CONCEPT
The Fishbowl of the Instrument
Every instrument reveals and conceals simultaneously—the set of limitations that determines not just what it shows but what it hides, and the capacity to distinguish instrument artifacts from genuine findings is the condition of being augmented rather than deceived by the tools one uses.
Vannevar Bush, who spent his early career designing
analog computers for solving differential equations, understood the fishbowl of computational instruments with a precision most users of modern computing have never acquired. The differential analyzer Bush designed between 1927 and 1931 could solve equations that were analytically intractable—but it could handle only certain classes of equations, and the researchers who depended on it learned, gradually and unconsciously, to stop asking questions the instrument could not address. The fishbowl is not a deficiency; it is a structural feature of instrumentation itself. The act of choosing an instrument is the act of choosing what to see and what to be blind to. The
large language models at the center of the AI transition have a fishbowl defined by their training data, architecture, and reinforcement learning process: they can generate text that is stylistically diverse and argumentatively sophisticated, but they cannot generate text