Cubic millimetre of brain mapped at nanoscale resolution
A one-cubic-millimetre fragment of human cortex has been mapped at nanoscale resolution, revealing tens of thousands of cells and 150 million synapses, along with strange, poorly understood microstructures. Commenters are struck by both the technical feat — including petabyte-scale data, advanced electron microscopy, and machine learning–based 3D reconstruction — and what it implies about the staggering complexity of even a tiny piece of brain tissue. The work fuels debate over how far we are from truly understanding cognition or emulating a human brain in silicon, and whether current AI approaches bear any meaningful resemblance to biological intelligence.
Reactions to the visualization
- Many find the interactive 3D viewer stunning and immersive; zooming and neuron highlighting are praised.
- Several report strong visceral reactions: awe, eeriness, “spiritual” feelings, or discomfort at seeing the physical substrate of mind.
- Some are unsettled by the idea that this tissue once belonged to a living person and that “this mess of fibers” implements consciousness.
Brain complexity and intelligence
- The density and tangled connectivity in just 1 mm³ reinforces a sense that brains are extraordinarily complex, perhaps the most complex known systems.
- Debate over whether a brain can ever fully understand itself; some argue we lack even a clear definition of “understanding.”
- Comparisons are made between human and bird brains; bird brains are noted as more neuron-dense and tightly packed, with implications for speed and efficiency but constraints on size.
Imaging, slicing, and reconstruction
- Sample prep: staining, embedding in resin, then ultra-thin slicing with diamond knives and automated tape collection (microtome/ATUM).
- EM resolution is discussed in the context of Feynman’s “plenty of room at the bottom”; progress is acknowledged but atomic-level detail is still challenging.
- Reconstructing 3D volumes and segmenting cells/synapses relies on ML; this is state-of-the-art but still requires laborious human proofreading, and only part of the dataset is fully checked.
Data volume, storage, and whole-brain emulation
- Back-of-the-envelope extrapolation: mapping a whole human brain at this resolution would require zettabytes of storage, far beyond any single system today.
- Some argue much of this data might be redundant for “mind uploading”; others respond we don’t yet know what can be discarded and even simple AI models are poorly understood.
- There is skepticism that whole-brain EM or uploads will arrive soon; even C. elegans remains poorly understood despite its tiny, mapped nervous system.
AI vs biological brains
- Rough parameter comparisons suggest human brains have orders of magnitude more “parameters” than current LLMs, even under very simplified assumptions.
- Others note neurons, dendrites, and even proteins perform complex local computation, so these estimates are likely still huge underestimates.
- Some see LLMs as surprisingly efficient per “connection”; others highlight their lack of embodiment, math rigor, and genuine understanding.
Philosophical and existential themes
- The dataset triggers reflections on consciousness as “just” physical wiring versus something more; some feel this diminishes human “magic,” others say the substrate doesn’t lessen the wonder.
- There is broad agreement that understanding intelligence will require advances in complexity theory and that both connectomics and AI alone are insufficient.