Neural Atlas · Dashboard

The eight regions, ranked by AI gap

Sort by AI gap to see where the difference between the human brain and the current state of the art is largest. Maturity values are editorial estimates anchored to the Stanford AI Index 2026 and the International AI Safety Report 2026.

Region	AI counterpart	Maturity	Gap↓	Direction
Hippocampus Medial temporal lobe (interior)	RAG · Vector DBs · Extended context · Agent memory	34/100	66	Large — no consolidation
Amygdala Limbic system · medial temporal lobe (deep)	Alignment · Constitutional AI · RSP · interpretability	42/100	58	Empirical — alignment is unsolved
Cerebellum Posterior-inferior cranial fossa	Humanoid robotics · Boston Dynamics · Figure · 1X · Apptronik	48/100	52	Wide — sim-to-real bottleneck
Prefrontal Cortex Frontal Lobe (anterior)	Reasoning models · Chain-of-thought · Extended thinking	54/100	46	Wide — emerging frontier
Basal Ganglia Deep cerebral nuclei (caudate, putamen, pallidum)	RLHF · DPO · RLVR · GRPO · DeepSeek R1 · π0	58/100	42	Closing — RL strong, selection fragile
Thalamus Diencephalon · deep central	GPT-4o · Gemini · V-JEPA 2 · audio-video fusion	72/100	28	Narrow — production-grade multimodal
Visual Cortex Occipital Lobe	Computer vision · ConvNets · ViT · SAM 2 · diffusion	78/100	22	Narrow but closing
Broca & Wernicke Inferior frontal & posterior superior temporal gyrus	Large language models · Transformer architecture	82/100	18	Production strong, grounding weak

Visual Cortex

78/100

Occipital Lobe

On narrow recognition tasks AI matches or exceeds humans. Robustness to distribution shift, adversarial noise and embodied vision still lag, but the gap is closing fastest of any cognitive function.

Deep dive →

Broca & Wernicke

82/100

Inferior frontal & posterior superior temporal gyrus

Language production is effectively mature. Long-horizon discourse coherence, factual grounding and pragmatic reasoning still trail competent humans, but the gap on canonical benchmarks is in single digits.

Deep dive →

Prefrontal Cortex

54/100

Frontal Lobe (anterior)

Frontier — pace of improvement is steep but raw capability still lags humans on open-ended planning, calibration of uncertainty and genuinely novel insight. ARC-AGI-2 has flipped from 'AI struggles below 50%' in early 2025 to 'AI now solves the majority' in early 2026.

Deep dive →

Hippocampus

34/100

Medial temporal lobe (interior)

Primitive — engineering substitutes work, but the system architecture is qualitatively unlike a hippocampus. The biggest open problem in modern AI: models that learn from experience without re-training.

Deep dive →

Cerebellum

48/100

Posterior-inferior cranial fossa

Developing — locomotion is solved on flat ground; dexterous manipulation, mobile manipulation in clutter and generalisation to new environments still trail mammals. 2026 is the year humanoids crossed the line from prototype to commercial deployment at automotive plants.

Deep dive →

Amygdala

42/100

Limbic system · medial temporal lobe (deep)

Developing — RLHF and Constitutional AI are production-grade for harmlessness on canonical refusals. The hard parts — robust refusal under jailbreak, scheming detection, alignment under distribution shift — remain partially solved at best.

Deep dive →

Thalamus

72/100

Diencephalon · deep central

Mature — native multimodal models ship at production scale across all frontier labs. Real-time audio-video understanding crossed sub-second latency in 2024 (GPT-4o ~320ms voice response). Cross-modal grounding still trails biology on edge cases, but the core capability is in production.

Deep dive →

Basal Ganglia

58/100

Deep cerebral nuclei (caudate, putamen, pallidum)

Developing — RL-based post-training is production-grade across every frontier lab. The procedural/agentic loop — pick the right action, repeat, build a skill — still fails on long horizons and unfamiliar environments, but the trend is steep.

Deep dive →