Doing the Feminist Work in AI: Reflections from an AI Project in Latin America

Research Background and Issues

• Issues and Challenges:

  1. Current AI development is primarily led by large corporations, lacking diversity and concentrated in the Global North, particularly employing a "logic of extraction" in the Global South.
  2. The authors focus on the critical issue of missing judicial data related to gender-based violence.
  3. The lack of transparency in gender-based violence data, coupled with mistrust in the judicial system and low reporting rates, further limits the understanding and response to this issue.

• Significance of the Issue:

  1. Gender-based violence is a widespread global issue, particularly in Latin America, where low reporting rates and lack of associated data further hinder policymaking and public understanding.
  2. There are currently few examples of critical AI development practices originating from the Global South, especially Latin America, highlighting the importance of showcasing endogenous knowledge from the region.

• Research Motivation and Related Work:

  1. The authors are inspired by the powerful feminist movements and long-standing social justice movements in Latin America, such as "Ni Una Menos" (Not One Less) and the Green Wave (Marea Verde).
  2. Technology development in Latin America is often influenced by technological colonialism, necessitating independent approaches rooted in regional perspectives.

Solution

• Methods and Solutions:

  1. The authors introduce a Latin American feminist AI development project, "AymurAI." This project developed an AI-driven tool to anonymize court data and generate datasets on gender-based violence.
  2. The project employs natural language processing (NLP) techniques to semi-automatically extract content from legal rulings and anonymize it.

• Innovative Features:

  1. Integrating a Latin American feminist perspective throughout the technology development process to counteract technological centralization and technosolutionism.
  2. Utilizing the Collaborative Autoethnography (CAE) method to reflect on the AI development process, bridging theory and practice.
  3. Emphasizing transparent, horizontal collaboration to ethically construct AI systems.

• Specific Implementation Steps and Technologies:

  1. Data Processing: Court data undergoes anonymization (court personnel replace all personal data with fabricated data).
  2. Model Development: The Flair framework and BiLSTM-CRF architecture are employed, utilizing Spanish-language corpora (BETO embeddings).
  3. System Design: The system targets low-resource environments, adopting a cross-platform, localized, lightweight architecture to ensure functionality in resource-limited court settings.
  4. Ethical Security: The code is open-source, allowing for audits; the development process avoids the transmission of real personal data entirely.

Research Outcomes

• Specific Achievements:

  1. Providing a tool to enhance transparency in gender-based violence data and support policymaking.
  2. Demonstrating the possibilities of feminist work in AI development, particularly through autonomous experiences in Latin America.

• Advantages Over Existing Solutions:

  1. Emphasizing ethics and collaboration throughout the development process, offering a counterbalance to "technology replacing humans" strategies.
  2. Proposing localized and differentiated approaches to technology development, balancing theory and practice.

• Experimental or Evaluation Results:

  1. Through collaborative autoethnography, the learning and challenges faced by team members were revealed.
  2. The tool received positive feedback and was proven to reduce the workload of judges and staff.

• Limitations and Future Directions:

  1. Limitations:
     • Dependence on limited annotated data, posing challenges for ML model performance.
     • Funding primarily sourced from the Global North, highlighting the lack of resources within Latin America.
  2. Future Directions:
     • Promoting new AI development methods based on feminism and social justice.
     • Exploring ways for AI development in Latin America and the Global South to break free from colonial technological logic and achieve genuine autonomous knowledge production.

Conclusion

Using the "AymurAI" project as an example, this paper demonstrates how feminist perspectives and practices can inform the development and application of AI systems. In the Latin American context, this project not only challenges the AI development narrative dominated by the Global North but also proposes new pathways to counter systemic technosolutionism, sparking extensive discussions on ethical and equitable technology platform construction. This experience-based and collaborative practice offers significant inspiration for researchers and social activists in other fields.