AI searches using the words “imagine entrepreneur,” “imagine inventor” and “imagine software engineer” predominantly yield images of men, with 1% images of women — results that reveal gender biases in neural networks. Composite graphic: KAUST
King Abdullah University of Science and Technology (KAUST) is launching "Dear AI," an International Women's Day campaign and hackathon to tackle gender bias and the under-representation of women, Saudi and Arab people in AI software.
A recent KAUST search of a popular AI imaging software tool returned an average of 1%
women when using the prompt “imagine entrepreneur,” “imagine inventor” and “imagine
software engineer.” Yet globally, women represent one in three early-stage entrepreneurs and 20% of computer
scientists, and in Saudi Arabia, 45% of startups are owned by women. The results
highlight the biases in neural networks and raise questions about how they are trained.
The name “Dear AI” reflects KAUST’s call for accurate representation of datasets and
demographics from KAUST, Saudi Arabia and beyond.
Portrait of Jürgen Schmidhuber, Director of the AI
Initiative; Professor of Computer Science at King Abdullah University of Science
and Technology. Photo: KAUST
"Gender bias is a well-known drawback of many learning AI systems based on
artificial neural networks, especially regarding women of color," said KAUST
Professor of Computer Science Jürgen Schmidhuber, director of the
KAUST AI Initiative.
"One solution is to retrain the algorithms on appropriately selected unbiased
data sets. Creating such sets, however, is a non-trivial task. Nevertheless,
I am encouraged by the fact that Saudi women are turning to tech as viable career paths
and 47% of the graduates in our AI academy program are women. That by itself may move
the needle in this space: more women working in AI may help to generate new and less
biased data sets as a natural by-product of their work."
Saudi Arabia is witnessing significant acceleration of women in scientific fields and KAUST is at the forefront, notably for being the first mixed gender university in the country, which heralded a new era of inclusion for women in academia. The University has continued to create professional opportunities for women on campus and in the Kingdom. For example, KAUST maintains a female student body population of 39% — a figure higher than the global average of women in STEM programs — and its AI extension courses by the KAUST Academy graduate 47% women.
KAUST also has a mandate to train young entrepreneurs and scientists from around Saudi Arabia. Its entrepreneurship programs have trained more than 24,000 people with an average female participation rate of 51%. It's MENA-based startup accelerator program, TAQADAM, now in its sixth year, has a female founder rate of 49%, well above the global average.
Portrait of Asrar Damdam, founder of Uvera; KAUST PhD
candidate and alumna (MS 2018; Electrical and Computer Engineering). Photo
courtesy of Asrar Damdam
Asrar Damdam, founder of Uvera, and Shahad Geoffrey, founder of Taffi, are both TAQADAM graduates who have become inspirational figureheads for women entrepreneurs in Saudi Arabia, recognized in Arabian Business' Most Influential Women and Forbes' 30 Under 30 in the Middle East, among other distinguished publications.
"Saudi Arabia has a booming entrepreneurship ecosystem for women," said Hattan Ahmed, director of the KAUST Entrepreneurship Center. "Nine out of ten women here view starting businesses as favorable. This is reflected in our programs where women from across the country represent over 50% of those trained. Not only are the AI results biased and out of touch at a global level, but they also don't reflect us locally – and we want to change that."
The "Dear AI" media campaign was developed by VMLY&R Dubai to highlight this lack of representation. "Right now, in the creative world, AI is met with either excitement or skepticism. I believe AI is a reflection of reality — it's what we teach it to be and our campaign reflects these biases. And while we can ask the machine to learn more, it is us that should do the teaching," said Fernando Miranda, creative director at VMLY&R Dubai.
A KAUST-hosted "Dear AI" hackathon will continue the conversation into summer,
convening AI and machine learning students and scientists from Saudi Arabia and the
globe to retrain these tools and mitigate the effects of gender bias. To this end,
hackathon participants will review demographic dataset categories, labels used and
modelling features, as well as create new datasets for training. The hackathon
underscores KAUST's commitment to drive change by creating innovative solutions to
tackle global challenges.