You have a fixed-capacity processor (your working memory holds about 4 items) and an enormous store of past experience (your long-term memory holds, effectively, everything you've ever attended to). When you need to think about something you're not currently thinking about, you retrieve it.

This is exactly the architecture we've been describing for language models. A bounded context window. An external store. Retrieval that connects them.

The brain got there first — and the hippocampus is the indexing system.

The parallel

Working memory is the context window. It's small, active, and everything in it is immediately available for computation. Baddeley's classic estimate is 4 ± 1 items. Whether you model it as slots, as a focus of attention, or as the currently active subset of long-term memory, the effective capacity is bounded. Fixed.

Long-term memory is the external store. It's vast, persistent, and not directly available for computation — you have to retrieve from it. The retrieval process takes time (hundreds of milliseconds) and is not always successful. It's a store you read from and write to, with latency.

The hippocampus is the index. It doesn't store the memories themselves — those live in neocortex. It stores the pointers: associations, contexts, temporal tags that let you navigate to the right memory when you need it. Damage the hippocampus and you can still hold things in working memory, still access old well-consolidated memories, but you cannot form new indexed associations. You cannot organize new experience for later retrieval.

This is the Library Theorem in biological hardware.

What the theory predicts

If the brain implements something like the Library Theorem, several things should be true:

Retrieval time should scale logarithmically with memory size. Not linearly. If hippocampal indexing works like a tree, then the time to retrieve a memory should grow as log N, where N is the number of stored items. There is evidence for this: recognition memory remains fast even as the number of studied items grows into the thousands. The set-size effect is weak — much weaker than linear scan would predict.

Hippocampal damage should produce the linear-scan deficit. Without the index, retrieval should degrade to something like sequential search — slow, effortful, and increasingly worse as the store grows. Amnesic patients with hippocampal damage show exactly this pattern: they can recognize familiar items (relying on a familiarity signal that doesn't require indexing) but cannot recollect specific episodes (which requires navigating the index).

Sleep should be index maintenance. Memory consolidation during sleep, particularly during sharp-wave ripples in the hippocampus, is well documented. The Library Theorem suggests a specific computational role: sleep is when the index gets reorganized. Replay isn't just strengthening memories — it's re-indexing them into a more navigable structure. This predicts that disrupting replay should impair organized retrieval more than raw storage.

Cognitive maps should be a special case. The hippocampus is also the brain's spatial navigation system. Place cells, grid cells, head-direction cells — these form a map. But a map is an index: a structure that lets you navigate efficiently to a location without visiting every intermediate point. Spatial navigation and memory retrieval may be the same computation: indexed lookup over a structured store.

Extended cognition

The parallel goes further. When you write something in a notebook, you are externalizing memory. When you organize your notes, you are building an external index. When you look something up, you are performing indexed retrieval with a physical store.

The extended cognition hypothesis (Clark & Chalmers, 1998) argued that cognitive processes can extend beyond the brain into the environment. The Library Theorem gives this argument formal teeth: external indexed memory doesn't merely supplement cognition. It changes the computational class. A brain with organized external notes is literally more powerful than a brain without them — not as metaphor, but as mathematics.

This is why writing was a civilizational threshold. Not because it preserved information (oral cultures preserved plenty), but because it made information indexable. You could organize it. Navigate it. Build on it without holding it all in your head. The Library Theorem quantifies the advantage: exponential.

The hippocampus is a biological index over a biological store. It solved the Library Theorem long before we proved it.

This paper connects the Library Theorem (Paper 1) and externalization completeness (Paper 7) to hippocampal indexing, memory consolidation, and cognitive maps. Literature review in progress. Target venue: Trends in Cognitive Sciences or Nature Neuroscience.