Fairness for Unobserved Characteristics: Insights from Technological Impacts on Queer Communities (2102.04257v3)

Abstract: Advances in algorithmic fairness have largely omitted sexual orientation and gender identity. We explore queer concerns in privacy, censorship, language, online safety, health, and employment to study the positive and negative effects of artificial intelligence on queer communities. These issues underscore the need for new directions in fairness research that take into account a multiplicity of considerations, from privacy preservation, context sensitivity and process fairness, to an awareness of sociotechnical impact and the increasingly important role of inclusive and participatory research processes. Most current approaches for algorithmic fairness assume that the target characteristics for fairness--frequently, race and legal gender--can be observed or recorded. Sexual orientation and gender identity are prototypical instances of unobserved characteristics, which are frequently missing, unknown or fundamentally unmeasurable. This paper highlights the importance of developing new approaches for algorithmic fairness that break away from the prevailing assumption of observed characteristics.

Fairness for Unobserved Characteristics: Insights from Technological Impacts on Queer Communities

This paper, authored by Nenad Tomasev, Kevin R. McKee, Jackie Kay, and Shakir Mohamed, addresses the critical need to broaden the scope of algorithmic fairness to include unobserved characteristics such as sexual orientation and gender identity. The authors argue that current methodologies predominantly focus on observable traits like race and legal gender, necessitating new frameworks for characteristics that are frequently unrecorded or fundamentally unmeasurable.

The omission of sexual orientation and gender identity from fairness discussions in ML has led to significant gaps in addressing the full spectrum of biases that AI systems can propagate. Despite the ethical importance, the constraints associated with data collection on these characteristics, including privacy concerns and legal limitations, exacerbate the challenge. The paper explores the potential positive impacts of AI on queer communities while simultaneously acknowledging the substantial risks.

Main Contributions

1. Technological Empowerment and Risks: The paper meticulously catalogs how AI can support queer communities, enhancing their online safety and health while also recognizing the risks stemming from systems prone to biased data, mishandling privacy, and intentional misuse. For instance, models that predict sexual orientation from images, despite methodological flaws, suggest potential risks if applied with better data and more sophisticated algorithms.
2. Algorithmic Censorship and Misinformation: The dual role of AI in moderating content is highlighted with particular concern for inadvertent censorship of queer content under well-meaning but ultimately flawed algorithms. This underscores the broader implications of algorithmic governance and digital autonomy.
3. Unique Considerations for Language and Communication: There is considerable focus on developing LLMs that respect the nuances of gender identity and sexual orientation, emphasizing a shift toward inclusive language that surpasses mere avoidance of derogatory terms.
4. Health and Mental Health Implications: AI holds the potential to revolutionize healthcare delivery for queer individuals by improving targeted healthcare interventions and prospective diagnostics. However, concerns about fairness and accuracy in the absence of sexual orientation or gender identity data emphasize the limitations and potential for unequal benefit distribution from AI development in healthcare.
5. Novel Fairness Frameworks: The authors advance the notion that tackling fairness for unobserved characteristics requires a departure from traditional demographic parity approaches, proposing instead more nuanced frameworks like individual, counterfactual, and distributional fairness.

Implications and Future Directions

The authors call for the expansion of fairness research to meaningfully incorporate the experiences and needs of queer communities. Central to this is the involvement of queer voices in AI development processes, advocating for an approach that is both inclusive and participative. The intersection between fairness and privacy emerges as a key theme, demanding that solutions honor both dimensions to protect individuals from potential privacy infringements.

The framework proposed in this paper has profound implications for future AI systems that aspire to be equitable. Solutions must be adaptable to account for the dynamic and often fluid nature of identity, which challenges conventional data-driven approaches and necessitates a deeper engagement with ethical frameworks and participatory research methodologies.

Conclusion

The paper contributes vital insights into the nascent discourse on queer fairness in AI, addressing both theoretical and practical concerns. By foregrounding the experiences of the queer community, it challenges researchers to rethink how fairness can be achieved for characteristics that elude quantification. The call for robust frameworks that address unobserved characteristics is essential for ensuring that the benefits of AI are distributed equitably, supporting diverse communities in a manner that is ethical and just. As AI technology continues to evolve, the insights from this paper will be critical in guiding future developments in fairness research that are inclusive of all identities.