From Algorithms to Accountability: Global AI Governance Needed

The International Telecommunication Union (ITU) is a specialized agency of the United Nations.

Recent research from Stanford’s Institute for Human-Centered AI warns that bias in artificial intelligence (AI) remains deeply rooted, even in models designed to avoid it, and can worsen as models grow. From bias in hiring men over women for leadership roles to misclassifying darker-skinned individuals as criminals, the stakes are high.

However, annual dialogues and multilateral processes, such as those outlined in UN Resolution A/RES/79/325, cannot keep pace with AI technological developments—and the cost of this lag is high. To enhance accountability, some experts suggest giving tech companies, whose operations now resemble state functions, participatory roles at the UN General Assembly.

In a landmark case, the plaintiff alleged that Workday’s AI-based applicant recommendation system violated U.S. federal antidiscrimination laws due to disparate impacts based on race, age, and disability. In July 2024, Judge Rita F. Lin ruled that AI vendors could be held directly responsible for discriminatory outcomes, not just the employers using the tools.

Similarly, University of Washington researchers found significant racial, gender, and intersectional bias in how three large language models ranked resumes, favoring white-associated names over equally qualified candidates with other racial associations.

The financial impact is substantial. A 2024 DataRobot survey of over 350 companies revealed that 62% lost revenue due to biased AI decisions, showing that discriminatory AI is not just a moral failure but a business risk.

A 2024 Stanford analysis of vision-language models found that larger models were up to 69% more likely to label Black and Latino men as criminals. In language models, women were more often associated with humanities than STEM, men were favored for leadership roles, and negative terms were disproportionately linked to Black individuals.

To prevent AI discrimination, the UN must lead by example and work with governments, tech companies, and civil society to establish global guardrails for ethical AI. Suggested actions include:

Working With Tech Companies: Treat tech firms as state-like entities with participatory privileges to ensure accountability.

Developing Clear Guidelines: Push for global standards on ethical AI, including inclusive data collection, transparency, and human oversight.

Promoting Inclusive Participation: Ensure AI policy development reflects the diversity of the world, including voices from the Global South, women, and underrepresented groups.

Human Rights Impact Assessments: Evaluate AI systems for potential human rights violations before deployment.

Holding Developers Accountable: Create an AI Accountability Tribunal within the UN to address discrimination and enforce penalties.

Digital Literacy and Rights Education: Equip policymakers and citizens to understand AI impacts and protect their rights.

Mandates for Intersectional Bias Audits: Require audits for combined biases in AI systems and fund open-source audit tools globally.

AI is neither inherently good nor bad. Its impact depends on how we use it. Without careful oversight, AI could exacerbate inequalities and create harder-to-detect forms of discrimination. But if human rights are prioritized, AI can become a tool to uplift, rather than exclude.

The UN remains uniquely positioned to lead this effort. Ethical AI governance is essential for safeguarding both human dignity and global fairness.

Chimdi Chukwukere is an advocate for digital justice, exploring technology, governance, Big Tech, and social justice. She holds a Masters in Diplomacy and International Relations from Seton Hall University.