tech industry, are still a very sad minority.

There are four times more men than women in Europe with ICT-related studies. There is a decrease in women taking up ICT related higher education when compared to 2011.

Actually, for every 1,000 female tertiary graduates in the EU, only 24 are graduates in ICT-related fields, while when you look at male figures the respective number is 92. What’s even more worrying, is that of these 24 graduates, only 6 women end up working in digital jobs (the respective number in men is 49).

What’s more, almost 9% of women yearly drop out from their digital jobs. The annual productivity loss for the European economy of women leaving their digital jobs to become inactive is calculated to be about EUR 16.2 billion.

When it comes to basic digital skills there is no significant gap between men and women. Greater gender differences arise when it comes to what is nowadays considered the “new literacy”, and that is coding. A recent study by Accenture found that 68% of female undergraduates have taken coding classes, compared to 83% of males. Women represent around 10% of one of the biggest international online coding communities, Stack Overflow. A survey carried out by them showed that women have, on average, less coding experience and, again, tend to underestimate their programming abilities compared to their male counterparts.

This is happening while the pandemic has boosted digital, automation and technology industry and has already created 800,000 new jobs across EU, UK, US, Japan and Australia. And while the tech sector is booming, OECD science technology directorate Mariagrazia Squicciarini expressed concerns that “the pandemic may be contributing to a widening of the gender gap, including the digital gender divide and the gender pay gap”.

A story of AI bias

Now that we looked into the data, let me share a story. A story that is focused on the “hottest” technology in our industry today, Artificial Intelligence.

A global e-commerce company is aiming to automate the search for talent by developing a bot that would review candidates’ resumes and would score them against some predefined criteria. However, at some point the company realized that their recruiting engine did not like women; it appeared that women CVs received much less score particularly for technical roles.

How did this happen?

A machine learning algorithm is trained based on a set of available data. Apparently, the algorithm in this case was trained to vet applicants by observing patterns in resumes submitted to the company over a period of 10 years. Most came from men, given the male dominance across the tech industry, so the system taught itself that male candidates were preferable and it disfavored resumes that included words associated to women.

This example is a case of building AI relying on limited datasets that reflected a historical bias. The result? Machine learning models that incorporated the biases inherent in the training data. Actually, the algorithms reinforced these biases.

There are more examples of gender biases inherent in data. Researchers from MIT and Stanford University published a study that shows that commercially available facial-analysis programs demonstrate gender biases due to neural networks being trained and evaluated against data sets that are heavily dominated by male data.

So, what can be done?

We discussed that AI can be discriminatory because of machine learning algorithms trained with historical data with inherent biases, but also because of the lack of diverse talent in the AI field.

The role of the tech industry is to stand up to this problem and ensure equal opportunities for all. And this will happen with a new ethical “contract” that will ensure that AI is built in a way that earns trust. How?

Diversity: AI systems should be based on diverse data. Yet we know that society is filled with biases, and that reflects into data. So, we need more. AI systems should be built and tested by diverse teams. Yet we know that there are few women engineers currently working with AI. Stronger gender representation will build more inclusive AI algorithms and will eliminate biases.

Inclusiveness. Advanced technologies like AI systems have the potential to create a great digital divide. Women are more vulnerable here as they do not feel confident that they have the capabilities to become an AI engineer. Girls during school years, but also women at universities should be encouraged to follow an engineering career in the tech field. More women coders can help create more inclusive systems.

Transparency. AI systems should be understandable. Today it is desirable to disclose the composition of the data on which models learn from and to demonstrate the probabilistic nature of the recommendations that the system is making.

Accountability. The designers and developers of AI systems carry a level of responsibility for the application of the solutions they create and should ultimately be accountable for their use. Businesses should have both the culture and the mechanisms to ensure AI applications are made taking into consideration the individuals affected by these systems.

Rethinking AI

I would argue that the problems in our AI systems are not simply limitations of data or developers’ blind spots, but rather stem from our entire approach to how these systems are built. We have approached AI development as any other information system development. In the AI era, we will need to get back to the basics. The ethical principles which we want to stand for.

Can AI be an opportunity? Sure, it can be. AI potential globally is estimated to reach 1.5-3 trillion dollars in the next 10 years. Why should it deprive women of the chance to flourish in the digital age?

Source: Women in the AI era

Tags: AI, diversity, Women