A Preface
Six Essays on Compression · Preface · How to read the next six
This series is about one idea, looked at from six angles. Communication and thinking are both kinds of compression and decompression, performed by finite processors using shared codes. Almost every interesting failure mode in human conversation, in clinical handoffs, in modern AI systems, is a failure of the code.
The root premise is simpler. No finite agent can carry reality itself. It can only carry a compressed representation of reality, shaped by the task in front of it. The question is never whether compression happens; the question is what gets preserved, what gets discarded, and whether the next agent can reconstruct what matters.
Compression fails in three ways. The wrong thing gets dropped: an action-changing fact never reaches the receiver. The wrong thing gets filled in: the receiver reconstructs a confident picture from a confident message, and the picture is wrong. Or the gap is never checked: no one verifies whether the receiver's picture matches what the sender meant. The work of safe communication, in any high-stakes setting, is the architecture that catches all three.
I came to the idea from medicine, where the failure modes are loud. A clinical sign-out is a wildly compressed message: a dozen words in place of a person, handed off in seconds at change of shift. The compression works when the receiver shares the code; it fails, sometimes catastrophically, when they don't. The same compression happens at every other scale of human communication. The same compression happens, with new failure modes, every time we put an AI in the loop.
Six essays, in order:
I. Compression Engines. Before any message is compressed, the world is. Perception, memory, language, science: all of them are codecs running on finite hardware.
II. The Receiver Is the Codec. Efficient communication is a property of the receiver, not the sentence. The work is to model what is already in the other head.
III. The Minimum Sufficient Message. What it looks like when a message has been tuned to a single point: every word still earning its place, nothing kept that the receiver could supply themself.
IV. The Algebra of Thought. To think is to manipulate compressed handles for things too large to hold whole. We are algebra all the way down.
V. A Third Processor. Communication is what happens when two finite processors share a codec. AI is another processor in the chain, and most "AI failures" are codec mismatches, not failures of intelligence.
VI. Closing the Loop. What a principled architecture looks like: six properties that high-stakes compression needs in order not to fail silently.
If you are a clinician, the first three essays are about the work you already do, named for what it is; the last three are about what is being built around you, and why most of it isn't built right yet. If you build AI, the first three are the missing background most ML papers assume but rarely write down; the last three give you a checklist your stack is probably failing somewhere. If you are neither, the whole sequence is a way of looking at how minds communicate across rooms, across professions, across substrates, that I have not seen put in one place before.
Each essay stands alone. The order is what I would recommend, but they are not chained. Pick whichever title looks most interesting and start there.
There is also a closing piece, the Coda, that puts the architecture from Essay VI into a single diagram and a checklist you can carry into a meeting. If you only have time for two pieces, read Essay I and the Coda.
Six Essays on Compression · Preface · I · II · III · IV · V · VI · Coda · Postscript