The Ghana Centre Democratic Development (CDD-Ghana) has launched its study report on Evaluating Gender Bias in Artificial Intelligence (AI) Applications using household survey data.
AI, which is the technology that enables machines to stimulate human-like cognitive processes such as learning, reasoning and decision-making, is rapidly transforming how people do things, from healthcare, finance and education, media to entertainment.
Over the past year, AidData and CDD-Ghana have been working on a gender related AI project that looks at the gender breakdown of commonly use wealth indexes.
AidData was awarded funding for the project through the United States Aid for International Development’s (USAID) Equitable AI Challenge.
The project sought to evaluate gender bias in AI Applications in measuring wealth using household survey data.
Data from the Ghana Demographic and Health Survey, along with satellite imagery were used to construct the module.
Findings of the study are intended to inform more accurate poverty estimates that take gender into account when allocating resources, undertaking impact evaluations and informing policy.
Dr Edem Selormey, Director of Research, CDD-Ghana, speaking at a workshop in Accra to launch the study report, reiterated that AI Applications had transcended various domains, ushering in a new era of efficiency and innovative solutions to complex problems; hence understanding and addressing the implications of AI on society had become ever more imperative.
“As we integrate AI into more facets of our lives, it is essential to critically examine its impact on different segments of our society, particularly on gender disparities,” she said.
“Recent research has shed light on instances where AI systems inadvertently perpetuate ethnic, disability and gender biases, therefore amplifying societal I qualities.”
Dr Selormey noted that over the past year, AidData, in collaboration CDD-Ghana, utilising geospatial data and the Demographic and Health Surveys (DHS) data as its foundation, had looked into the nuances of gender bias in wealth estimates generated by AI.
She said the project team analysed the intricate relationship between AI models, wealth and gender dynamics; adding that they used AI’s capacity to discern and classify to reveal biases that might lie hidden with the DHS data.
She said the gender lens through which the project had been meticulously executed exposes the disparities that might evade conventional analyses; saying “for instance, we noticed that digital realm can mirror and amplify the inqualities ingrained in our societies”.
She said gender bias was insidiously woven into the algorithms that underpin AI system.
She cited that the study found that models trained on male household data consistently outperform their female counterparts.
“This study marks only the beginning; I believe the methods and findings presented today will lay the groundwork for others to continue by unraveling more questions and answers toward a future where AI’s power is harnessed for the betterment of all; regardless of gender and where AI becomes a driving force for empowerment rather than one that perpetuate disparities,” the Director said.
Dr Rachel Sayers, Research Scientist, AidData, in her presentation, said the study also considered how gender-specific wealth indexes compare to household wealth index used by the DHS.
She said the study found that DHS wealth index overestimated the wealth of the poorest female households, as compared to a female-specific wealth index.
Dr Sayers said the study also found that the DHS wealth index underestimated the wealth of the poorest male households, as compared to a male-specific wealth index.
Dr Rita Udor, Gender Inclusive Officer, Responsible Artificial Intelligence Lab (RAIL), Kwame Nkrumah University of Science and Technology, advocated the inclusion of women on the boards of AI developers to help improve the quality of decision-making as part of efforts to ensure that all segment of society enjoy the full benefits of technology.
Madam Deborah Dormah Kanubala, Machine Learning (ML) Researcher, Saarland University, Germany, in virtual presentation, said AI played a very critical role in the medical field such as its usage in quick diagnosis of breast cancer.
GNA