Parallels Between CAD/CAM and BIM with XBRL

I have an observation: People are overhyping the role of AI in humanity's future. The way that I see it is that what is actually happening is a revolution in information and AI is one component of that revolution; not the revolution itself.

AI operating on raw text is impressive.

AI operating on a model of the domain is transformative.

The deepest revolution is not intelligence. It is the conversion of human knowledge into stable computable structures. (Knowledge | Stable | Computable | Structures)

As I pointed out in another blog post; there are parallels between CAD/CAM and BIM which related to artifacts related to design, engineering, architecture, and the construction of physical things and XBRL which relates to artifacts related to accounting, reporting, auditing, analysis of financial information.

I think the parallel is deeper than "both are machine interpretable structured information formats."  What I see is that they are both instances of the same fundamental shift: replacing a human-interpretable document artifact with a machine-interpretable logical model artifact and a projection that humans can still interpret.

Here's how I see it:

CAD/CAM, BIM, and XBRL are three instances of the same change: each converts a body of human knowledge; physical design, construction, financial reporting; from documents that humans interpret into computable structures that machines can reason over but humans can still interpret. They are the revolution; AI is what arrives afterward to exploit them.

The shared pattern. In each domain, the artifact of record was once a drawing or a document: blueprints, sheets, financial statements. In each, it was demoted to a view of a structured model; BIM-generated drawing sheets, iXBRL renderings of tagged facts. In each, objects carry semantics, not just appearance: a beam carries load, material, and cost; an XBRL fact carries concept, period, unit, and dimensional axes. In each, standardization made the structures interoperable (IFC for buildings, US GAAP/IFRS taxonomies for reporting), machine validation caught errors before they became expensive (clash detection; consistency checks), and adoption stalled until mandates forced the coordination; because the party paying to model rarely was the party reaping the value from having the model.

The divergences matter too. BIM models are constrained by physics and close the loop to fabrication; the model becomes toolpaths and steel. XBRL is constrained by interpretive rules and informs capital allocation rather than manufacturing it. A building that violates gravity fails visibly; a financial filing can be internally consistent yet economically misleading. And XBRL's ambition is market-wide comparability across every reporting entity; a significantly harder coordination problem than any single project.

Now apply the framing. AI operating on raw text is impressive; an LLM reading 10-Ks and blueprints PDFs, summarizing, guessing structure after the fact. AI operating on a model of the domain is transformative; the same intelligence querying tagged facts that tie to a taxonomy, or querying a BIM model with actual materials and costs, where every answer is traceable to a defined concept rather than inferred from prose. The intelligence is the same; the substrate is not. Raw text forces the machine to reconstruct the model on every pass. A computable structure makes the model a permanent, shared asset that intelligence compounds against.

The deepest revolution, then, is not that machines became intelligent. It is that human knowledge; how we build, what we own, what we owe; became structured, versioned, and queryable. CAD/CAM and BIM did this for physical objects. XBRL is doing it for financial facts. AI is the beneficiary of that conversion, not its cause; and its transformative power will be proportional to how much of our knowledge has actually been made computable, not to how clever the models become.

Engineering, Design, Architecture, Construction

Consider the history of engineering. The revolutionary step was not that computers became smart. The revolutionary step was that engineers stopped expressing knowledge primarily as:

  • drawings
  • documents
  • specifications

and started expressing knowledge as:

  • objects
  • relationships
  • constraints
  • rules
  • models

BIM became a machine-readable representation of reality. Once that happened, many technologies suddenly became possible:

  • automated validation (clash detection between objects)
  • simulation
  • optimization
  • generative design
  • digital twins
  • AI assistants

AI came later and benefited from the information model.

Accounting, Reporting, Auditing, Analysis

The same thing is happening in accounting. Historically, accounting knowledge lived in:

  • reports
  • spreadsheets
  • audit workpapers
  • accounting manuals
  • human expertise

AI can read those artifacts. But AI reading documents is still somewhat analogous to an engineer trying to understand a building by reading blueprints. The bigger transition is moving accounting knowledge into:

  • XBRL taxonomies
  • reporting frameworks
  • semantic models
  • knowledge graphs
  • computable rules
  • interoperable ontologies

Once that occurs, AI has something much richer to work with.

An analogy

Consider the following analogy. Imagine we have two worlds.

  • World A: The AI has access to every engineering textbook ever written and/or every accounting textbook ever written.
  • World B: The AI has access to: a BIM model, structural rules, material properties, building relationships, occupancy information and/or the US GAAP reporting framework model, the IFRS reporting framework model, structural rules of disclosures, business event properties, audit evidence graphs and records of auditor judgment.

Which system is likely to produce better answers?

Obviously World B. Not because the AI is smarter. Because the information is better organized.

Information Models Amplify Intelligence

One way to think about it is: Intelligence × Information Quality = Capability

If information quality is low: 100 × 1 = 100

If information quality is high: 100 × 100 = 10,000

The biggest gain often comes not from increasing intelligence but from increasing the quality and structure of information.

Why AI Appears Revolutionary

AI is highly visible because it "talks". People interact directly with ChatGPT, Copilot, Claude, Gemini, and others.

Information models are largely invisible. Nobody gets excited about:

  • schemas
  • ontologies
  • taxonomies
  • semantic relationships
  • metadata

Yet these often create the conditions that make AI useful. This is the stuff that supercharges artificial intelligence.

It is similar to electricity. People notice the light bulb. They don't notice the power grid.

BIM and XBRL Suggest the Same Future

BIM taught us: The model becomes more important than the drawing.

XBRL is teaching us: The model becomes more important than the report.

The next step may be: The model becomes more important than the AI. Why? Because multiple AIs can operate on the same model.

The enduring asset is not the AI. The enduring asset is the structured representation of knowledge.

The Pacioli Parallel

This connects directly to something that seems to run through the Seattle Method, REA, Data Centric Accounting, ISO/IEC Accounting and Economic Ontology, SBRM:

The truly scarce resource is not intelligence. Human intelligence has always existed. The scarce resource is a shared, explicit, computable representation of domain knowledge. Engineering created CAD/CAM and then BIM. Medicine is creating clinical ontologies. Law is beginning to create computable legal rules. Accounting created bookkeeping and the notion of the ledger; then double-entry bookkeeping to manage accountability and the notion of the manifold. Accounting is moving toward computable reporting frameworks. Computable business events. Computable governance.  Computable accountability.  Computable audit evidence. Computable records of auditor judgement. Computable financial statements. Computable accounting working papers.

AI then becomes the reasoning layer that operates on those representations. The key is the information pattern.  Also, this is not really about XBRL as an information organization approach; PROLOG can work, the Semantic Web Stack can work,  Labeled Property Graphs can work. Even relational databases can be made to work. There are many different representation or implementation approaches.  Each as pros and cons which should be well understood.

Digital Information Organisms

I refer to these stable computable structures as digital information organisms. They are engineered to be easy to use by creating them at the appropriate level for business professionals to use by leveraging the Atomic Design Methodology. All artifacts business professionals work with are at the level of the logic the business professional works with, not the level of the physical technical format which is (a) ugly, (b) difficult to understand, (c) unnecessarily challenging to actually work with.

Think "Lego" brinks of information.  Complexity is beaten down; accidental complexity eliminated. Elegant software is the result, not like the current kludge.


Additional Information:

Comments

Popular posts from this blog

Internationalized Resource Identifier (IRI)

Example Financial Statement Holon

Inhabiting Babel, A Manifesto for Responsible Meaning Engineering