How AI leaves minority communities behind

Artificial intelligence is increasingly shaping decisions that affect everyday life. Governments use it to process immigration applications, distribute welfare, support policing, and improve public services, while businesses rely on it to screen job applicants, assess loan applications and recommend products. Humanitarian organizations have also begun integrating AI into aid distribution and digital identity systems. Yet AI is only as representative as the data it learns from. Built on vast amounts of information gathered from books, websites, government records, healthcare systems and financial databases, AI reflects the societies that produced those records. Many minority communities, including Indigenous peoples, refugees, stateless populations, and speakers of minority languages, have historically been underrepresented or excluded from official data. As AI becomes increasingly responsible for decisions that shape access to services and opportunities, a growing concern is not simply how these systems make decisions, but whether they recognize communities that have long existed at the margins of official records.

The minorities AI cannot see

Artificial intelligence increasingly assumes people possess stable legal identities, birth certificates, passports, tax histories and banking records. These assumptions reflect the realities of citizens living within established administrative systems. Millions of people do not fit those assumptions. Stateless populations, displaced communities, refugees, and undocumented migrants often live for years without consistent documentation despite working, studying and participating in society. As governments expand digital identity systems and automated verification processes, access to employment, healthcare, financial services, and humanitarian assistance increasingly depends on appearing within official databases. For these communities, exclusion can occur simply because digital systems cannot verify identities that have never been comprehensively recorded. Recognition itself is becoming dependent on data. For minority communities that have historically existed outside formal administrative systems, digital invisibility risks becoming a new form of exclusion.

When artificial intelligence becomes a tool of surveillance

Artificial intelligence can also amplify discrimination when deployed by governments. In studies on AI and minority rights, it is argued that AI possesses a dual nature. Depending on how it is governed, it can either strengthen minority rights or reinforce discrimination and exclusion. The technology itself is neither inherently beneficial nor inherently harmful. Its impact depends on the political institutions and human decisions surrounding its use. Perhaps the most widely cited example is China's treatment of the Uyghur population in Xinjiang. Authorities have combined facial recognition technology, biometric databases, DNA records, fingerprints, iris scans, voiceprints, banking information, healthcare records and police databases into the Integrated Joint Operations Platform (IJOP), an AI-assisted surveillance system designed to identify individuals considered potential security threats. According to the paper, the platform does more than monitor behavior. It uses artificial intelligence to process enormous quantities of personal information while helping authorities identify members of ethnic minorities. The U.S. Treasury has described the system as one of the first examples of governments using AI for racial profiling because it can identify Uyghurs based on physical characteristics and continuously monitor their movements. The significance extends beyond China. This is an example of how AI can transform discrimination from a human process into an automated one capable of operating continuously and at unprecedented scale.

Who controls the data?

A relatively small number of technology companies, including OpenAI, Google, Microsoft, Meta, Amazon and Anthropic, possess the computing infrastructure, proprietary datasets and financial resources required to build the world's most advanced AI systems. Their datasets increasingly extend beyond information generated by their own platforms. Over the past several years, companies such as OpenAI, Google, Anthropic and Perplexity have signed licensing agreements with newspapers, magazines, and image libraries, allowing AI systems to learn from professionally produced journalism and other trusted information. These agreements create new revenue opportunities for publishers, but they also contribute to a new information economy in which a relatively small number of companies control access to much of the information used to train advanced AI systems. This concentration matters because whoever controls the data increasingly influences what AI knows and which histories, cultures and communities become visible within future technologies. Minority communities whose histories, languages and experiences generate comparatively little digital content risk remaining underrepresented in these systems.

Digital languages, inequality, and inequality

Minority exclusion extends beyond race and citizenship to language itself. The concept of digital minoritisation describes how languages that are widely spoken in everyday life can become digitally marginalized because AI systems are overwhelmingly trained on English and a handful of dominant languages. This has significant consequences. Communities speaking Indigenous or minority languages often receive poorer machine translations, less accurate speech recognition and fewer AI-powered educational resources. As AI becomes increasingly integrated into search engines, virtual assistants and public services, language itself risks becoming another dimension of digital inequality. At the same time, AI also offers opportunities for preservation. At the same time AI can support language revitalization by helping document endangered languages, improve translation systems and expand access to digital tools for communities that have historically lacked technological representation. Governments such as Iceland have already partnered with AI developers to strengthen digital support for their national language, demonstrating that AI can also contribute to cultural preservation when inclusion is intentionally prioritized.

Artificial intelligence will continue transforming societies over the coming decades. The question is not whether AI will become more capable, but whose realities it will be capable of recognizing. For minority communities, the future of AI extends far beyond questions of technological innovation. It concerns visibility, representation, and participation in increasingly automated societies. If historical inequalities continue shaping the datasets used to build AI, the technology risks reinforcing the very forms of exclusion it is often expected to eliminate. In an increasingly data-driven world, visibility is becoming a form of power. Ensuring minority communities are represented within the digital architecture of artificial intelligence may ultimately prove just as important as improving the algorithms themselves.