Page 46 - AI for Good - Impact Report

AI for Good

Caitlin Kraft Buchman from Women at The Table shares recommendations on ensuring gender
equality in AI development and deployment (2024):

• Construct large new unbiased datasets with a focus not only on quantity but on quality for
the public good. It is important to actively produce open, gender disaggregated datasets;
which will better enable an understanding of the sources of bias in AI, and ultimately
improve the performance of machine learning systems. We need to Invest in controls
to oversee data collection processes and human-in-the-loop verification, so that data is
not collected at the expense of women and other traditionally excluded groups. And of
course, it is also vital to engage in more inclusive data collection processes that focus,
again, not only on quantity but on quality of datasets.
• Pilot AI that allocates 21st century social protection, subsidies, and scholarships where
women and girls have traditionally been left behind. Encourage public institutions to
innovate and lead in this domain. We need to be creative with small, targeted, impactful
pilots based on social science research that allocate social incentives, subsidies, or
scholarships where women have traditionally been excluded in prior systems. This is a
positive agenda to advance values of equality we have long embraced, to correct for
the visibility, quality, and influence of women proportionate to the population. STEM
education alone will not get us where we want to go.
• Enact gender responsive public procurement guidelines for organizations and all levels
of government with hard targets and the outline of roles and responsibilities of those
organizations required to apply these principles. This could jumpstart new industries
and value creation, invented and owned by women and girls, expanding definitions of
‘expertise’ so that those with lived experience and communities affected by technologies
can influence the design, deployment, and control of new technologies.
• Mandate algorithmic impact assessments with an integrated approach, and holistically
include gender, human rights, and environmental impact. These assessments need to
be done beforehand and continuously throughout the lifecycle of the system. We need
rigorous testing across the lifecycle and this testing should account for the origins and
use of training data, test data, models, Application Program Interface (APIs), and other
components over the product lifecycle. AI should improve the quality of, not control, the
human experience.
• Enshrine the public’s right to know the systems that impact their lives if algorithmic
decisions have been made that affect an individual and that this right includes continuous
consent and ends with contestability of the systems.

36