From Shelves to Patterns: What Dewey and Ranganathan Can Teach Us About Intelligence in the Age of AI

Long before artificial intelligence began organizing, retrieving, and generating knowledge, librarians confronted a remarkably similar problem: How do we impose useful structure on an expanding universe of knowledge? Melvil Dewey and S. R. Ranganathan offered two profoundly different answers. Seen through the lens of Pattern AI, their ideas reveal something surprisingly contemporary about classification, composition, and intelligence.

---

A Problem Much Older Than AI

We often describe information overload as a problem of the digital age.

It isn't.

Human beings have struggled for centuries with a fundamental problem:

As knowledge grows, how do we organize it so that people can find it, understand it, and use it?

Libraries confronted this problem long before search engines, databases, recommendation systems, large language models, or generative AI existed.

Two thinkers stand out in the history of library classification.

Melvil Dewey developed the Dewey Decimal Classification (DDC), first published in 1876.

More than half a century later, Indian mathematician and librarian S. R. Ranganathan developed Colon Classification, first published in 1933. (OCLC Library)

Both wanted to organize knowledge.

But the patterns underlying their approaches were strikingly different.

Dewey's system asks, in effect:

Where does this belong?

Ranganathan's system leads us toward another question:

What is this subject made of?

And Pattern AI adds a third:

What patterns are present here, and what can we accomplish by recognizing and recombining them?

The journey from Dewey to Ranganathan to AI can therefore be understood not simply as a history of classification.

It is a story about the changing architecture of knowledge.

---

1. Dewey's Great Pattern: Knowledge as Hierarchy

The power of the Dewey Decimal Classification comes from an elegant organizing idea:

knowledge can be arranged hierarchically.

DDC divides recorded knowledge into ten main classes. Each main class contains ten divisions, producing 100 divisions; these are further divided into 1,000 sections. OCLC, which maintains DDC today, explicitly describes this top-down organization. (OCLC Support)

Consider a simplified Dewey sequence:

600 — Technology
↓
630 — Agriculture and related technologies
↓
636 — Animal husbandry
↓
636.7 — Dogs

The structure moves from general to increasingly specific concepts.

Importantly, the notation itself normally carries information about hierarchy. OCLC explains that numbers with longer notation are generally subordinate to numbers with shorter notation, subject to some exceptions and special devices. (OCLC)

The underlying pattern is beautifully simple:

GENERAL → SPECIFIC → MORE SPECIFIC

Or, through the Pattern AI lens:

UNIVERSE → DOMAIN → CATEGORY → SUBCATEGORY → ENTITY

This is one of humanity's most enduring organizational patterns.

We see it everywhere.

A university might organize itself as:

University → Faculty → Department → Programme → Course

A company might use:

Company → Division → Department → Team → Role

A computer file system gives us:

Drive → Folder → Subfolder → File

A biological taxonomy uses:

Kingdom → Phylum → Class → Order → Family → Genus → Species

The specific classifications differ.

The underlying pattern does not.

The Dewey Pattern

We can therefore compress Dewey's logic into:

LOCATE → NARROW → CLASSIFY

That is our first Pattern AI insight.

Dewey was not merely assigning numbers to books. He was applying a reusable structural pattern to knowledge:

hierarchical decomposition.

---

2. Dewey and Pattern Recognition

Now consider what happens when a librarian receives a new book.

The classifier must determine what the work is about.

OCLC's official introduction to DDC states that classification requires determining the work's subject, disciplinary focus, and, where applicable, approach or form. It also warns that the title alone may not be enough to determine the actual subject. (OCLC)

Abstractly, the process looks something like this:

BOOK

↓

Identify subject

↓

Identify disciplinary context

↓

Find appropriate class

↓

Move toward appropriate specificity

↓

ASSIGN CLASSIFICATION

That begins to resemble a familiar machine-learning pattern:

INPUT

↓

FEATURES / EVIDENCE

↓

PATTERN RECOGNITION

↓

CLASS

Consider:

Email → characteristics → spam/not spam

Image → visual characteristics → cat/dog

Transaction → behavioral characteristics → fraudulent/legitimate

Document → subject characteristics → classification

This does not mean Dewey was doing machine learning.

He wasn't.

It means something more interesting:

Classification itself is a recurring pattern of intelligence.

The same abstract pattern can appear in librarianship, biology, medicine, computing and machine learning.

Pattern AI allows us to see the common structure underneath the different applications.

---

3. But Knowledge Does Not Always Behave Like a Tree

Hierarchies are extraordinarily useful.

But reality frequently refuses to stay neatly inside them.

Consider:

Artificial intelligence for early detection of crop disease using smartphone images in Karnataka.

Where does it belong?

Artificial intelligence?

Computer science?

Agriculture?

Plant pathology?

Image processing?

Mobile computing?

India?

The problem becomes increasingly difficult as knowledge becomes interdisciplinary.

This was one of the deeper problems that classification theorists had to confront.

A subject is not always simply a branch hanging from one larger branch.

Sometimes a subject is a combination of dimensions.

This is where Ranganathan becomes fascinating.

---

4. Ranganathan Changed the Question

S. R. Ranganathan's Colon Classification introduced an influential facet-analytical approach to knowledge organization.

The International Society for Knowledge Organization describes his facet analytical approach as perhaps his most important contribution to twentieth-century knowledge organization and notes its portability across technologies into the Internet era. (ISKO)

The intellectual shift is profound.

Instead of only asking:

Where should this subject be placed?

facet analysis asks:

What concepts constitute this subject?

That changes the operation from predominantly location toward analysis and synthesis.

The International Encyclopedia of Knowledge Organization describes the analytico-synthetic method in two operations:

Analysis: break the subject into its basic concepts.

Synthesis: combine the relevant conceptual units to represent the subject. (ISKO)

Through Pattern AI, we can express the process as:

SUBJECT → DECOMPOSE → IDENTIFY COMPONENTS → ORGANIZE → COMPOSE

This is a fundamentally different knowledge pattern.

---

5. PMEST: Looking for Recurring Dimensions

Ranganathan eventually generalized facets into five famous fundamental categories:

P — Personality
M — Matter
E — Energy
S — Space
T — Time

Together:

PMEST

The mature PMEST formulation emerged through the development of Colon Classification; ISKO dates Ranganathan's explicit theory of the five fundamental categories to 1952 rather than treating PMEST as a fully formed feature of the first 1933 edition. (ISKO)

Broadly interpreted, these categories ask questions such as:

Personality — What is the focal entity?

Matter — What material, property or substance is involved?

Energy — What process, operation or action occurs?

Space — Where?

Time — When?

This was an extraordinary conceptual move.

Because Ranganathan was searching not merely for subjects.

He was searching for recurring dimensions through which subjects could be constructed and understood.

And that is deeply compatible with Pattern Thinking.

---

6. Dewey Finds the Branch. Ranganathan Finds the Components.

Imagine knowledge as a tree.

Dewey's problem becomes:

Which branch should I follow?

The pattern is:

Knowledge
↓
Domain
↓
Discipline
↓
Subject
↓
Topic

Now imagine knowledge as something constructed from LEGO or Meccano pieces.

Ranganathan's question becomes:

Which components are present, and how should they be assembled?

The pattern changes:

Subject

↓

Analyze

↓

Identify facets

↓

Arrange relationships

↓

Synthesize

↓

Specific representation

The Meccano analogy is particularly appropriate historically. Scholarship on facet analysis reports that Ranganathan himself drew inspiration from seeing Meccano while studying in London. Facet analysis similarly allows concepts to be combined from reusable building blocks. (ISKO)

That gives us a memorable distinction:

Dewey sees the tree.

Ranganathan sees the building blocks.

---

7. The Supermarket and LEGO

Another analogy makes the difference easier to see.

Imagine entering a supermarket.

Products are organized:

Food → Beverages → Tea → Green Tea

You navigate the existing structure until you find what you want.

That is broadly Dewey-like.

The governing pattern is:

THING → CATEGORY → SUBCATEGORY → LOCATION

Now imagine opening a LEGO set.

There isn't necessarily one finished object waiting for you.

Instead, you have components.

You select pieces and combine them according to relationships and rules.

That is broadly Ranganathan-like.

The pattern becomes:

DECOMPOSE → SELECT → RELATE → COMPOSE

And suddenly we have moved from classification toward composition.

That distinction becomes very important when we arrive at generative AI.

---

8. A Necessary Correction: Dewey Is Not Merely a Set of Ready-Made Boxes

There is an important fact-checking qualification here.

It would be tempting to say:

"Dewey is completely enumerative while Ranganathan is completely synthetic."

That would oversimplify modern DDC.

DDC includes substantial mechanisms for number building.

OCLC defines number building as constructing a number by adding notation from tables or other parts of the schedules to a base number. It even describes this as notational synthesis. (OCLC Support)

WebDewey provides explicit workflows for incrementally building classification numbers when the required specific number is not directly present in the schedules. (OCLC Support)

So the comparison needs nuance.

The stronger distinction is:

DDC's dominant architecture is hierarchical, although it includes synthetic mechanisms.

while:

Colon Classification was deliberately developed around facet analysis and analytico-synthetic construction.

This distinction matters because Pattern AI should illuminate reality rather than force history into a convenient analogy.

---

9. Two Different Kinds of Pattern Thinking

We can now identify two important modes of Pattern Thinking.

Pattern Thinking Type 1: Find the pattern that contains something

This is predominantly Dewey-like reasoning.

The question is:

Where does X fit within the structure I already have?

Pattern:

INPUT → RECOGNIZE → MATCH → CLASSIFY

Examples:

Applicant → characteristics → job category

Disease symptoms → pattern → diagnostic category

Transaction → behavior → risk category

Document → content → subject category

This is the logic behind a great deal of classification AI.

---

Pattern Thinking Type 2: Find the patterns that constitute something

This is more Ranganathan-like.

The question changes:

What components make X what it is?

Pattern:

WHO/WHAT + PROPERTY + PROCESS + PLACE + TIME

or more generally:

WHO/WHAT → WHAT ABOUT IT → WHAT HAPPENS → WHERE → WHEN

Now we are no longer merely assigning something to a bucket.

We are exposing its internal structure.

This is compositional Pattern Thinking.

---

10. Why Ranganathan Feels Surprisingly Modern

Consider a modern e-commerce website.

Search for:

Laptop

You might then encounter filters for:

Brand

Processor

RAM

Storage

Screen size

Price

Rating

You are no longer navigating only:

Electronics → Computers → Laptops

You are exploring the object through multiple dimensions.

This is the broader logic of faceted navigation.

Ranganathan did not invent today's e-commerce filters, of course. Nor should we draw a direct technological lineage from Colon Classification to every modern filtering interface.

But the intellectual resemblance is powerful.

Facet analysis proved adaptable beyond its original library context. ISKO notes that Ranganathan's facet-analytical contribution has remained portable across changing technologies, while the broader facet-analysis literature explains how subjects can be constructed from combinations of conceptual building blocks. (ISKO)

The important pattern is:

Don't enumerate every possible combination in advance.

Instead:

Identify reusable dimensions and construct meaningful combinations when required.

That is a remarkably powerful design principle.

---

11. Now Enter Pattern AI

Pattern AI starts from a simple view of artificial intelligence:

AI derives patterns from data, recognizes patterns in new inputs, and generates new outputs by applying learned patterns.

That gives us three core capabilities:

1. DERIVE PATTERNS

Data → Relationships → Pattern

2. RECOGNIZE PATTERNS

New Input → Pattern Matching → Interpretation

3. GENERATE USING PATTERNS

Context + Learned Patterns → New Output

Now place Dewey and Ranganathan beside these capabilities.

Lens	Dewey	Ranganathan	Pattern AI
Primary organizing idea	Hierarchy	Facets	Patterns
Core question	Where does it belong?	What constitutes it?	What pattern is present/useful?
Operation	Locate/classify	Analyze/synthesize	Derive/recognize/generate
Representation	Hierarchical notation	Composed faceted notation	Learned/contextual representation
Strength	Position within structure	Composition from dimensions	Recognition, transfer and generation
Mental model	Tree	Building blocks	Dynamic pattern space

The table should not be interpreted as a technological evolutionary ladder.

Dewey and Ranganathan designed knowledge organization systems.

AI systems learn statistical representations from data.

These are fundamentally different technologies.

But at the level of patterns of thought, the comparison becomes highly illuminating.

---

12. From "Where?" to "What?" to "What Next?"

We can now compress almost 150 years of knowledge-organization thinking into three questions.

DEWEY

WHERE DOES IT BELONG?

Knowledge → Hierarchy → Location

---

RANGANATHAN

WHAT IS IT MADE OF?

Knowledge → Analysis → Facets → Synthesis

---

PATTERN AI

WHAT PATTERN IS HERE—AND WHAT CAN WE DO WITH IT?

Data/Knowledge → Pattern → Recognition → Transfer/Generation → Outcome

This final shift is crucial.

The first problem is organization.

The second is composition.

The third is application.

---

13. Classification Is Not the End of Intelligence

For centuries, organizing knowledge was itself a major intellectual challenge.

If a library contained a million books, finding the correct book was extraordinarily valuable.

Digital technology changed part of that equation.

Search engines made retrieval easier.

Databases made filtering easier.

Knowledge graphs made relationships increasingly explicit.

AI introduces another possibility.

Instead of merely asking:

Where is the knowledge?

we increasingly ask:

What can I accomplish with the knowledge?

That changes the role of knowledge infrastructure.

We move from:

STORE → ORGANIZE → RETRIEVE

toward:

UNDERSTAND → RECOGNIZE → COMBINE → GENERATE → APPLY

This is where Pattern AI becomes more than another classification method.

It becomes a way of working with knowledge.

---

14. A Pattern AI Example

Suppose an executive says:

"Customer complaints are increasing."

A purely categorical approach might classify the complaints:

Delivery

Quality

Billing

Service

Returns

Useful.

But Pattern Thinking asks another question:

What recurring structure is underneath these complaints?

Perhaps we discover:

Delay → Customer uncertainty → Repeated contact → Escalation

Now we have identified a pattern.

Call it:

THE ESCALATION LOOP

Once the pattern becomes visible, we can look for it elsewhere:

IT support

Hospital appointments

Banking queries

University admissions

Government services

The surface contexts differ.

The underlying pattern may recur.

This is the leap Pattern AI is interested in.

Knowledge becomes more powerful when a pattern discovered in one context can be recognized and transferred to another.

---

15. Ranganathan's Deeper Lesson for Generative AI

There is an especially important idea we can take from Ranganathan without claiming that he anticipated large language models.

His work demonstrates the intellectual power of compositionality.

Instead of attempting to list every possible complex subject beforehand, facet analysis provides components that can be combined as required. ISKO's explanation explicitly contrasts enumerated classes with faceted systems in which classes can be constructed from building blocks when needed. (ISKO)

That principle has much broader relevance:

Complexity can sometimes be managed not by storing every possible combination, but by discovering reusable components and rules of combination.

Consider language.

We don't memorize every sentence we will ever speak.

A finite vocabulary and grammatical structures enable enormous expressive variety.

Consider music.

A limited number of notes and structural relationships produce countless compositions.

Consider software.

Functions, APIs and modules can be recombined into applications.

Consider design.

Reusable components generate many interfaces.

The deeper pattern is:

PRIMITIVES + RELATIONSHIPS + COMPOSITION → VARIETY

That is one reason Ranganathan's thinking remains intellectually fascinating in the age of AI.

---

16. But Dewey Has Something AI Still Needs

It would be equally mistaken to conclude that hierarchy has become obsolete.

Quite the opposite.

AI systems operate in environments where users still need:

categories,

taxonomies,

ontologies,

metadata,

permissions,

business structures,

domain boundaries,

and organized context.

Dewey reminds us of something fundamental:

Structure reduces complexity.

A hierarchy answers:

What is broader?

What is narrower?

What belongs together?

Where should I begin looking?

Modern DDC itself continues to evolve. OCLC maintains WebDewey and updates the classification continuously; in 2026 it also released an updated print-on-demand edition. OCLC has additionally introduced Dewey linked-data identifiers intended to connect classification concepts with entities and knowledge-graph relationships. (OCLC)

So this isn't a story of:

old Dewey → better Ranganathan → superior AI.

It is something more useful:

Each exposes a different intelligence pattern.

---

17. Three Architectures of Knowledge

Seen through Pattern AI, we can identify three architectures.

Architecture 1 — Hierarchical Knowledge

Dewey

Knowledge
   ↓
Domain
   ↓
Category
   ↓
Subcategory
   ↓
Topic

Its power is navigation and position.

---

Architecture 2 — Faceted Knowledge

Ranganathan

Subject
   ↓
Analyze
   ↓
Facets
   ↓
Relationships
   ↓
Synthesize

Its power is decomposition and composition.

---

Architecture 3 — Pattern Knowledge

Pattern AI

Data + Context
      ↓
Patterns
      ↓
Recognize
      ↓
Relate
      ↓
Transfer / Generate
      ↓
Outcome

Its power is recognition, transfer and productive application.

And perhaps the most interesting future systems will combine all three.

---

18. From Knowledge Organization to Knowledge Intelligence

This suggests a broader historical arc.

LIBRARY AGE

Where is the knowledge?

The challenge:

ORGANIZE

---

DIGITAL AGE

How quickly can I find the knowledge?

The challenge:

SEARCH

---

AI AGE

What can I understand and accomplish using this knowledge?

The challenge:

INTELLIGENCE

This doesn't make libraries obsolete.

It reveals how fundamental their original problem was.

Before we could ask machines to reason over oceans of information, someone first had to confront the problem of giving knowledge structure.

Dewey and Ranganathan approached that problem from different directions.

Pattern AI allows us to see the intellectual continuity.

---

19. The Classification-to-Composition Principle

From this comparison we can derive what I would call:

The Classification-to-Composition Principle

Intelligence expands when we move from merely locating knowledge within existing structures to identifying reusable components and patterns that can be recombined for new contexts.

This does not eliminate classification.

It builds upon it.

The progression becomes:

CLASSIFY

Where does this belong?

↓

DECOMPOSE

What is this made of?

↓

RECOGNIZE

What recurring pattern is operating?

↓

TRANSFER

Where else does this pattern occur?

↓

COMPOSE

How can these patterns be recombined?

↓

GENERATE

What new output can we create?

↓

ACCOMPLISH

What outcome can we achieve?

That last step matters.

AI becomes valuable not when it merely generates something.

It becomes valuable when generated intelligence contributes to an outcome.

---

20. A New Way to Think About Knowledge

Dewey gave us an extraordinarily powerful idea:

Knowledge can be navigated through structure.

Ranganathan gave us another:

Complex subjects can be understood by analyzing and recombining dimensions.

Pattern AI adds:

Knowledge can also be approached through recurring, recognizable and reusable patterns.

Put together:

Hierarchy helps us locate knowledge.

Facets help us decompose and compose knowledge.

Patterns help us recognize, transfer and generate with knowledge.

That is not merely a comparison of two library classification systems.

It points toward a broader theory of how humans—and increasingly machines—deal with complexity.

---

21. The Library Was Already an Intelligence Laboratory

There is something wonderfully ironic here.

We tend to imagine the AI revolution as beginning inside computer laboratories.

But one of its deepest intellectual problems has been sitting inside libraries for centuries:

How do we represent an ever-growing universe of knowledge without being overwhelmed by it?

Dewey's answer was, broadly:

Build a structure.

Ranganathan's answer added:

Discover the facets and compose the subject.

Pattern AI suggests another step:

Discover the recurring pattern, recognize it across contexts, and use it to produce an outcome.

So perhaps the history of classification contains a lesson for the future of artificial intelligence.

The real breakthrough is not simply having more information.

Nor is it merely retrieving information faster.

It is learning to see the structure beneath information.

Because once we see the structure, something changes.

A million individual cases can become a handful of reusable patterns.

A complicated subject can become a composition of intelligible components.

And knowledge that once had to be searched for can increasingly become knowledge that can be recognized, recombined, transferred and applied.

That may be one of the defining shifts of the AI age.

---

One Idea to Take With You

**Dewey taught us to find the right shelf.

Ranganathan taught us to find the building blocks.
Pattern AI asks us to find the reusable intelligence hidden underneath both.**

Or, in three questions:

Dewey:
Where does it belong?

Ranganathan:
What is it made of?

Pattern AI:
What pattern is operating—and what can we accomplish with it?

Perhaps that is the journey from organizing knowledge to working intelligently with knowledge.

---

Fact-Check & Source Note

The historical and classification-system claims in this article were checked against authoritative or scholarly sources. OCLC's official DDC documentation confirms Dewey's hierarchical structure, ten main classes, notational hierarchy, subject-analysis procedure, tables, and number-building/notational-synthesis mechanisms. (OCLC)

The discussion of Ranganathan, Colon Classification, analytico-synthetic classification, PMEST and facet analysis is grounded primarily in the International Society for Knowledge Organization's Encyclopedia of Knowledge Organization. Its Colon Classification entry documents the development of CC, PMEST and Ranganathan's classification theory, while its facet-analysis entry defines analysis and synthesis and documents the Meccano connection. (ISKO)

The article deliberately treats the connection to Pattern AI as an interpretive framework, not a historical claim that Dewey or Ranganathan anticipated machine learning or generative AI. Likewise, Ranganathan's influence on facet analysis should not be interpreted to mean that every contemporary faceted-search interface descends directly from Colon Classification.

Verified reference gateways: OCLC — Dewey Decimal Classification · ISKO — Colon Classification · ISKO — Facet Analysis