AI is built on systems that are often hidden from view, such as data drawn from communities, infrastructure that can burden local people, and labour that is frequently exploited or made invisible. At the same time, AI tools can support important work, including language revitalization, research, accessibility, education, service delivery, and community planning. This duality requires careful consideration of how AI is used, who benefits from it, who is burdened by it, and who has the authority to make those decisions. Communities need the knowledge, authority, and resources to decide how, when, and whether AI should be used, and to demand that any tools they choose are built, governed, and used in ways that serve their priorities rather than extract from them.
The Hidden Systems Behind AI
Data: The Fuel of AI
AI systems are built on data. Every AI tool learns from examples and records collected over many years. The quality, origin, and governance of this data shape how the system behaves. If the data is incomplete, biased, or taken without consent, the AI system will reflect those problems. The scale of data collection required to build modern AI raises serious questions that every community should ask.
Data ownership - Who owns it?
Who holds legal and cultural authority over information — including community, cultural, and personal records?
Consent - Who consented?
Was permission given for data to be collected and reused? Was it informed, voluntary, and culturally appropriate?
Governance - Who decides?
Who controls how data is stored, accessed, shared, and used over time?
Where Does AI Training Data Come From?
- Most data used to train AI is collected without direct consent from the people whose content, records, or cultural material is included. It can come from:
- Web scraping: automated collection of text, images, video, audio, and other media from across the internet
- Historical archives: government, academic, corporate, and cultural records.
- User-generated content: social media posts, forums, comments, uploaded media, and other online activity.
For Indigenous Nations, this creates the risk that cultural knowledge, language, imagery, records, and community information are absorbed into AI systems without permission, governance or benefit to the Nation they come from. This mirrors older patterns of extraction where resources were taken without consent and used to create value elsewhere. This is one way to understand data colonialism.
Infrastructure and Environmental Considerations
Every prompt entered into an AI tool depends on physical infrastructure: networks, servers, energy systems, water systems, and data centres filled with powerful computers. These facilities make AI possible, but they also carry real environmental, economic, and social costs.
Data centres can require significant amounts of electricity and water, especially to power servers and keep equipment cool. In some places, this can place pressure on local energy grids, water supplies, land use, and community infrastructure. These impacts are not the same everywhere, but they raise important questions about who benefits from AI infrastructure, who carries the costs, and whether local communities have a meaningful role in decisions that affect them.
For Indigenous Nations, these questions are especially important when infrastructure is proposed, built, or powered on lands and waters connected to Indigenous rights, title, jurisdiction, and stewardship responsibilities. Understanding AI means looking beyond the software itself and asking where its infrastructure is located, what resources it depends on, who governs it, and whether its benefits return to the communities affected by it.
The Labour Behind AI
AI can appear autonomous because the human work behind it is hidden from view. Behind every system are people who label data, moderate harmful content, test tools, review outputs, maintain infrastructure, and provide oversight. Much of this labour is poorly recognized, poorly compensated, or located far from where the profits are made. [1]
Data workers
People annotate, label, and moderate the data that trains AI models. This includes content moderation, which can include reviewing deeply disturbing material to train safety filters. Many major tech companies subcontract this work to poorly compensated workers in low-income countries. These "ghost workers" are essential but largely unrecognized. [2]
Mineral extraction
AI hardware depends on lithium, cobalt, and rare earth minerals mined in under-resourced regions. These operations carry heavy environmental costs and weak labour protections. The regions that bear the ecological and social burden rarely share in the profits. [3]
The Incentives Behind AI Development
AI is being shaped by powerful incentives: technology companies are racing to dominate markets, capture users’ attention, collect valuable data, and reduce labor costs through automation. These incentives influence not only which AI tools get built, but also how quickly they are released and whose interests they serve. [4]
Because advanced AI requires enormous computing power, large datasets, specialized talent, and expensive infrastructure, wealth and decision-making power can become concentrated in a small number of large corporations. As these companies gain more control over the tools, platforms, and systems that shape work and communication, workers, creators, educators, and communities may be left with less income, less bargaining power, and less say in how AI affects their lives. [5]
The Center for Humane Technology, along with a growing number of other voices, argues that a healthier AI future requires rules, business models, and design choices that share benefits more broadly, protect meaningful work, and keep human dignity at the center. This means asking not only what AI can do, but who benefits, who is harmed, and how society can shape AI in the public interest. [6]
References
[1] Mary L. Gray and Siddharth Suri, Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass (New York: Houghton Mifflin Harcourt, 2019).
[2] Gray and Suri, Ghost Work.
[3] Chinasa T. Okolo, "Global Majority Countries Must Embed Critical Minerals into AI Governance," Science 391, no. 6789 (March 5, 2026): eaef6678, .
[4] Center for Humane Technology. (2026). AI in society.
[5] Center for Humane Technology. (2026). Preserving what makes us deeply human in the age of AI.
[6] Center for Humane Technology, “Preserving What Makes Us Deeply Human.”