The evolution of Gender Equality % the role of AI

For centuries, gender inequality has plagued our society and continues to do so. Be it voting rights, right to marriage and abortion, or right to work or secure financial assistance, women have had to fight for their fundamental rights. With more and more women entering the workforce, we now face the problem of income inequality, lack of women representation and other deeply ingrained stereotypes. Some of our society functions through gender hierarchies and it is we, who have created these hierarchies and imbalances. So, while complete gender equality is a goal that is yet to be achieved, high hopes have been placed on artificial intelligence to address them. But with machines and AI, there is this misconception that they are perfect and cannot have human like flaws. Therefore, it is crucial to understand what do we mean by gender bias in AI.

What are the various components of gender bias in AI?

While artificial intelligence is still in its nascent stage, it continues to affect our daily lives in several different ways- job recruitment, approval of bank loans, search engines and more. While AI governs a relatively small part of our lives in the present, it is set to play an extremely important role in every aspect of our daily lives. Hence, it is critical for us to understand how AI affects and could affect our lives and its potential contributions towards gender equality.

Artificial Intelligence essentially functions through a set of rules called algorithms. These algorithms use large data sets to identify patterns, make predictions and recommend a course of action. Over time, AI improves automatically through experience and this phenomenon is called Machine Learning. Therefore, while AI improves on its own, its foundations are laid by the creators whose inherent gender biases, if any, would affect the AI system development. This would lead to an ‘AI application in making biased decision’ regarding certain groups of people thus making it imperative for us to study it closely. Due to these inherent assumptions and biases of the algorithm developers, AI has a massive impact on gender equality, particularly women.

Second is how the popular AI applications we see today are spreading and reinforcing gender stereotypes. For example, most of the virtual assistants powered by AI have women’s voices, and the most powerful computer in the world, Watson, is named after a man.These gender biases risk further stigmatizing and disregarding women on a global scale. Considering the increasing ubiquity of AI in our societies, such biases put women at risk of being left behind in all realms of economic, political and social life.

This takes us to some of pertinent questions that must be answered while we are unlocking the discussion around biases in AI: 1. In what way is AI negatively affecting gender equality and women in particular? 2. How can we overcome these built-in biases found in AI? 3. What can we do to create robust and gender-inclusive AI principles and ethics?

While we know that emerging technology is bringing about a number of benefits and has the potential to contribute to a large number of development goals. It is important to bear in mind as we design, develop and deploy these technologies that their benefits are not always equal. And that in some instances, there are major negative, often unforeseen impacts on gender.

1. Swati Agarwal is Heading Programs and Partnerships at AI Policy Labs, Ruchi Yemul is the Research Associate at AI Policy Labs. This document is developed jointly with Dr Jerome Dubbery, Lecturer, Graduate Institute Geneva.

2. https://en.unesco.org/system/files/artificial_intelligence_and_gender_equality.pdf

In what way is AI counterintuitively affecting gender equality and women in particular?

Several gender biases have been identified with the emergence of AI. While some have little effect on gender equality, others are effectively widening the gender divide.

1. Perpetuating stereotypes: here have been clear examples of AI perpetuating gender biases and stereotypes. We have seen for example, image labeling software that identifies gender associated with carrying out activities i.e they are perceived as female for activities such as baking or shopping. There are language translation softwares that systematically translate women voices to male in male dominated professions. Machine learning software used for these applications, not only mirror existing biases in society, but actually amplify them.

2. Gender-based discrimination: Going beyond harmful stereotypes, AI has been seen in the past to also perpetuate discrimination. For example, screening applications using AI has in some cases replicated gender profiling. So it has made gendered assumptions based on data that were fed such as favoring male candidates for appointments in male dominated sectors.

3. Lack of women representation: Beyond the issue of bias, we also have technology developed with little or no input from women that can simply result in products that do not serve women. We have had for example, voice recognition technologies which were not tested on women’s voices and therefore prove less effective in recognizing female voices. To take another instance, health apps that were launched as comprehensive profiling to track everything to do with health, actually omitted menstruation. This would mean, these applications actually did not serve half or more of its consumer base.

4. Online safety of women: Another major concern in a more digitalized society is the large number of issues related to online safety for women. We’ve seen a rise in online harassment, stalking, sexual predation, and even fake power. There are multiple forms of software and apps that allow the face of one person to be superimposed on pornographic images or even videos. A study in 2018 found that 96% of the use of deep fake technologies is for the creation of pornographic or pedophilic content and 100% of those victims are women and girls.

5. Dangers of job automation: AI also has a negative impact on the economic empowerment of women and their position in the labour market. Recent studies conducted by the IMF shows that women are at higher risk of job displacement due to job automation than men.On average, women face an 11 percent risk of losing their jobs due to automation, compared to 9 percent of their male counterparts.

How can we overcome these built-in biases found in AI?

While AI poses a significant threat to gender equality, it is important to recognize its potential to bring abouta positive change. We asked a few experts from the field to discuss ways in which the biases in AI can be overcome and after careful deliberation, the following recommendations emerged:

1. Design assessment First of all, while designing an AI application, one needs to be acutely aware of the risks mentioned above. This will enable one to identify and build in appropriate checks and corrections at each step. Secondly, the product design and web design teams need to be diverse and inclusive to be able to incorporate suggestions and views from all groups, including women in different processes thereby avoiding unintended consequences of technology. When designing a product, we can seek the inputs of women and specific user groups or consumer groups. There is a need to look at the trialing processes and to properly pilot new technologies before they are actually deployed and put on the market.One such organization that looks into this aspect is the Design Justice Network. The organization challenges the ways that design and designers can harm those who are marginalized by systems of power.

3. https://www.tilburguniversity.edu/sites/default/files/download/TPCS_237-Farago.pdf

4. https://blogs.imf.org/2018/11/16/women-technology-and-the-future-of-work

5. AI for SDGs event on 5th March. Experts included Lowri Angharad Rees from UN Women, Geneva; Anna Maj, Senior Advisor Truffle Capital | Jury Member, European Innovation; Allison Cohen, Co-founder Women Leading in AI; Vaishali Neotia, Cofounder and CEO of Merxius; Jerome Dubbery, Lecturer, Graduate Institute Geneva; Swati Agarwal, Head of Programmes, AI Policy Labs

2. Women education: Like mentioned above, some of these issues may be partly due to the less representation of women in relevant careers. Increasing women representation and diversifying workforce can potentially mitigate the risks of biases when developing these technologies. Therefore, education of women also plays a very important role here. Countries must provide STEM education to women in order to equip them with the appropriate skills needed to enter the field of AI.

3. Augment human capabilities: Instead of substituting jobholders particularly women, AI should be used to augment their existing skills. AI has the potential to replace people in their jobs. But AI also has the potential to educate, train, and augment people, making them better at their tasks and activities. AI can make the quality of an individual’s work better, resulting in better writing, design, healthcare, communication, teaching, and art.

4. Assessing funds: There should be a requirement that if a firm wants the money to fund their research, they are going to have to ensure that their design is truly a developing and trustworthy, technological solution. And part of that is to make sure they have a meaningful diverse team, and stakeholder interaction. For instance, governments could issue value-based financial assessment and relate finance to gender equality norms. This would mean the corporates that meet these gender equality norms and include women in the process of AI development will be granted financial aid.

5. Multidisciplinary groups: We need more multidisciplinary groups working on AI data problems because it isnot for technologists to solve problems. It for technologists along with policymakers, economists, social scientists, anthropologists, behavioral economists, people to really think about how they are going to construct new models that are not just digitization what is already existing. Thus, including individuals particularly women, from diverse backgrounds will help reduce biases in AI.

6. Involving men in the movement: Gender equality should not be a woman’s battle alone, we must get men on board and change their perspective as well. To bring about social progress, we must involve men from various backgrounds. This is again about diversity but not gender diversity alone. It is about age, cultural background, professional background, and ethnicity. In recent years, there has been progress in the AI feminist movement. One such organization led by this movement is Women Leading in AI which essentially a global think tank for women in AI with the aim to address the bias that can occur within algorithms due to a lack of diversity and inclusivity within the field of Artificial Intelligence.

7. Investing in women founded initiatives: There is a need to encourage more support for women lead or female founded initiatives, because some of these initiatives create a lot of good ethical answers about AI today. According to a report , only 2.8% of the total startup funding was invested in women founded initiatives. Boosting investments is necessary to encourage more women to take initiatives and help diversify this field.

8. Reframing AI and gender ethics: Back in 1984, there were about 37% of women working in tech, which then reduced to about 18% within computer science. It has dropped even lower with AI which is roughly about 12% and since about 1984 it really has not changed very much at all. Identifying these gender gaps is the first to step finding a solution to this problem. Tech industries must diversify and provide more opportunities to women to incorporate gender equality in their organizational ethics. Therefore, reconfiguring AI ethics and principles was a unanimous recommendation among our panelists. To make women a larger part of AI and technology in general, there is a dire need to reframe AI ethics and principles. This is further elaborated in the next section.

6.https://designjustice.org/about-us

7.https://stanford.io/3cDewdf

8.https://www.un.org/womenwatch/daw/egm/financing_gender_equality/EGM%20Report%20Final.pdf

9. https://womenleadinginai.org/about

10.https://about.crunchbase.com/blog/business-investment-to-women/

 

What can we do to create robust and gender-inclusive AI principle and ethics?

The need for ethics was identified in 1948 by Norbert Weiner who was a professor at MIT but these ethics had several gaps which were not addressed. For example, the ethics behind exclusion or who should have a say in the design of an algorithm was excluded. Ethics of AI has now emerged as a new field but these ethical guidelines still stand toothless unless they actually fulfill the moral code and become inclusive. There is a need for international cooperation and creating stricter norms and regulations. The aim should be to popularize the idea of using human rights frameworks for all of what we are doing in terms of our AI, and then identify and plug the gaps with ethical solutions. But we do not need to reinvent the wheel and develop new ethical codes. We just start from human rights frameworks from what we all agreed to in 1948 and have been adding to ever since, to make a robust life that affirms to useful tech that helps those in need and all of us, so that we can all thrive.

Way Forward

Overall, AI models and algorithms are touted as being objective and neutral, which is clearly far from the truth. Biases of the people that control the AI systems, continue to affect the system in several ways as discussed above.

It is imperative for this biased system to evolve and change for the good and the first step towards that is to include more women in the process and diversify the teams involved. Moreover, care needs to be taken with the data that we feed into our systems and it should be updated constantly with changing times. Women are slowly coming to the forefront and the machines cannot learn of this inclusion on their own. So while feminist AI movements seek to create a network of diverse women in AI and promote fairness and equality and remove the bias in AI, design assessment teams can come into the picture and check for any biases involved. Such a framework will effectively help examine the aspects of who Is responsible of designing technology, who benefits from it and who is likely to get harmed from it.