WaiONSTAGE: Shaping gender-inclusive AI
Last week, the inaugural event of Les matinées de la science focused on AI and its related societal impacts. Our Ambassadors’ Lead Emilie Labidoire shared Women In AI point on view and initiatives on gender gap and bias in AI as well as guidelines to tackle them.
“AI models and systems we create and train are a reflection of ourselves. Yet, only 22% of AI professionals globally are female.”
AI systems are trained and learn from human and real-world data to inform their decisions, meaning there are at risk of inadvertently replicating the social biases and even worse amplifying them incrementally over time.
“Gartner predicts that through 2022, 85 percent of AI projects will deliver erroneous outcomes due to bias in data, algorithms or the teams responsible for managing them.”
Some illustrations of algorithms creating the same unfairness as people with unconscious biases or more worryingly exacerbating gender stereotypes:
· AI recruiting tool prototype developed by Tech company dropped because it was favoring male applicants for technical jobs and discriminating CVs including “women college” or “women association”
· Facial analysis AI programs displaying gender bias demonstrating high level of accuracy for determining the gender of lighter-skinned men versus high level of errors for darker-skinned women
· Targeted ads where algorithms are perpetuating gender stereotypes for housing and employment despite neutral targeting parameters, e.g. listings for secretarial work mainly pushed to female users.
While there is no bullet-proof approach, we believe in holistic approaches to build gender-inclusive AI that can benefit the global society.
Some guidelines to increase Workplace inclusion1:
· Pursue multiple talent approaches and initiatives - including hiring people returning to the workplace or reskilling in AI -to maximize diversity
· Get Executives involved, e.g. ask them to share their stories of gender inclusive workplace
· Foster dialogue by organizing and participating to panel events, encouraging female role models to share how they succeed and all participants to contribute to their personal stories on the benefits of diversity
· Propose training on unconscious bias to raise awareness of diversity
· Promote a culture of accountability and transparency (including around compensation and promotion)
Some guidelines to address gender bias and discrimination in AI systems2:
· Assess upfront how AI will be used in context and whether it should be used for the targeted outcomes
· Involve diverse and representative development teams - different perspectives, backgrounds, etc. - at every stage of the product cycle
· Screen historical data to detect pre-existing issues/biases, e.g. stereotypical concepts of gender such as woman as homemaker vs. man as engineer
· Ensure diversity in the training samples
· Encourage teams to measure accuracy levels separately for different demographic categories to identify potential discriminatory outcomes
· Monitor algorithms and systems on a regular basis to ensure there are not becoming biased - even if not biased at inception, they can learn the biases -with action plan to address them if detected in AI’s actions, decisions, etc.
“In addition to empowering AI female pioneers and identifying good and next gen gender inclusive practices, we strongly believe in building skills and confidence in young women to increase the percentage of women who study and work in AI — and help to address AI’s potential biases.” — presented Emilie Labidoire, Women in AI Ambassadors’ Lead.
For instance, WaiCAMP — our flagship initiation and orientation workshop — provides introduction to AI through hands-on projects to young girls in order to highlight AI’s potential for good and their role in driving change.
AI can also help increase inclusion. Our 2019 WAI award3 rewarded Goshaba an AI hiring solution which automates the pre-qualification of job candidates by combining cognitive science, gaming and intelligent data to assess experience, competence, personality rather than just CV’s experiences. It identifies candidates who may have been ignored in the traditional recruiting process.
Our Talk was followed by a captivating presentation from Luc Julia to demystify AI and the related hypes and fears based on its book “L’intelligence artificielle n’existe pas” (soon to be released in English : “There is no such thing as Artificial Intelligence”). His demonstration showcased how Augmented Intelligence amplifies human potential by complementing the other’s strengths. Luc also stressed out that biases in AI may act as a barrier to the functionning of the related systems.
This insightful session concluded with a debate moderated by François Euvé, Rédacteur en chef de la Revue Études et Professeur au Centre de Sèvres.
Many thanks to Cécile Du Verne for offering us the opportunity to present our return of experience and latest initiatives to increase female representation and participation in AI!
www.womeninai.co — Join us in building the future of AI!
#artificialintelligence #ai #AI #biasinAI #diversity_in_tech #wai #womeninai #wai_france
1. See more insights in AI NOW report: “DISCRIMINATING SYSTEMS Gender, Race, and Power in AI”, April 2019
2. See more insights in Josh Feast article “4 Ways to Address Gender Bias in AI”, November 2019
3. https://www.wherewomenwork.com/Career/1649/Capgemini-unveils-Women-in-AI-Awards-winners
