Representational Transformations

In the AI Story

The navigation team's cognitive work consists of a chain of representational transformations: visual scene becomes bearing, bearing becomes verbal report, report becomes log entry, entry becomes chart line, multiple chart lines intersect to produce a position fix. Each step is a translation between media with different properties — continuous visual to discrete numerical, spoken verbal to written textual, numerical to geometric.

The AI-augmented builder's desk operates through a representational chain of dramatically different structure. The builder's intention — a mental representation partly tacit, partly visual, partly kinesthetic — is transformed into natural language. This transformation is already cognitively demanding: aspects of the intention that resist linguistic expression are filtered out. The AI then transforms the linguistic representation into code, a formal representation whose properties differ from natural language along nearly every dimension. The code is compiled to produce a running system — another transformation, this one performed by the machine without human involvement.

Each eliminated transformation in the compressed chain is an eliminated source of delay and noise — a genuine efficiency gain. But each eliminated transformation is also an eliminated cognitive checkpoint, a lost opportunity for a specialist's trained perception to detect a problem no other agent in the system would catch. This is the collapse of translation chains that Segal describes in The Orange Pill — liberation from friction, and the invisible loss of distributed intelligence that the friction also produced.

The materiality of representations is the point. A chart is not information about the coastline; it is a physical artifact with specific properties that shape what can be done with it. Orient the chart to match the ship's heading, and translation between chart and visual field becomes cognitively cheap. Leave it misaligned, and every translation introduces error. The convention of orientation is a cognitive design choice that generations of navigators refined through experience with the alternative.

Origin

Hutchins developed the concept of representational transformation through close observation of navigation practice. Watching a bearing move from pelorus observation through verbal call, written record, and chart plot revealed that the information was not simply being preserved across media — it was being actively transformed, with each transformation involving cognitive work that the previous form had not required.

The theoretical lineage draws on Shannon's information theory, Peirce's semiotics, and the cognitive-anthropological tradition. But Hutchins's innovation was insisting that the transformations be studied as situated, physical operations rather than as abstract information processing — that the medium matters, and matters cognitively.

Key Ideas

Medium as computational substrate. Different media support different cognitive operations — the numerical affords arithmetic, the geometric affords spatial reasoning, the visual affords pattern recognition.

Transformation as checkpoint. Each translation between media requires information to be examined in its new form, creating natural opportunities for error detection.

The cost of compression. When AI collapses multiple transformations into a single human-to-machine step, the speed gain comes with the loss of intermediate checkpoints.

Representational diversity as design principle. Reliable cognitive systems employ multiple formats that make different properties salient and catch different categories of error.

Materiality of cognition. Representations are physical objects in specific media; their material properties shape the cognitive work they can support.