TL;DR
This paper argues that building trustworthy AI requires moving beyond traditional metrics like transparency and fairness to actively incorporate principles of justice, power redistribution, and feminist, non-exploitative participatory design. It proposes a "queer epistemology" for AI development and regulation, emphasizing the need to center the experiences and needs of marginalized communities, particularly queer, gender-diverse, and trans individuals, to create AI that is truly inclusive and beneficial.
What Problem Does This Paper Address?
The paper addresses the critical issue of lost and absent trust in AI systems. It highlights that historical exclusion of marginalized groups from AI development has resulted in technologies that are not only less useful but actively harmful to these communities. The authors contend that current approaches to AI trustworthiness, focusing solely on transparency, fairness, and accountability, are insufficient. They argue for a more radical shift that prioritizes justice and power-sharing, and specifically calls for the integration of queer, gender-diverse, and trans perspectives throughout the entire AI lifecycle, from design to regulation.
Methods
The paper doesn't present a traditional empirical study with novel algorithms or datasets. Instead, its methodology is primarily theoretical and advocacy-oriented. It draws on:
- Feminist, Queer, and Postcolonial Theory: To critique existing AI paradigms and propose alternative frameworks.
- Participatory Design Principles: Emphasizing the need for deep engagement with affected communities, moving beyond superficial "participation-washing."
- Expert Knowledge from Queer in AI: Leveraging the organization's years of advocacy and research on queer issues in AI.
- Policy Analysis: Examining existing and emerging laws and policies related to AI, gender, and privacy.
- Multilateral Conversation: Advocating for dialogue between technologists, policymakers, and diverse communities.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Trustworthy AI is More Than Transparency: True trustworthiness requires actively addressing power imbalances and ensuring justice, not just making systems understandable or fair on the surface.
- Marginalized Voices are Essential: AI development that excludes or marginalizes communities will inevitably produce harmful or ineffective technologies. Empowering organizations like Queer in AI is crucial for building a more inclusive field.
- "Queer Epistemology" for AI: This framework suggests that AI development and regulation should be informed by the unique insights and experiences of queer, gender-diverse, and trans individuals, challenging cisnormative assumptions and promoting a more nuanced understanding of identity and data.
- Participatory Design Needs Depth: Superficial engagement with communities is insufficient. True participatory design must be feminist, non-exploitative, and address power dynamics to avoid further marginalization.
- Privacy is Foundational: For queer, gender-diverse, and trans individuals, privacy is not just a feature but a fundamental requirement for safe and equitable engagement with AI.
- Beyond Binary Gender: AI systems often rely on binary gender assumptions, which are inherently harmful and exclusionary to non-binary individuals. This necessitates significant structural changes in how AI is designed and deployed.
- Grassroots Efforts are Key: Empowering community-led initiatives is vital for developing AI that is truly accountable and responsive to the needs of marginalized groups.
Key Results
The paper presents a series of recommendations and a proposed framework rather than quantitative results. Key takeaways include:
- Recommendations for Design:
- Design with and for queer, gender-diverse, trans-specific privacy considerations.
- Develop features for queer, gender-diverse, trans-specific purposes.
- Create representative and gender-affirming AI personas.
- Base development in grounded and participatory design processes.
- Recommendations for Policy:
- Analyze legislation through a gender lens to promote equality.
- Develop research-based, gender-specific policy recommendations.
- Foster collaboration between policymakers, technologists, and researchers.
- Adopt an intersectional approach to policy development.
- Policy Options for Privacy and Data Protection:
- Directly prohibit or monitor algorithmic discrimination.
- Implement accountability measures like transparency, explainability, risk assessments, and audits.
Limitations and Future Work
- Structural Barriers: The authors acknowledge that while incorporating queer perspectives is crucial, it's unlikely to fundamentally change power distribution within the current profit-driven AI industry without more substantive structural changes.
- "Pinkwashing" and Surveillance Traps: Even well-intentioned efforts can be susceptible to "pinkwashing" or become surveillance traps if not carefully designed with privacy and genuine community empowerment at their core.
- Need for Deeper Engagement: The paper calls for researchers and designers to go beyond current structures and work towards developing grassroots AI accountable to communities, while also disrupting harmful existing systems.
- Future Research Directions: The authors suggest further research into:
- Analyzing legislation and public policy related to AI and gender equality.
- Exploring existing and emerging policies concerning AI and gender, particularly in data and privacy, technological design, and labor.
- Conducting theoretical analysis and multi-lateral conversations to understand how theory can speak to practice in AI design.
- Developing research-based tools for designers and technologists to embed pluralism and inclusion into AI systems.
Summary updated