In a system I worked on, "partner" originally meant an external organization with which we had a business relationship. When the system had to support brands we ran ourselves, the term was carried forward to them too, because the infrastructure was already built around it. The consequences were minor; however, separating partners from non-partners was not always straightforward, and the term sometimes led to confusing discussions. This drift was overlooked because nobody was testing whether "partner" still meant what it was defined to mean.

This overview describes how I apply five sources to problems like that one: Stamper's semiotic ladder, Peirce, Wittgenstein, Date and Darwen, and Dietz and Mulder. It states what each says and how it applies.

The Semiotic Ladder

The Semiotic Ladder was developed by computer scientist Ronald Stamper based on earlier work by the philosopher and semiotician Charles Morris. It provides a framework to break down signs and communication into distinct interconnected levels. The result is a way to say which level of meaning broke down, rather than that "communication failed" in some general sense.

In our example, the word "partner" is a sign that stands for an external organization with which we have a business relationship.

Level What it concerns Dietz and Mulder's grouping
Social world Norms, roles, and commitments Performa
Pragmatics Intent: why something is said and what it commits anyone to Informa
Semantics Meaning: what a sign refers to Informa
Syntactics Structure: grammar, protocols, data formats Forma
Empirics Encoding: letters, codes, bits Forma
Physical world The medium: marks, sound, signals Forma (as its carrier)

How it applies: a database checks structure, and by Date and Darwen's account that is all it can do (see below); meaning and commitment have to come from somewhere else. A system can be well formed at every lower level and still carry the wrong meaning, as "partner" did. AI systems do not escape the ladder either. They handle form and content, but commitments sit at the social level, where only people can be held to account. In the example I provided, the syntax and encoding were fine but the semantics drifted (i.e. changed from external organizations to include internal organizations), and the social commitment layer also changed, since owned and operated brands did not require agreements or payouts.

Peirce: Meaning Is in the Consequences

Charles Sanders Peirce held that meaning is triadic: a sign, the object it stands for, and the interpretant, which is the effect the sign produces (an action, a response, or a further sign). His pragmatic maxim, from "How to Make Our Ideas Clear" (1878), says that our idea of anything is our idea of its practical effects; his own example there is what it means for a diamond to be hard. For Peirce the final interpretant is a habit: a stable way of acting under certain conditions.

He also held that any interpretation can be wrong (fallibilism), and that meaning is stabilized over time by a community of inquiry that tests and corrects interpretations against experience.

How it applies: a term in a schema means what it leads people and systems to do. A definition that nothing ever queries, enforces, or decides with has little meaning in practice, and organizations rarely have the community of inquiry Peirce assumed for those small day-to-day business-as-usual changes an organization subtly encounters. People move on, priorities change, and no one is responsible for asking whether a term still means what it was defined to mean, so drift goes uncorrected.

Wittgenstein: Meaning Is Use

Ludwig Wittgenstein held that for a large class of cases the meaning of a word is its use in the language (Philosophical Investigations, §43). Words get their meaning in particular activities, which he called language games, and definitions do not govern their own application. He was describing how meaning works, not prescribing how it should.

How it applies: a term defined in a governance document means what people do with it. Semantics is intended at modeling time and tested and corrected in use. In the working example, partner no longer meant "an external organization with which we have a business relationship.". Wittgenstein might say the game changed.

Date and Darwen: The Predicate Lives Outside the System

C.J. Date and Hugh Darwen treat each relation variable (roughly, a table) as standing for a predicate. A sentence with places to fill, such as "Supplier S is located in City C." Each row asserts one instance of that predicate. The system checks constraints as its approximation of the predicate; but the meaning of the predicate itself lives outside the system (The Third Manifesto).

How it applies: a conceptual schema, in the sense I use it, is a structured specification of the distinctions a business makes, stated as predicates and the rules over them, so that people can refer to the same things, reason about them, and act together. It records the meaning that has been agreed; but it is not the meaning itself; but again, that is consistent with Wittgenstein (meaning is fixed in use) and with Peirce (it shows in consequences). The schema is where agreed meaning is written down so it can be checked. In the example, the predicate was "P is an external organization with which we have a business relationship", yet when owned and operated brands were added, every constraint still passed.

Dietz and Mulder: Commitments Bring Facts Into Existence

Jan Dietz and Hans Mulder build Enterprise Ontology on three foundations: Stamper's ladder, Habermas's theory of communicative action, and Bunge's ontology. The two most important insights from my perspective were the following:

First, model only what is essential. The essential model concerns the commitments the organization makes and the facts they produce, not the systems and documents that support them, so it changes when the organization's commitments change, not whenever its implementation does. This means the essential models change less frequently than implementation models.

Second, facts come into existence through commitments. In their transaction pattern, one party requests a result, another promises it, delivers it, and the first accepts; through the accept act, the result starts to exist. What travels between the parties is only the signal; the meaning, intent, and commitment are reconstructed by each side, and if any level is misaligned, the commitment one party intended is not the one the other understood.

How it applies: an organizational fact is not just a stored value; it is something someone with standing has accepted. Concepts anchored to those commitments change deliberately and visibly when the commitment changes, rather than drifting through informal use. In my view, Dietz overstates the case when he describes the essence as simply there to be revealed; I treat it as agreed and recorded, which is closer to Peirce and Wittgenstein. In the example, there are commitments specific to a partner that do not apply to internal brands. The essence of the organization changed in this case but the conceptual schema was not explicitly changed with it.

What This Means for Enterprises

The practical side of this, how to establish that structure before deploying AI. I will explore methods to do just this in future essays.

Open Question

A language model does appear to produce an interpretant in Peirce's sense: a sign causes a response, and the response can become a sign for the next system or person. Whether that is a full triadic sign relation or a partial one, and what follows for AI governance, is a question this series will return to.

References