< Incubating Feminist AI > From paper to prototype

Papers go through 3 stages of development

PAPER ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯   PROTOTYPE  ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯  PILOT

We are so incredibly proud of our Alliance’s FA+IR Feminist AI Research Network cohorts of projects, which include papers and now a series of prototypes that emanate from applied research funded by the IDRC.

f<A+I>r aims to support the skill and imagination of Global South/Majority World feminists in producing effective, innovative, interdisciplinary models that harness emerging technologies which correct for real life bias and barriers to women’s rights, representation and equality.

Ongoing Pilots | Cohort 4

Team: Ivana Feldfeber, Yasmín Belén Quiroga, Marianela Ciolfi Felice (Data Género, Argentina)

AymurAI is an open-source software that helps criminal courts collect open data on GBV from court rulings in Latin America.

Gender-based violence takes on many forms: physical, psychological, and economical among others. While the harms are known, the impact of GBV has not yet adequately been measured. The absence of this data in Latin America has led, in part, to the championing of open data initiatives in the judiciary. When gathered this data could be used in machine learning to identify the patterns of violence that might ultimately lead to feminicide – and then to policy and potential remedies to hinder the violence and death.

AymurAI as a paper:  “Feminisms in Artificial Intelligence: Automation Tools Towards A Feminist Judiciary Reform in Argentina and Mexico” 

AymurAI as a prototype:  “Prototype for an open and gender-sensitive justice in Latin America – AymurAI” — 

Team: Hazel T. Biana, Rosallia Domingo, Homer J. Yabut, Melvin A. Jabar, Kevin Lazarito and Shaina Dimaapi (De La Salle University, The Philippines) 

Focused on transport systems in Metro Manila,  SafeHer is simultaneously a women’s empowerment and AI safety app. Focused on transport systems in Metro Manila in the Philippines, SafeHer challenges societal norms, raises awareness, and will potentially influence policy changes concerning women’s safety and community action. 

SafeHer as a paper: Strong Girl AI: Frameworks for the Empowered Mobility of Women in Southeast Asia” focused on transport systems in Metro Manila,  which have been hailed as dangerous and unsafe for women.

SafeHer as a prototype:  The prototype features AI-powered functionalities, including SOS Alert, Nearby female commuter identification, live location sharing, Medical ID, User Verification and Invitation, and scream detection. It takes into account women’s– socio-cultural, psychological, personal, and spatial factors for effective urban transit safety.

Team: Fundación Datos Protegidos and Observatorio de Estadísticas de Género e Interseccionalidades, Chile 

SOF+IA supported with Generative Artificial Intelligence (AI) is a chatbot that is able to dialogue and address situations of violence and digital harassment that women, of any age, often experience when expressing themselves, giving opinions or when uploading content on digital social media platforms.

SOF+IA as a paper: Digital Gendered Violence in Chile: Development of a system for report, monitoring and response-orientation based on a feminist chatbot”

SOF+IA as a prototype: reporting and guidance information system for technology-facilitated gender violence (tfGBV) situations in Chile.  It is based on a conversational agent (‘chatbot’ type) using feminist principles that consider ethical issues and put the needs and context of women who are exercising the right to freedom of expression and opinion on social networking platforms –  particularly women with a public voice –  activists, academics, women involved in politics, among others who live this type of situations daily – at the center of the prototype.”SOF+IA” was named after a public consultation in social media. 

Team: Laura Alonso Alemany, Luciana Benotti, Beatriz Busaniche (Fundación Vía Libre, Argentina)

E.D.I.A. is a toolkit that allows people without technical expertise, but with lived experience to explore, characterize and audit biases and stereotypes in language models. During this newest pilot phase E.D.I.A. will co-create then publish structured content and teaching materials so that E.D.I.A. methods can be replicated for other languages and contexts. 

E.D.I.A. as a paper:  A tool to overcome technical barriers for bias assessment in human language technologies”

E.D.I.A., as a prototype model,was co-created with teachers, workshops where conducted and structured content and teaching materials were developed for their 2023 workshops. 

Ongoing Prototypes

Team:  Técnicas Rudas, Mexico & Diversa Studio, Ecuador

In this prototype phase, Técnicas Rudas and Diversa Studio will make use of the field data generated, which formed the basis of  their paper “Indigenous Natural Resources Governance:  Indigenous rights, water conservation & AI Exploring active participation of the Yaqui community in water resource management”. 

Water Governance project conducted in collaboration with the Yaqui community of Vícam unveiled the challenges inherent in water management in Mexico. The methodology and amassed image, audio, satellite and analyses can now serve as a foundational resource for devising strategies and policies that advocate for sustainable water management while upholding the traditions and values of indigenous communities.

Working in collaboration, Diversa Studio leads the development of all the technical analysis of AI as well as data analysis while Técnicas Rudas guides the workshops and field engagement with the Yaqui community and across Sonora territories. 

Team: Marwa Soudi, Esraa Ali, Maha Bali and Nihal Mabrouk (IdeasGym, Egypt)

The ongoing prototype is a platform that combines the expertise of teachers with the capabilities of AI technology to enhance the educational experience through resource allocation to underprivileged schools, particularly under-resourced one room community-based schools that have 80% girl students.  The prototype emerges out of the team’s paper: “Human-Centered Artificial Intelligence Tutoring Systems.”

A human-centered approach to intelligent tutoring systems focuses on needs + preferences of students + teachers in Upper Egypt’s  one room community schools populated predominantly by girls.  By tailoring the system to the specific environment of under-resourced schools, there is potential to reduce dropout rates and improve learning outcomes for students. Key considerations include system alignment with the localized context;  utilizing GenAI for content creation in different languages; emphasizing the importance of human oversight. The proposed AI-based tutoring system is positioned as an assistant, enhancing teachers’ capabilities rather than replacing teacher-student interaction., acknowledging the irreplaceable role of human teachers in providing motivation, guidance, and dialogue, harnessing the potential of AI for education while ensuring a human-centered approach. 

Living Prototype & Papers

How can AI Crowd work become more feminist and fair? In order to function, algorithms require millions of labeled training data to learn, recognize and categorize information. This labeling is often done by crowd workers who are neglected in system designs and whose well being is seldom considered. Latin American and Caribbean workers have been identified as significant crowd work contributors depending on crowd work as a steady source of income. However most studies on crowd work center around Western women and fail to take into account the personal and professional advancement of Latin American women.

To facilitate communication between these women crowd workers and help them identify relevant conversations, an intelligent a conversational agent was developed to emulate the personality of Latin American heroines and assist users in searching for specific advice, articulating their interests, and navigating the platform. Furthermore, the conversational agent will recommend other crowd working women who might be valuable connections based on shared interests, expertise, and experiences.

A practical approach, with a feminist perspective located in Latin America, focused on the development of Artificial Intelligence (AI). It questions if it is possible to develop AI that does not reproduce logics of oppression. To answer, they focus on the power relations immersed in the field of AI and make an interpretative analysis of the day-to-day experiences of seven women working in some field of AI or data science in the region, in dialogue with different statements of feminist principles and guidelines for the development and deployment of digital technologies. The prototype turns the paper into a living document and new methodology. Workshops will explore deepening of the basic guide of questions from the initial paper, and practices for development with projects actively in development.  

Read the full paper in English  | Read the full paper in Spanish 

Reimagining automated Violence Interventions Through Participatory Technology Design (Point of View & Digital Futures Lab & Design Beku, India)

Point of View launched TechSakhi, an informational helpline service on digital safety and security, to address the gendered experiences of marginalized women, girls, and queer individuals in grassroots communities across India. The helpline, operational since February 2022, provides information, step-by-step guidance, referrals, and support in Hindi and Bangla languages. Callers connect with one of seven available responders through a cloud-based telephony system, ensuring privacy. The system includes an automatic ticketing system for efficiency and a knowledge database for standardization. The system utilizes AI powered technologies to facilitate intelligent triage of the cases including assisting with image and sound processing as well as AI powered searches.

Analyzing Public Procurement Anomalies in Ecuador with a Gender Perspective (Datalat,Ecuador)

A study of the regulations issued in Latin America related to public procurement and the gender approach. Using focus groups with women leaders from economy enterprises and microenterprises to identify barriers to participation in public procurement processes, and potential use of technological tools and artificial intelligence applications. The paper recommends a participatory approach to identifying barriers to entry and addressing gaps in data to develop artificial intelligence systems that ultimately address unequal participation in public procurement processes.

Read English paper  |  Read Spanish paper 

An exploration of how AI and a Digital Public Good for Gender Justice could contribute to better outcomes for Girls (PEKKA, Indonesia)

This paper focused on the use of technology to enhance capabilities of grassroots paralegals to deliver concrete, relevant, and persuasive information to parents and young people about the risks of child marriage. Information is intended to be used in advocacy efforts and to support policy-makers in establishing evidence-based and gender-responsive policies. A critical challenge in addressing child marriage in Indonesia is the scarcity of data, especially considering that a significant number of marriages occur outside the official state mechanism. The absence of published, disaggregated, and gender-responsive data contributes to a lack of targeted interventions by policy-makers and advocates. The envisioned prototype offers a gender-responsive and comprehensive data collection method, facilitating the reproduction and visualization of this data for stakeholders, including PEKKA Paralegals and policy-makers.

Feminist NLP: An Annotated Corpus to Evaluate Sex Differences in Work Related Diseases in Chile (Catholic University of Chile,Chile)​

Evidence from the US and Europe has highlighted significant differences in work accidents and occupational disease rates between women and men. However, such gender-disaggregated data is lacking for countries in the global south. In collaboration with the Chilean Safety Association (ACHS), which extends occupational health services to half the contract employees in the country, this paper focuses on two entities: the agent causing the accident (e.g., a walking surface or stairs) and the mechanism (e.g., a fall or a blow). 

When AI is not the Solution: Towards Integral Co-Design Processes (By Sofía Trejo, Hamlet Antonio García Zúñiga, Ivan Vladimir Meza Ruiz, and Gaby León Ortiz)

Artificial Intelligence (AI) is currently promoted as a tool for the development of numerous knowledge areas and as a possible solution to several global issues, from gender inequality to climate change. However, data requirements, computing infrastructures, and access to specialized knowledge have led to these systems being developed, to a large extent, in the private sector or in collaboration with it.

The concentration of AI development on these actors has led to the production of technology that emulates the ideologies, practices, and issues of the contexts and groups that develop them. In particular, the hegemonic AI systems we have today reproduce colonialism and extractivism, in addition to operating under a power centralization logic, where its profits concentrate in a few hands. Therefore, it is common for AI systems not to align with the context and needs of the populations on which they operate, which are generally not included in the various stages of their life cycle, from design to operation.

Design of Data Science Projects for Inclusive Public Policies (Virginia Brussa, Maria Paz Hermosilla Cornejo, Jesus Santorcuato Munoz)

This study aimed to reformulate and validate changes to a design methodology for data science projects for public officials based on collective contributions. The analysis integrates approaches from the field of data justice, intersectionality and public design justice. Promoting the critical implementation of data science in the public arena based on awareness of structural changes in public decision-making is a non-technical instrument and an alternative technology for rethinking AI and inclusive data science. Efforts to bridge the existing data gaps, reconcile public innovation agendas and feminist AI, and raise awareness of new rights and knowledge matrices are elements taken into account to review the design of public data-intensive projects.