I Reinvented Critical Realism by Accident
Building an AI truth engine walked me into a fifty-year-old philosophical school. I didn't know it existed until last week.
The axiom
I started Korbia with a three-part claim.
There is only one reality. Everyone operates from a perspective. The usefulness of a perspective is its proximity to reality.
This was not philosophy. It was a product requirement.
Korbia is supposed to be the system of record for a business. That means it has to represent something real.
A single underlying state that sits outside any one person's view. Otherwise there is nothing to be accurate about.
I wrote the axiom, moved on, and started building.
The thread
A month later I stumbled across Alfred Korzybski. He is the general semantics guy from the 1930s.
His famous line is "the map is not the territory." That line parked itself in my head.
It was my axiom, compressed into seven words.
So I pulled on the thread. What makes one map better than another?
How many maps can a territory support? Is the territory stable enough to be mapped at all?
Those questions led me into the philosophy of science. Which led me to Roy Bhaskar.
The collision
Bhaskar founded a school called Critical Realism. Its foundation rests on three claims, published across two books in the 1970s.
- Ontological realism. There is one reality, independent of anyone's knowledge of it.
- Epistemic relativism. All knowledge comes from a perspective.
- Judgmental rationality. Some perspectives are better than others. Proximity to reality is a real property, not a matter of taste.
Those are my three claims. Same order. Same substance. Fifty years older.
Why this isn't a coincidence
I did not derive a position adjacent to Critical Realism. I derived Critical Realism.
The reason is not that I am clever. It is that any serious attempt to build a knowledge product about a real business hits the same three constraints.
Reality has to be one thing, or there is no ground truth to approximate. Perspectives have to be real, or your users evaporate.
Proximity to reality has to matter, or you cannot tell a good record from a bad one.
Take any one away. The product breaks.
What this means for AI enterprise software
Most AI tooling being built right now is implicitly empiricist. It treats data as the thing.
Patterns in data are treated as knowledge. Confidence is treated as truth.
This works for narrow tasks. It does not work for systems that claim to represent a business.
A business is not the set of events its systems recorded. It is the underlying state that generated those events, and the mechanisms that keep generating them.
The events are the shadow. The state is the thing.
Critical Realism has a vocabulary for this. It calls it a stratified ontology. Three domains.
- The Empirical is what gets observed.
- The Actual is what happens, observed or not.
- The Real is the mechanisms and structures that produce what happens.
Every truth engine worth the name has to operate across all three.
The practical takeaway
If you are a founder or owner running a business, you are already doing Critical Realism. You just may not have the vocabulary for it.
You know there is one real state of your company. You know everyone on your team sees a different slice. You know some views are closer to true than others.
That is the three-part axiom. You live it every week.
The vocabulary helps. It tells you where your mental map of the business is over-claiming.
It tells you where your dashboards are flattening reality into pattern. It tells you which disagreements on your team are about perspective and which are about mechanism.
It is a diagnostic tool for running the company, not just a philosophy for building the software.