 In 1869, when the first transcontinental railroad was completed in the United States, newspapers declared it a triumph of progress — a steel ribbon that would bind a vast nation together. Yet within a generation, it became clear that the same tracks that united some also isolated others. Towns bypassed by the main line faded into obscurity; Indigenous communities were displaced from ancestral lands; and wealth pooled along the routes determined by a handful of engineers and financiers. The railroad, celebrated as a symbol of connection, had quietly encoded inequality into geography itself.
A century later, when the Internet was first imagined, its architects believed they had learned from such lessons of centralization. This time, the network would have no single track, no privileged junctions, no gatekeepers. It would be the most democratic technology in human history — a decentralized web of networks without a center, without a master, without prejudice. Its architecture itself was supposed to guarantee fairness. No single entity could control it, no government could dictate its operation, no corporation could own its core. Anyone, anywhere, with a connection and a modest machine could join and be part of a global community of equals. That was the idea. Decentralization was the design’s moral claim. Paul Baran’s 1964 vision of “distributed communications networks” was born from the Cold War fear of nuclear annihilation — a system resilient because it had no single point of failure. When Vint Cerf and Bob Kahn later translated that idea into TCP/IP, the guiding principle was indifference: routers didn’t care who you were or where you came from, only that you spoke the common language of packets. This technical neutrality was meant to be a structural defense against bias. If no one was in charge, then no one could discriminate. Yet as a Cloudflare research makes clear, the neutrality of design did not guarantee the equality of experience. Bias crept in not through censorship or control, but through the quiet accumulation of technical and economic asymmetries. The Internet’s architecture — designed for openness — was gradually shaped by the unequal distribution of resources, timing, and attention. The story begins, ironically, at the very level of the Internet’s atomic structure: the IP address. In the 1980s and early 1990s, before the Internet was a household word, address space was freely given to early adopters — universities, corporations, and government agencies, overwhelmingly in the United States and Western Europe. These allocations were generous, sometimes recklessly so: Stanford University once held more IPv4 addresses than the entire continent of Africa. The assumption was that there would always be more to go around, and that those connecting early were the natural stewards of the network’s growth. It was a small group of people making decisions for a world they could barely imagine. By the time the Internet reached the Global South in full force, the pool of IPv4 addresses had nearly run dry. The early open-handedness had become a zero-sum game. To keep new users online, Internet service providers in Africa, Asia, and Latin America were forced to rely on Carrier-Grade Network Address Translation (CGNAT), a workaround that allows thousands of users to share a single public IP address. Cloudflare’s study shows the consequences of that workaround: requests from CGNAT-shared addresses — common in Nigeria, India, or Indonesia — are three times more likely to be rate-limited or blocked by online services than those from the unique addresses typical of Europe or North America. On paper, the Internet still functions as a universal network. In practice, it has become a layered one. People in the Global South often connect through labyrinths of shared addresses and intermediate proxies, while those in the Global North usually enjoy direct, unique routes. The result is not just slower service, but structural misrecognition. When multiple users share an IP, the misbehavior of one can lead to collective punishment: an entire neighborhood or region suddenly locked out of websites or platforms. The Internet was built to treat all packets equally, but the packets coming from some parts of the world are more likely to be mistrusted. This bias is not ideological but infrastructural. It arises from an asymmetry embedded at the birth of the network. The Internet’s early designers believed that decentralization itself would prevent inequality — that if no one controlled the system, it would naturally remain fair. But history offers many examples of systems built on similar optimism that later hardened into inequality. In the early 20th century, electricity was hailed as the “great democratizer” — a force that would light every home and level every economy. But grids were laid first in industrial and urban centers, leaving vast rural regions literally in the dark for decades. Infrastructure designed to connect the many instead deepened the divide between those who could afford early access and those who could not. The Internet’s topology, both logical and physical, has produced a digital parallel: the grids of privilege were laid early, and they are hard to reroute. As the network commercialized in the 1990s, another layer of bias took hold — economic rather than technical. The prevailing wisdom was that the Internet’s evolution should be left to the market. The U.S. government’s decision to privatize core Internet functions in the mid-1990s reflected a deep faith in what economist Milton Friedman called “the invisible hand of innovation.” Infrastructure was treated as a business opportunity, not a public good. Under this model, questions of fairness — who got addresses, who built cables, who could afford backbone connections — became questions of profit. Decentralization, which had been a safeguard against control, became a justification for disengagement. No one was responsible for ensuring global equity because, supposedly, no one was in charge. The architecture that once guaranteed openness became a substrate for market concentration: today, a handful of companies control the majority of the world’s cloud and content delivery infrastructure. The Internet remains decentralized in form, but deeply centralized in function. The Global South feels this contradiction most acutely. The digital divide has long been framed in terms of access and skills, as if the solution were simply to connect more people and train them to use technology. But the divide now runs deeper — into the operational logic of the network itself. A user in Nairobi or Dhaka may have the same smartphone and broadband plan as someone in Berlin, yet their connection traverses more intermediaries, carries more latency, and is more likely to be misclassified or penalized. This is not a failure of connectivity, but of architecture — a kind of invisible infrastructure of inequality. The rise of artificial intelligence threatens to magnify that gap. AI depends on abundant, stable, and uniquely identifiable data streams. But in CGNAT environments, where thousands of devices share a single address, the “uniqueness” of users becomes mathematically blurred. AI systems trained on data from the Global North may interpret this shared connectivity as noise or fraud, misreading the realities of the Global South as anomalies. In this way, architectural bias feeds algorithmic bias, which then reinforces architectural disadvantage — a feedback loop of inequity. These are not merely engineering problems; they are political ones. Many governments in the Global South have turned to the United Nations and other multilateral fora for redress, recognizing that no single nation can correct an imbalance rooted in global infrastructure. Processes such as the World Summit on the Information Society (WSIS+20) review, the implementation of the Global Digital Compact, and debates around Digital Public Infrastructure (DPI) reflect a growing insistence that the Internet’s evolution must be guided by public interest, not market expedience. The challenge is not only to connect more people, but to rebuild the network so that its architecture does not systematically disadvantage them. Here, the framework of human rights may offer a moral compass. The UN Human Rights Council has affirmed that the same rights people enjoy offline must be protected online, recognizing Internet access as a facilitator of fundamental freedoms. But access, in the human-rights sense, cannot mean mere connectivity. It must mean meaningful connectivity — a quality of access that allows individuals to participate fully and equally. If the Internet’s architecture prevents that, then the inequality is not just technical but moral. Human-rights law offers an important distinction between intentional and structural discrimination. The latter occurs when neutral rules produce unequal outcomes — when systems designed without malice nonetheless disadvantage some groups. The Internet’s architecture fits this definition precisely. The early allocation of IPv4 addresses, the market-driven approach to infrastructure investment, and the indifference to the lived realities of late-connecting regions have created a pattern of structural discrimination, even if no one meant for it to happen. Recognizing that is the first step toward redress. Correcting architectural inequality will require more than policy declarations. It will mean accelerating IPv6 adoption in underserved regions, rethinking how global Internet governance allocates resources, and ensuring that emerging technologies — from AI to edge computing — are built with diverse network conditions in mind. It also means rejecting the idea that the Internet’s evolution should be guided solely by what is financially viable. The notion that the market will automatically optimize for human well-being has been tested, and it has failed. Some decisions about the Internet — how it connects people, who it empowers, and what it values — must be made because they are good for humans, even if they are not profitable. To ask whether the Internet is biased, then, is to confront a deeper irony. The Internet’s design — decentralized, open, resilient — was meant to immunize it against precisely the kinds of bias we now see. And in theory, it still could. Decentralization is not the problem; the problem lies in the early choices that structured who got to participate in that decentralization and on what terms. Like a democratic constitution written by a narrow elite, the Internet’s founding architecture enshrined equality in principle but distributed power unevenly in practice. To ask Is the Internet biased? is to confront the legacy of choices made when the network was young — choices that privileged business over humanity, speed over solidarity. The question, then, is not merely diagnostic but prophetic: will the next evolution of the Internet — through AI, digital public infrastructure, and global governance — repeat those choices, or finally redeem the promise of a truly universal network? Because if the Internet is biased, it is only because we made it so. And that means we can unmake it, too. Comments are closed.
|