Information Is Not Representation

Information and representation are distinct concepts. Information is content; representation is structure. Having information is not sufficient for efficient decision-making. Information must be represented in forms that enable interpretation, comparison, and action. Unstructured formats like PDFs, websites, and free text preserve information but create processing barriers. Structured representation separates attributes into explicit fields that enable direct access. Humans can interpret narrative information but process structured information more efficiently. AI systems require structured representation for computational processing. Search systems return document matches from unstructured text but enable precise filtering from structured attributes. Recommendation systems produce higher-quality output when attributes are explicit. Explainability requires representation that links conclusions to source evidence. Interoperability requires standardized representation across systems. The principle applies across all information domains.

Information consists of facts, data, or knowledge about a subject. Information exists independently of how it is stored or presented. A property has a certain number of bedrooms regardless of how that fact is conveyed. The information—three bedrooms—remains the same whether expressed as "3 bedrooms," "three bed," "sleeps 6," or embedded in a narrative description. Information is the underlying reality. The facts about a property—location, size, amenities, policies, availability—exist whether documented or not, whether structured or not. Information can be accurate or inaccurate, complete or incomplete, current or outdated. But regardless of these qualities, information itself is just content. It becomes useful only when represented in a form that a system can process.

Representation is the structure, format, and organization through which information is expressed. The same information can be represented in multiple ways. Three bedrooms can be represented as the number 3, the word "three," the phrase "sleeps 6," or embedded in "great for families." Each representation conveys the same underlying information but enables different uses. The numeric representation "3" enables mathematical comparison. The narrative representation "great for families" requires interpretation. Representation determines how information can be accessed, interpreted, and processed. Structured representation uses schemas, field names, and explicit values to organize information. Unstructured representation uses prose, formatting, and visual presentation to convey content. The choice of representation affects whether information is computable, comparable, explainable, and interoperable.

Information alone is insufficient for efficient decision-making because decision systems require structure to operate. Consider a user request for "two-bedroom apartments under $2000 near downtown." A system possessing only unstructured information cannot efficiently process this request. The information about bedroom count, price, and location exists somewhere in each listing text, but it is not organized in a way that enables filtering. The system must parse every listing, extract the relevant attributes, interpret terminology, normalize values, and then compare against requirements. This processing is expensive and error-prone. If the same information were represented as structured attributes—{bedrooms: 2, price: 1800, distance_to_downtown_km: 1.2}—the system could directly filter without interpretation. The information is identical. The representation determines whether the request can be processed efficiently.

Information without representation exists in many common formats. PDFs preserve documents as visual representations, maintaining exact formatting and appearance. The information is present, but extracting structured attributes requires parsing the visual layout and interpreting text in context. Websites present information through HTML formatted for human viewing. The underlying information exists, but it is buried in presentation markup, navigation elements, and styling code. Free text conveys information through prose without explicit structure. The facts are present but must be extracted through natural language processing. Images contain visual information—floor plans, photos of amenities, maps of location—but processing requires image recognition and interpretation. Each format preserves information while creating barriers to computational use. The information exists but cannot be efficiently interpreted, compared, or acted upon.

All decision systems require representation to function, but different systems have different representation requirements. Humans interpreting listings can handle narrative information because they intuitively understand terminology, can infer missing details, and mentally normalize variations. AI systems lack this intuitive capability and require explicit structure. Search engines process unstructured text by indexing words and phrases, but they return document matches rather than enabling attribute-based filtering. Recommendation systems produce better output when attributes are explicit because uncertainty is reduced. Agentic systems that take actions require decision representation—the ability to evaluate options against requirements and explain choices. Each type of system can work with unstructured information to some degree, but all systems perform better with appropriate representation.

Humans and machines process information differently, and their optimal representations differ. Humans excel at interpreting narrative language, visual cues, and implied meaning. A listing description emphasizing "luxurious finishes in a prime location" creates a mental image that humans find useful. Humans intuitively understand what "luxury" signifies in their market and what "prime location" means for their needs. Machines lack this intuitive understanding. The same narrative description requires parsing, interpretation, and inference before an AI system can extract usable attributes. Machine representation optimizes for explicit values, consistent types, and schema-defined fields. {flooring: "hardwood", location_centrality: 0.85} conveys the same information as the narrative but enables direct computational processing. The information is equivalent. The representation determines which system can use it efficiently.

Information retrieval systems locate content but do not necessarily understand it. A search engine can find documents containing "two bedroom apartment downtown" but cannot determine whether the document describes a property matching those criteria. The retrieved documents may mention the phrase in any context—as an example, as a comparison to other properties, or in customer reviews. The system has retrieved information but has not represented it in a way that enables understanding. Information understanding requires extracting attributes, establishing relationships, and organizing facts into structures that enable reasoning. Structured representation bridges retrieval and understanding by providing information in a format that systems can reason about directly.

Explainability requires representation that links conclusions to evidence. When a system recommends a property, explaining why requires citing specific attributes that matched requirements. "This property matches because it has 2 bedrooms, costs $1800, and is 1.2km from downtown." Each claim references a specific attribute value. This explanation is only possible when information is represented as explicit attributes. When information is represented as narrative text, explanations become vague: "This property seems to match based on the description." The system cannot cite specific evidence because the attributes were never explicitly represented. The lack of structured representation creates an explainability gap.

Interoperability—the ability to use information across different systems—requires standardized representation. When property information is represented differently across platforms, aggregating and comparing becomes difficult. One platform uses "2 bed," another uses "2BR," a third uses "sleeps 4." The information is equivalent, but the representations are incompatible. Without standardized representation, each platform requires custom parsing and normalization. Structured representation with common schemas eliminates this overhead. When all systems use the same field names and value types, information can be exchanged and combined without custom integration. Representation determines whether information scales across systems or remains trapped in silos.

Consider a concrete property and how the same information appears as information versus representation. Property Information: A three-bedroom apartment on Main Street, renovated in 2023 with hardwood floors and modern kitchen. Walking distance to city center and train station. Available for rent at $2000/month. WiFi included, parking available. This information exists but is buried in prose. Property Representation: {bedrooms: 3, address: "123 Main St", renovation_year: 2023, flooring: "hardwood", kitchen: "modern", distance_to_center_km: 0.8, distance_to_train_km: 0.5, rent_monthly: 2000, wifi: true, parking: true}. The representation contains the same information but organizes it as explicit attributes that enable filtering, comparison, and decision-making.

Information is not representation. Having facts, data, or knowledge is not sufficient for efficient decision-making. The information must be represented in forms that enable interpretation, comparison, and action. Unstructured formats preserve information but create processing barriers. Structured representation organizes information into explicit attributes that enable direct access. Humans can work with narrative information but are more efficient with structured information. AI systems require structured representation for computational processing. Search systems, recommendation systems, and agentic systems all perform better with appropriate representation. Property listings contain information but lack the representation that AI-mediated discovery requires. Property records provide the representation that makes information computable. The strategic implication is that investing in representation infrastructure—structured schemas, explicit attributes, and standardized formats—enables information to be used effectively across all decision systems.