For nearly three decades, the internet has been built around a simple assumption: Humans are the primary consumers of information. Articles were written for us to read, not machines. Every major information system in modern society was built around human cognition and human communication.
The Economist is now preparing for what it calls a “two-track internet,” which may prove to be one of the most important publishing stories of the AI era. The publication is exploring ways to create content that serves both human readers and AI agents, recognizing that increasingly, the first consumer of information may not be a person at all. Instead, it may be an artificial intelligence system that reads, interprets, summarizes and repackages information before a human ever encounters it.
For centuries, every major communication system was designed around the limitations and strengths of the human mind. Artificial intelligence does not think that way.
Large language models and other AI systems perform best when information is highly structured, explicitly defined and organized into relationships that machines can process efficiently. While humans thrive on ambiguity, context, metaphor and inference, machines generally perform better when information is categorized, tagged, labeled and linked in ways that reduce uncertainty. What appears elegant to a human reader often appears inefficient to an AI system.
As a result, organizations throughout society are beginning to ask an extraordinary question. If AI systems are becoming major consumers of information, should information itself be redesigned for AI?
The publishing industry may simply be the first sector confronting this reality publicly.
The deeper issue is not whether publishers will create separate websites for humans and machines. The deeper issue is whether every form of human knowledge will eventually require translation into a second language: the language of artificial intelligence.
For decades, a clinical note was intended to communicate observations, diagnostic reasoning, treatment decisions, uncertainty and professional judgment. Increasingly, however, the medical record serves multiple audiences simultaneously. AI systems review documentation for coding, quality metrics, risk adjustment, utilization management, clinical decision support, predictive analytics and revenue cycle management. The physician may believe that he or she is documenting a patient encounter, but the note is increasingly being written for both a human reader and a machine reader.
The implications are significant. If documentation is optimized for machine interpretation, physicians may gradually alter how they communicate. Clinical narratives may become more structured. Diagnoses may become more standardized. Observations may be translated into categories and data fields that are easier for AI systems to process.
Imagine a future judicial opinion that exists in two forms. The traditional opinion would remain available for judges, attorneys, scholars and law students. Alongside it would exist a machine-readable version in which every fact, precedent, legal principle, jurisdictional limitation and judicial holding is explicitly tagged and categorized. An AI system could instantly determine what portion of an opinion constitutes binding precedent, what portion represents dicta and how the ruling connects to thousands of other decisions.
Such a system could dramatically improve legal research and reduce errors. At the same time, it raises important questions about authority and interpretation. If humans read one version and machines consume another, which version ultimately becomes the source of truth?
Business schools have relied on case studies that teach students through narrative. Readers learn not simply from facts but from context, uncertainty, leadership dilemmas and competing priorities. Yet AI systems often struggle to identify the subtle causal relationships that human readers extract naturally. In a two-track world, every case study could have a companion version specifically designed for machine consumption. Stakeholders, decisions, outcomes, timelines, risks and causal relationships could be encoded in structured formats that AI systems understand immediately.
This raises a question that sounds philosophical but is becoming increasingly practical. Are we creating a second language for knowledge itself?
Throughout history, human civilization has repeatedly developed new systems for organizing information. We invented writing to preserve knowledge beyond memory. We developed printing to distribute knowledge at scale. We created libraries, databases, search engines and digital networks to manage ever-growing volumes of information.
Artificial intelligence may represent the next stage in that evolution. However, unlike previous information technologies, AI is not merely storing information or transmitting it. AI is actively consuming information as an independent participant within the information ecosystem. Consequently, information must increasingly be structured in ways that support machine understanding.
The result is the emergence of what could be described as a parallel informational universe. The more capable AI becomes, the more important this invisible layer becomes.
Ironically, one of the goals of this transformation is greater accuracy. AI hallucinations often occur because machines must infer relationships from unstructured information. Publishers, researchers and institutions are increasingly discovering that when information is presented in structured formats, AI systems produce more reliable outputs. The creation of AI-native content is therefore not simply about efficiency. It is also about reducing ambiguity and improving trustworthiness.
Yet there is a potential paradox embedded within this strategy.
The more we optimize information for machines, the more we may alter the way humans communicate. For the first time in history, humanity is not merely teaching machines our language. We are beginning to learn theirs.
The Economist’s experiment, therefore, represents something far larger than a publishing initiative. It represents a glimpse into a future in which every organization must decide whether its information exists solely for humans or whether it must also exist for machines. It suggests a world in which every important document may have two audiences, two formats and perhaps eventually two realities.
The critical question is whether these realities remain connected.
If they do, the result could be extraordinary. Medical knowledge could become more accessible. Legal research could become more precise. Scientific discovery could accelerate. Educational content could become more personalized and effective. Artificial intelligence could function as a trusted intermediary between people and information.
If that occurs, The Economist’s two-track internet may be remembered as one of the earliest signs that humanity had begun constructing a parallel universe of knowledge for a new form of intelligence.
Sreedhar Potarazu is an ophthalmologist and former health care executive.
