Foundations: The Theory Under Modern AI

Connectome

Q: Where the Word Comes From

Definition: A connectome is a complete map of the connections between neurons in a nervous system — which neuron connects to which, through which synapse, in which direction, and how many times. It is structural rather than functional: the circuit schematic for a brain, not a recording of the brain running. TL;DR: Several complete connectomes now exist, and the scope has grown with each one: a 302-neuron roundworm (1986), a whole adult fly brain (2024), and whole fly central nervous systems in 2026 — female in June, then the first complete male map in September with 166,700 neurons. All of them show the same thing: wiring gives you paths, never strengths. Build an AI app free → The term was coined independently in 2005 by Olaf Sporns and by Patric Hagmann, by analogy with genome. The analogy is useful in one specific way and misleading in another. Useful: both a genome and a connectome are reference resources that thousands of later experiments can query without re-deriving them. That is the Human Genome Project pattern, and it is the real argument for the expense. Misleading: a genome is a sequence of instructions an organism is built from. A connectome is a graph that resulted. One is closer to a program, the other closer to a circuit board.

Q: What a Connectome Shows and Hides

flowchart TB C["The connectome"]:::pink G["GIVES YOU morphology · location · cell type direction · connection counts neurotransmitter clues"]:::green H["DOES NOT GIVE YOU effective synaptic strength neuromodulatory and hormonal state plasticity · learning history gene expression · live activity"]:::orange B["Behaving organism"]:::blue P["+ dynamics + state + plasticity + body + environment + feedback"]:::muted C --> G C --> H G --> P P --> B classDef pink fill:#2a1225,color:#ff8fa3,stroke:#ff2d60,stroke-width:2px classDef blue fill:#0d1b2a,color:#7dd3fc,stroke:#38bdf8,stroke-width:1.5px classDef green fill:#0d1a0e,color:#86efac,stroke:#22c55e,stroke-width:1.5px classDef orange fill:#1a140a,color:#ffb74d,stroke:#e65100,stroke-width:1.5px classDef muted fill:#1e1e2e,color:#94a3b8,stroke:#475569 The cleanest framing is a computing one: a connectome is a netlist, not a running program. A netlist is a complete, typed, directed description of components and their connections, and it is silent about behavior until you supply timing, state and input. See Von Neumann Architecture for the layer below it.

Q: How AI Builds a Connectome

flowchart LR S["Slice + electron microscope"]:::muted --> M["Machine learning: align, segment, trace, detect synapses"]:::pink M --> P["Human proofreading and annotation (years)"]:::orange P --> G["3D connectome graph + database"]:::green classDef pink fill:#2a1225,color:#ff8fa3,stroke:#ff2d60,stroke-width:2px classDef green fill:#0d1a0e,color:#86efac,stroke:#22c55e,stroke-width:1.5px classDef orange fill:#1a140a,color:#ffb74d,stroke:#e65100,stroke-width:1.5px classDef muted fill:#1e1e2e,color:#94a3b8,stroke:#475569 The core technique is flood-filling networks — introduced in 2016 and validated in Nature Methods in 2018 — which start at a single pixel and identify every other pixel belonging to the same object. The later PATHFINDER system automates more of the assembly. The honest part: proofreading and annotation still take years of expert human effort. AI produced the candidate; humans produced the truth. That division is the same shape as every reliable AI system shipping today, and it is worth remembering whenever a headline says AI mapped a brain.

Q: Is the fly connectome the largest brain map?

It is the largest by number of neurons, which is the qualifier the researchers use. MICrONS mapped a cubic millimeter of mouse visual cortex in 2025 with more than 200,000 cells and roughly half a billion synapses — larger by synapse count, but a fragment rather than a complete nervous system.

7 min read

On this page (13)

Definition: A connectome is a complete map of the connections between neurons in a nervous system — which neuron connects to which, through which synapse, in which direction, and how many times. It is structural rather than functional: the circuit schematic for a brain, not a recording of the brain running.

TL;DR: Several complete connectomes now exist, and the scope has grown with each one: a 302-neuron roundworm (1986), a whole adult fly brain (2024), and whole fly central nervous systems in 2026 — female in June, then the first complete male map in September with 166,700 neurons. All of them show the same thing: wiring gives you paths, never strengths. Build an AI app free →

Where the Word Comes From

The term was coined independently in 2005 by Olaf Sporns and by Patric Hagmann, by analogy with genome. The analogy is useful in one specific way and misleading in another.

Useful: both a genome and a connectome are reference resources that thousands of later experiments can query without re-deriving them. That is the Human Genome Project pattern, and it is the real argument for the expense.

Misleading: a genome is a sequence of instructions an organism is built from. A connectome is a graph that resulted. One is closer to a program, the other closer to a circuit board.

The Major Maps So Far

Year	Subject	Neurons	Connections	Scope
1986	C. elegans roundworm	302	~7,000	Whole nervous system
2020	Fly "hemibrain"	~25,000	~20 million	Most of the central brain
2024	Adult female fly	139,255	54.5 million	Whole brain
Jun 2026	Female fly	—	~10⁸	Brain + nerve cord
Sep 2026	Male fly	166,700	~125 million	Brain + nerve cord

Only the 2020 hemibrain is partial. It stays in the table because it bridged the worm and the whole fly. Three qualifiers are frequently misreported:

302 is the hermaphrodite count. White, Southgate, Thomson and Brenner mapped C. elegans in 1986. The male worm connectome only arrived in 2019 (Cook et al., Nature).
The "hemibrain" is half of the central brain, not half of a fly brain. It excludes the optic lobes. The whole fly brain holds over 100,000 neurons.
The female whole-CNS map came first, in June 2026. The male dataset is the first male one, which is what makes a sex comparison possible.

What a Connectome Shows and Hides

The cleanest framing is a computing one: a connectome is a netlist, not a running program. A netlist is a complete, typed, directed description of components and their connections, and it is silent about behavior until you supply timing, state and input. See Von Neumann Architecture for the layer below it.

The Effectome

This limit has a name in the literature. A 2024 Nature paper, The fly connectome reveals a path to the effectome, states it exactly: the connectome "specifies the synaptic paths by which neurons can affect each other, but not how strongly they do affect each other in vivo."

The proposed fix is not a better map. It is perturbation — pairing the wiring diagram with stochastic optogenetic experiments and using the connectome as a prior to recover causal strengths.

Structure gives you the paths. Only the experiment gives you the strengths. That principle generalizes well beyond neuroscience, and it is why evals matter more than architecture diagrams when assessing any complex system.

What the Finished Maps Revealed

The most consequential result is architectural rather than numeric. The June 2026 whole-CNS paper found:

Motor, endocrine and visceral effectors are driven mainly by sensory neurons in the same body part, forming local feedback loops.
Ascending and descending neurons join those loops into behavior-centric modules.
Brain regions for learning and navigation supervise the modules rather than commanding them.

The authors describe the architecture as "distributed, parallelized and embodied, reminiscent of distributed control architectures in engineered systems." There is no central controller. This maps onto how multi-agent systems and autonomous agents are designed, and it is empirical rather than theoretical.

How AI Builds a Connectome

The core technique is flood-filling networks — introduced in 2016 and validated in Nature Methods in 2018 — which start at a single pixel and identify every other pixel belonging to the same object. The later PATHFINDER system automates more of the assembly.

The honest part: proofreading and annotation still take years of expert human effort. AI produced the candidate; humans produced the truth. That division is the same shape as every reliable AI system shipping today, and it is worth remembering whenever a headline says AI mapped a brain.

Common Questions

Is the human connectome finished?

No. Only a coarse, region-level map exists, made with diffusion MRI rather than electron microscopy. A neuron-level human connectome is roughly five orders of magnitude beyond current methods — estimates place the adult human brain near 86 billion neurons.

Is the fly connectome the largest brain map?

It is the largest by number of neurons, which is the qualifier the researchers use. MICrONS mapped a cubic millimeter of mouse visual cortex in 2025 with more than 200,000 cells and roughly half a billion synapses — larger by synapse count, but a fragment rather than a complete nervous system.

Can you simulate a brain from its connectome?

Partially, and only with added assumptions. A 2024 whole-fly-brain model built from the connectome plus predicted neurotransmitter identities correctly predicted the neurons required to start feeding. Every such model supplies parameters the connectome does not contain.

Why map a fly instead of a human?

Scale and tractability. A fly nervous system is small enough to finish and complex enough to produce navigation, learning, courtship and feeding. It has also been a genetics workhorse for a century, so there is deep prior work to validate a map against.

What is the difference between the male and female fly connectome?

The published abstract states that sex-specific and dimorphic neurons are concentrated in higher brain centers while the sensory and motor periphery is largely isomorphic, and that dimorphic neurons reroute information across sexes. The input and output hardware is nearly identical; the differences sit in the integrative middle and work by changing routing.