By Guest Writer
Artificial intelligence and machine learning based solutions hold a promise of becoming a transformative force in multiple facets of humanitarian and development work. However, alongside this lies the potential of negative implications in the form of furthering entrenched biases and exacerbating inequality.
The the Gender Equitable AI Toolkit from NetHope presents principles developed to provide a framework for ensuring gender equitable AI. The principles laid out in this toolkit are sourced from a synthesis of input from the NetHope Member and Partner community.
NetHope identified 5 sets of principles for gender equitable AI development in the following areas:
- Fairness and inclusivity.
- Transparency.
- Design and development.
- Governance and autonomy.
- Collaboration and capacity building.
To mitigate risk, it is critical that implementing organizations adopt principles – like the above – and thereby champion gender equity. At the core of these intertwined principles is the concept of inclusiveness and each following principle serves to promote the visibility of underrepresented gender groups.
Principles of Fairness and Inclusivity
Embracing fairness and inclusivity principles enable organizations to tackle biases, combat discrimination, and advance inclusive technology solutions. This entails considering the impact of AI on priority gender groups, involving diverse perspectives in decision-making, and ensuring for representative data, algorithms, and models.
Inclusive AI solutions are characterized by empathy, cultural sensitivity, and a contextualized understanding of human complexities. They should also aim to bridge existing digital divides, empower marginalized communities, and foster social justice. By championing inclusivity, organizations foster a digital ecosystem that distributes AI benefits in an equitable manner, and that no one is left behind.
1. Accessibility
AI initiatives must tackle challenges related to equitable access to digital resources. This involves bridging the gap for individuals across the globe, especially those in underserved regions. The overarching goal is to protect the rights of individuals and prevent harm in the design and deployment of AI solutions.
2. Gender-Centricity
AI initiatives should have a dual objective to first support existing program work as well as to actively address existing inequities by considering historical disparities in humanitarian aid.
By establishing comprehensive metrics and objectives that effectively advance both facets, organizations can systematically measure the impact of AI on programmatic effectiveness and gender equity, thereby fostering fairness in the conception and execution of AI technologies.
3. Bias Mitigation
AI initiatives should acknowledge the inherent biases present within culture and shared languages thus dictating the creation of data sources or algorithms which can perpetuate and reinforce gender stereotypes. It underscores the need to take proactive measures in countering an inherently biased world.
Recognizing that language itself can carry historical and societal biases, the principle emphasizes the significance of addressing the potential biases ingrained in the data used to train AI models.
Principles for Design and Development
Design and development principles guide the creation of AI systems that are accessible, unbiased, and responsive to the needs and aspirations of diverse individuals and communities. By embracing these principles, organizations can ensure that AI technologies are designed with empathy, cultural sensitivity, and the deliberate inclusion of diverse perspectives.
This approach fosters the development of AI systems that are impactful and capable of addressing societal challenges while leaving no one behind.
1. Gender Centered Design
AI initiatives must center solutions around the needs, perspectives, and leadership
of priority gender groups at the local level. Participatory design, in this context, involves ensuring that the voices and experiences of gender are represented in decision-making processes.
Organizations should actively engage with communities, seeking their input, feedback, and co-creation throughout the design and development of AI technologies. By embracing participatory design, AI initiatives empower communities to influence the technology landscape resulting in solutions that genuinely meet their aspirations and needs.
2. Intersectionality
AI initiatives should recognize that gender is a critical aspect of an individual’s identity, and that like any human group, deserves dignity and visibility in digital ecosystems. This principle emphasizes moving beyond a monolithic understanding of gender and pushes practitioners to consider gender identity during data collection and curation.
This entails intentionally implementing data collection methods that seek out diverse perspectives
and experiences, ensuring comprehensive representation. By embracing an intersectional lens in data collection organizations can ensure that AI models are built upon an equitable foundation of representative data sources.
3. Model Awareness
AI initiatives should emphasize the importance of training models in ways that raise awareness of fairness, inclusivity, and equal representation of diverse gender groups. They should acknowledge the potential biases in data and algorithms that can perpetuate gender inequalities and take steps to mitigate these biases during the training process.
By incorporating conscious model awareness, AI initiatives enhance the ethical and social responsibility of their systems, ensuring that AI technologies contribute positively to gender equity and broader social progress.
Source, ictworks.org
About The Author
