Gender Bias in AI
How to address the Bias in AI & ML
Today on March 8th, we celebrate international women’s day. This year the women’s day theme is #BreakTheBias. The Bias that is there in our communities, neighborhood, workplaces, and more importantly in our thoughts and actions. AI workforce is no exception.
On one side AI and ML are transforming the world around us and we are relying on machines to do everyday tasks for us intelligently. Whether it’s automated chatbots, social media/e-commerce recommendations, medical treatments, or taking investment decisions we have knowingly/unknowingly adopted AI in our daily lives.
But does AI around us is fair? NO. Most often it is biased toward the world that we want to escape from, which has discrimination based on age, race, and gender.
Let’s see the numbers closely.
According to research by Wired, only 12% of machine learning researchers are female and less than 7% of single-author papers are written by women.
Only 18% of authors at leading AI conferences are women.
This gap also exists in the workforce. According to the research by the World Economic Forum and LinkedIn, only 22% of jobs in artificial intelligence are held by women, this number goes down with more senior roles.
According to the AI Now Institute, only 10% of AI research staff at Google and 15% of AI research staff at Facebook are women.
How Gender Bias affects AI Systems?
With this gender inequality, gender bias is baked into the AI systems. Stephen Hawking once said, “A superintelligent AI will be extremely good at accomplishing its goals, and if those goals aren’t aligned with ours, we’re in trouble.”
Are we ready to live in an AI-powered world that is biased toward a vision that is coming from only half of the human population? Many examples show how harmful this bias can be. Let's look at a few-
- According to a report from Reuters, AI recruiting tools show bias against women and advantaged male candidates.
- Another report shows gender bias with natural language processing in Amazon’s Alexa and Apple’s Siri.
- Another study shows how speech-to-text technology performed poorly for female speakers as compared to males.
- Word embeddings also cite a biased aspect of AI. Systems most often associate ‘man’ with ‘doctor’ and ‘woman’ with ‘nurse’.
- Another study shows a similar example with vector analogy, “man is to computer programmer as woman is to x” was completed with x=homemaker.
- The bias is not only present in text but in photos as well. According to a report, Computer vision systems report higher error rates to recognize women.
Are we ready to live with these outdated views in today’s modern AI systems?
Imagine these biased AI systems teaching in schools, providing biased opportunities and access rights, operating in hospitals and generating medical reports. Gender bias in these systems is inescapable and it can have intense impacts on women’s psychological, economic and medical safety. It can also enhance existing harmful gender stereotypes. That is why bringing diversity in AI is not only important for AI systems but for human well-being as well.
Factors influencing gender bias in AI
Eliminating bias completely is not possible as many factors can influence it. Let’s look deep.
Biased Data
The data chosen to train models is the biggest factor. Imbalanced data which is skewed on male vision or examples will introduce gender bias. Though it's hard to remove this bias, recording more data that is not skewed for males is beneficial. There is a good example of the COVID-19 dataset on how gender-biased data can affect the overall performance of the ML model.
Human Bias
The next bias comes from people who choose the data, create ML models, and test and interpret the results are biased. Consciously/Subconsciously the society we live in is biased and thus the human behaviors and actions. Being aware of the biases can help us to reduce bias in algorithms.
What we can do?
To present AI systems that are fairer, there has to be a combined effort from industry leaders, researchers, and organizations to maintain the balance. Fortunately, there are many growing initiatives on a global scale to empower women in AI, but there is still much to do.
Women’s Day is perfect for reminding ourselves that we need to change and promote gender diversity in AI not only because it is fair for AI Systems, but because we need it for the overall well-being of society.
Few steps-
- Ensure gender diversity in the training data. Solve unfairness by collecting more data.
- Ensure that human labeling of sample data is coming from diverse backgrounds.
- Apply de-biasing techniques to penalize for producing unfairness in ML models.
- Create gender diversity, inclusion, and equity in the AI workforce.
- Advocate for AI literacy for women within and outside workplaces.
- Ensure to have the voices of women in the development and management of AI systems.
Conclusion
Creating technology is not enough, we have an obligation to use technology that is effective and fair for everyone. The responsibility not only goes to the AI teams and organizations but also to researchers and leaders in the field to develop solutions to reduce gender bias in AI. The benefits of AI can outweigh the risks if we can address them correctly.
Lastly, AI only shows us the mirror of what we are as a society. According to a report by Financial Times, “Addressing issues of AI and fairness rests on the fundamental idea that if you do not know how to solve a problem, AI will not be able to solve it for you.” AI can’t solve the gender bias problem without humans doing it in the first place. So let’s start addressing this gender bias in our homes, workplaces & societies first and then hope in coming years we can shape a bias-free AI world.
Thanks for the read. If you like the story please like, share, and follow for more such content. As always, please reach out for any questions/comments/feedback.
Github: https://github.com/charumakhijani
LinkedIn: https://www.linkedin.com/in/charumakhijani/
