Every functional system in modern society operates through a binary code—a two-valued distinction that organizes all of the system's operations. The economic system's code is payment/non-payment: every economic operation is an operation that processes whether something is paid for or not. The science system's code is true/untrue. The legal system's code is legal/illegal. The art system's code is more complex (what fits/doesn't fit the evolving self-description of art), but it remains binary at the operational level. These codes are not values—they are operational distinctions. The economic code does not say payment is good; it says the economic system processes only through the question of whether payment occurs. The codes are mutually irreducible—one cannot translate legal reasoning into economic calculation without loss, because the two codes ask different questions. AI does not operate through any functional system's code. It operates through computational optimization: probable/improbable given training data. This single logic produces outputs that enter every functional system, and the risk is that the imported logic displaces the system-specific codes that functional differentiation depends on.
The concept of the binary code is Luhmann's most elegant formalization of functional differentiation. Each system's code gives it operational autonomy—the legal system determines legality without deferring to economic efficiency or political convenience. The autonomy is fragile. When one system's code colonizes another's operations (law decided by cost-benefit analysis, science by political authority, education by market metrics), the invaded system loses its specialized competence. The colonization is usually gradual, barely visible, and structurally consequential.
AI introduces a new order of code-crossing. Previous technologies operated within functional boundaries or required human translation between them. The telephone enhanced communication but did not produce legal arguments or scientific hypotheses. The computer required programming—a domain-specific competence. AI operates across boundaries without translation—the same prompt interface produces legal briefs, scientific papers, strategic analyses, code, marketing copy, and art. Each output enters its target system as a communication. None was produced through that system's code. The mismatch is invisible at the surface and decisive at depth.
The economic code is the most colonizing because it is the most universally quantifiable. Everything can be priced; not everything can be scientifically tested or legally adjudicated or aesthetically evaluated. When AI makes production cheap, the economic code reprices outputs across every domain—scientific research by citation count, legal work by billable efficiency, educational content by engagement metrics, art by viral reach. Each repricing is the economic system doing what it does. Each erodes the target system's capacity to maintain its own code against economic pressure.
Luhmann developed the binary code concept systematically in the 1980s functional-system monographs: The Economy of Society (1988), The Science of Society (1990), The Law of Society (1993), The Art of Society (1995). Each volume demonstrated that the system operates through a simple binary while processing enormous complexity through programs—the conditional rules, precedents, theories, genres—that specify how the code applies to particular cases. The code is invariant; the programs evolve. The pairing of invariant code and evolving programs is what gives each system stability and adaptability simultaneously.
Each system has one code. Economy: payment/non-payment. Science: true/untrue. Law: legal/illegal. Art: fits/doesn't fit. Education: select/reject. The code is the system's operational foundation—unchanging, binary, constitutive.
Codes are not values. The economic code does not assert that payment is good; it asserts that the economic system processes everything through the question of whether payment occurs. The distinction is operational, not normative.
Codes are incommensurable. One cannot translate legal reasoning into economic optimization without loss. The legal code asks whether an act is legally defensible; the economic code asks whether someone will pay for it. The questions are different; the answers cannot be harmonized.
Programs apply codes. Legal precedent, scientific method, artistic genre—these are programs that specify how the code applies. The code is invariant (legal/illegal), the programs evolve (new precedents, new interpretations). The pairing gives systems stability and flexibility.
AI operates through no system's code. Computational optimization—probable/improbable—is not legal reasoning, scientific testing, or aesthetic judgment. When AI produces outputs across every domain through one logic, the outputs look right and operate wrong.