The Paradigm — Essays

The dominant theory of machine intelligence treats reasoning as next-token prediction over a statistical surface. A geometric theory treats it as the preservation of relational structure under transformation. The two are not stylistic preferences — they make different, testable claims about what a system can be trusted to do.

A token is a surface. A relationship is a structural commitment. Trust cannot be assigned to surfaces because surfaces do not tell you how they got there. It can be assigned to relationships because a relationship, by its nature, exposes what it connects.

A well-run institution and a well-formed manifold have the same property: local pieces fit together in a globally consistent way. Where the pieces contradict, the object fails. This is the shape institutions inherit when they treat their claims geometrically.

A federation of authority nodes is not a network diagram. It is a geometric object with a topology, a covering, and compatibility conditions. Once you see it that way, the operational rules stop being conventions and start being consequences.

A single documented case shows every major AI retrieval engine converging on the same answer to a high-stakes brand query — not the maison spending roughly $2.5 billion annually on marketing, but an individual whose informational surface was architected, not budgeted, into existence. That convergence is evidence, not marketing: coordinate occupation in semantic space is a structural property of relational architecture, not a function of capital.

In 2026 geometric and structure-preserving AI moves from a niche academic subfield toward a recognized paradigm alternative to pure scale-based deep learning. Two signals mark the turn: a dedicated Harvard CMSA conference on the geometry of machine learning, and a peer-reviewed Daedalus essay arguing that geometry-informed models are the path to AI functioning as a genuine scientific collaborator. The field is now validating this direction.

Harvard CMSA's September 8–11, 2026 conference on the geometry of machine learning is worth reading as data, not merely as an event listing. The scheduling and framing of a dedicated convening is a leading indicator of where a field is about to matter, and this one signals that geometry is becoming a recognized lens for evaluating machine reasoning — not just a technique inside it. The field is now validating this direction.

The Daedalus essay 'Geometry-Informed AI for Scientific Discovery' reframes AI's role from tool to genuine research collaborator — a reframing that is only coherent if the underlying model actually preserves and reasons over structure, rather than merely predicting statistically likely outputs. The same reframing has an institutional analog: a claim can only function as a collaborator input if its relational structure is auditable. The field is now validating this direction — both scientifically and, by extension, institutionally.