Ontology: Preparing the World to Be Seen

How Alex Karp Productized an Obscure Philosophical Concept and Turned it into one of the most Consequential Technologies of Our Time

In 2002, Alex Karp defended his doctoral dissertation in German at Goethe University in Frankfurt. He had been living in Germany since the mid-90s, and he wanted the argument conducted in the language it was about.

Karp’s thesis was about how aggression moves through a culture inside ordinary speech, absorbed rather than erupting, and how a person can break a rule in a manner that demonstrates he knows the rule from the inside. His empirical case was postwar German resentment of the Jews, what the Frankfurt sociologists called secondary antisemitism: the anger of the guilty at whoever holds the receipt.

The title was Aggression in der Lebenswelt (aggression in the lifeworld). The subtitle promised an extension of Talcott Parsons, the mid-century American sociologist who modeled society as a system of functions, by describing the relation between jargon, aggression, and culture.

Jargon was Adorno’s word. Theodor Adorno had returned to Frankfurt after the war and written The Jargon of Authenticity to show how a debased Heideggerian vocabulary of rootedness and depth could circulate as respectable German while doing other work underneath. The concept of the Life-world was one introduced by Wilhelm Dilthey, and became a major influence on Husserl and Heidegger, as well as cultural anthropologists like Clifford Geertz. The idea of a life-world is that you have to understand beings in their habitat.

Karp had gone to Frankfurt hoping to work with Jürgen Habermas, Adorno’s successor and the most consequential living German philosopher. He attended Habermas’s colloquium, but Habermas did not supervise the dissertation. Its official readers were the social psychologist Karola Brede and the sociologist Hans-Joachim Busch.

Today, Karp is recognized not as an academic specialist in German social thought but as the wavy-haired billionaire in a white t-shirt, who wields ancient Japanese swords, runs long distances, and summons retail traders to punish short-sellers on Squawk Box.

Within a few years Karp was running Palantir, whose software would eventually place an abstraction called the Ontology at its center.

The lifeworld, Lebenswelt, is a concept Habermas inherited from phenomenology and transformed into a theory of social communication. It names the tacit background that makes explicit communication possible: the meanings, assumptions, practices, and competencies people already share before anyone opens his mouth. We do not notice it because we are standing in it. It is the reason a shrug lands, a joke works, an order is understood without being spelled out.

Every institution has one. An airline has aircraft, crews, routes, maintenance protocols, safety requirements, and practical knowledge distributed across thousands of people who have never met. A library has authors, editions, subjects, citations, interpretive traditions, curatorial judgments, and communities of readers. Neither is a pile of data. Each is an environment of meaning in which certain actions make sense and others do not.

Palantir’s achievement was to render part of that background explicit and computable. Its Ontology binds the nouns of an institution to its verbs: what an aircraft, a shipment, a patient, a work order is, and also what can happen to it, who may act on it, under which rules, with what consequence. Objects acquire permissions. Relationships acquire operations. The map becomes kinetic.

Palantir answers a question the philosophers left standing: How does a shared world of meaning become a shared world of action?

Twenty years later, Karp supplied a political answer. In The Technological Republic, written with Nicholas Zamiska, he argues that the West has not lost its technological abilities so much as the confidence to direct them toward collective ends. Silicon Valley’s retreat into advertising, entertainment, and consumer convenience is, on this account, a symptom of a deeper failure: a society uncertain of what it believes cannot tell its engineers what is worth building.

The argument extends his dissertation’s central intuition. Aggression does not simply erupt from outside the social order. It circulates within culture and can even become a force of integration. Societies are held together not by rational agreement alone but by affect: solidarity, rivalry, humiliation, taboo, sacrifice, and the distinction between those who belong and those who threaten the shared world.

The Technological Republic scales that problem from ordinary language to geopolitics. How can the West recover enough common purpose to organize its power? How can a liberal society acknowledge the aggression necessary for self-preservation without allowing aggression to become its organizing principle? If the dissertation asked how culture gives aggression a social form, the book asks how a republic gives power a legitimate purpose.

The conventional story about AI concerns models. Bigger models ingest more text, recognize more patterns, produce more fluent answers. Access to information turns out to be a different thing from understanding a particular world, from participating in a lifeworld. A base model can tell you that a document exists, summarize it, name its author. But it has no native way to know which interpretive tradition surrounds it, which quarrel it records, which obscure text it answers, or why a specific reader should meet it this morning. That knowledge lives in distinctions, provenance, institutional memory, and judgment. It lives in an ontology.

A system does not find the needle merely because it has been handed the haystack. Someone first has to establish what could count as a needle.

Heidegger understood this philosophically.

Being and Time distinguishes the ontic from the ontological: ontic inquiry concerns particular beings and facts—what exists, what happened, what properties a thing has; ontological inquiry asks after the structures that let anything show up as the kind of thing it is. Facts are indispensable. Facts do not interpret themselves. Every inquiry starts from a prior sense of what it is looking for and why an answer would matter.

The later Heidegger became preoccupied with a specific mode of disclosure produced by modern technology. He called it Gestell, “Enframing,” a way of revealing in which everything shows up as measurable, orderable, available for use. The forest appears as lumber, the river as head of water, the person as human capital. Things stop confronting us in their own presence and become standing-reserve, stock held ready.

Applied to a computational ontology, the critique is this: Once a manuscript or a tradition or a community or a man becomes legible to the system, it can be classified, scored, optimized, and acted upon. The risk is not confined to miscategorization. The deeper risk is that the thing begins to appear as the object the system represents, and nothing else.

Palantir makes this vivid because its Ontology refuses to stop at description. The aircraft can be reassigned. The shipment can be rerouted. The target can be selected. The world is disclosed as a field of available intervention.

This is also Palantir’s value proposition to Western governments. A modern state rarely lacks information. It lacks the ability to convert information distributed across agencies, databases, security regimes, and chains of command into common knowledge and coordinated action. Palantir gives the state a representation in which it can perceive itself as a coherent actor.

Here ontology becomes the hinge between what The Technological Republic calls soft belief and hard power.

Soft belief determines what a political community values, which threats it recognizes, which actions it considers legitimate, and what sacrifices it is prepared to make. Hard power turns those commitments into intelligence, logistics, industrial capacity, deterrence, and force. The ontology translates one into the other. It renders a political interpretation of the world operational.

Palantir is therefore not simply selling governments better data integration. It is selling state capacity: the ability of a political community to bind its knowledge, judgment, and machinery into action. The company’s alignment with the United States and its allies is not branding applied to an otherwise neutral product. It belongs to the product’s meaning. Palantir exists to make the West capable of defending the form of life from which the software itself emerged.

Karp read Adorno on the language through which aggression hides inside culture, studied Habermas’s account of the shared world that makes action intelligible, and then spent twenty years building a machine for making that world explicit. Palantir can be understood as an extraordinarily complete instrument of enframing. But it can also be understood as Karp’s correction to Frankfurt: shared meaning is politically defenseless unless it can be converted into organized power.

This is where Karp parts company with Habermas. Habermas placed his hope in communicative action: the possibility that legitimate purposes might emerge through public reasoning among equals. Karp retains the need for legitimate common purposes but doubts that communication is sufficient to preserve the world in which communication occurs. A republic must give reasons. It must also deter enemies, secure infrastructure, manufacture weapons, and survive.

Communication may legitimate the republic. In Karp’s account, hard power preserves the conditions under which communication remains possible. The Frankfurt School’s rejoinder would be that systems built to defend a lifeworld have a persistent tendency to colonize it.

Adorno had a name for the second half of the problem. He called it identity-thinking: the habit of a concept presenting itself as exhaustive of the object it names. The object always spills over. The person exceeds the demographic category, the book exceeds its metadata, the tradition exceeds the doctrines attributed to it. What Adorno wanted preserved was the remainder, the nonidentical, the part that will not go into the box. Every ontology generates that remainder by construction. It draws boundaries, fixes identities, decides which differences count. Whatever it cannot represent becomes progressively harder to see.

The tempting conclusion is that we should decline to build ontologies at all and let things disclose themselves unclassified.

There is no ontology-free search. Every archive has a catalog. Every curriculum has a canon. Every interface privileges some paths over others. Every embedding model encodes a metric of similarity, which is to say a theory of what resembles what. Every search engine decides which features are relevant. The apparently universal base model carries an implicit ontology inherited from its training corpus, its architecture, and whatever its reinforcement rewarded. Refusing to build one leaves these decisions in place and moves them out of view.

The live distinction runs elsewhere. Between ontologies that are implicit and ontologies that are explicit. Between ontologies that are closed and ontologies that expect to be revised. Between systems that conceal their exclusions and systems that publish them. Between a map that prescribes the conclusion and a map that prepares the conditions for finding one.

A bad ontology confuses itself with the world. It hardens inherited categories, suppresses ambiguity, and admits only the questions its designers already knew how to ask.

A good one remembers that it is partial. And enables openness to a changing world. It is self-revising. It keeps provenance. It separates evidence from inference. It tolerates rival classifications. It records what it does not know.

Maria Montessori used the phrase “the prepared environment” for a room arranged so a child can act and explore without being told what to find. The arrangement does not determine the discovery. It puts real possibilities within reach and then gets out of the way. An ontology can work the same way on a body of knowledge. It supplies coordinates and leaves the destination open.

Heidegger is right that every revealing is also a concealing. Concealment is a reason to build systems that hold onto the difference between what they display and what is there.

P.S. - The institutions of the wisdom economy pose different problems from the institutions for which Palantir was built, yet the architecture is similar and requires a point of view on your life-world. That is our work at Lightning. We work with libraries, museums, and cultural institutions to define and map this life-world, to enable better surfaces for inference and insight, and to build internal tools and outward facing products that bring your data to life. If this describes you, get in touch: zohar@lightningstudios.ai