A Brief History of Intelligence

Knowledge Graph: A Brief History of Intelligence (Max Bennett, 2023)

Editorial spotlight: ↑ the fifth breakthrough: language transforms thought itself

Concepts

• Bennett's Breakthrough #1: Steering (importance 5): First neural innovation ~600 MYA in early bilaterians. Ability to navigate toward food/away from danger using simple steering circuits.. Source: (from training memory of book).
• Bennett's Breakthrough #2: Reinforcement Learning (importance 5): ~500 MYA vertebrates evolve basal ganglia. Trial-and-error learning, dopamine-based value updates, model-free RL.. Source: (from training memory of book).
• Bennett's Breakthrough #3: Simulating (importance 5): ~300 MYA early mammals develop neocortex. Mental models of the world, counterfactual reasoning, model-based planning.. Source: (from training memory of book).
• Bennett's Breakthrough #4: Mentalizing (importance 5): ~40 MYA primates evolve theory of mind. Modeling other agents' beliefs, intentions, knowledge states. Prefrontal cortex expansion.. Source: (from training memory of book).
• Bennett's Breakthrough #5: Language (importance 5): ~100 KYA humans acquire language. Symbolic representation, compositional thought, cultural transmission. Transforms cognition itself.. Source: (from training memory of book).
• Bennett's dopamine prediction error (importance 4): Neural mechanism encoding reward prediction errors. Core of vertebrate reinforcement learning. Discovered by Wolfram Schultz.. Source: (from training memory of book).
• Bennett's theory of mind (ToM) (importance 4): Ability to attribute mental states to others. False belief understanding. Distinguishes primates from other mammals.. Source: (from training memory of book).
• Bennett's language compositionality (importance 4): Infinite generativity from finite elements. Core property of human language. Enables abstract, recursive thought.. Source: (from training memory of book).
• Bennett's symbolic representation (importance 4): Using arbitrary signs to stand for concepts. Fundamental to language. Transforms nature of thought itself.. Source: (from training memory of book).
• Bennett's world model (neocortical construct) (importance 4): Internal representation of environment structure. Predicts sensory consequences of actions. Basis of mental simulation.. Source: (from training memory of book).
• Bennett's evolutionary accumulation principle (importance 4): Each breakthrough builds on previous ones. Intelligence is cumulative, not a single leap. Tinkering, not design.. Source: (from training memory of book).
• Bennett's working memory (PFC function) (importance 3): Active maintenance of information in prefrontal cortex. Enables planning, reasoning, mental manipulation.. Source: (from training memory of book).
• Bennett's credit assignment problem (importance 3): Challenge of determining which past actions caused current rewards. Solved by eligibility traces in basal ganglia.. Source: (from training memory of book).
• Bennett's social intelligence (primate specialization) (importance 3): Cognitive abilities for navigating complex social worlds. Theory of mind, deception detection, reputation tracking.. Source: (from training memory of book).
• Bennett's cultural ratchet (importance 3): Language enables cumulative cultural evolution. Knowledge accumulates across generations. Unique to humans.. Source: (from training memory of book).
• Bennett's value function (RL core) (importance 3): Mapping from states to expected future reward. Learned via temporal difference updates. Cached in basal ganglia.. Source: (from training memory of book).
• Bennett's episodic memory (hippocampal) (importance 3): Memory for specific events in space and time. Enables mental time travel. Hippocampus-dependent.. Source: (from training memory of book).
• Bennett's recursion (language property) (importance 3): Self-embedding of linguistic structures. Enables infinite expression from finite rules. Debated language universal.. Source: (from training memory of book).
• Bennett's inner speech (importance 3): Internal verbalization of thought. Language transforms cognition by providing 'software' layer over neural 'hardware'.. Source: (from training memory of book).
• Bennett's abstraction hierarchy (cortical) (importance 3): Processing progresses from concrete sensory to abstract conceptual across cortical hierarchy. Ventral visual stream example.. Source: (from training memory of book).
• Bennett's mental time travel (importance 3): Episodic memory + prospection. Mentally simulating past and future. Unique to humans (and possibly some animals).. Source: (from training memory of book).
• Bennett's cognitive map (hippocampal) (importance 3): Spatial representation of environment. O'Keefe's place cells. Foundation for mental simulation.. Source: (from training memory of book).
• Bennett's cortical hierarchy (importance 3): Sensory → association → executive areas. Increasing abstraction and timescales. Predictive coding framework.. Source: (from training memory of book).
• Bennett's executive function (PFC role) (importance 3): Top-down control of thought and action. Planning, inhibition, cognitive flexibility. Prefrontal specialization.. Source: (from training memory of book).
• Bennett's distributional semantics (LLM basis) (importance 3): Meaning from word co-occurrence patterns. 'You shall know a word by the company it keeps.' Foundation of language models.. Source: (from training memory of book).
• Bennett's emergent abilities (LLM phenomenon) (importance 3): Capabilities appearing suddenly at scale. Few-shot learning, chain-of-thought. Unpredicted from smaller models.. Source: (from training memory of book).
• Bennett's symbol grounding problem (importance 3): How symbols connect to real-world meaning. LLMs lack sensorimotor grounding. Chinese Room argument.. Source: (from training memory of book).
• Bennett's AGI (artificial general intelligence) (importance 3): AI matching human cognitive breadth. Would require all five breakthroughs. Current systems far from this.. Source: (from training memory of book).
• Bennett's exploration-exploitation tradeoff (importance 2): Balance between trying new actions and exploiting known rewards. Fundamental RL problem. Modulated by dopamine/norepinephrine.. Source: (from training memory of book).
• Bennett's habit formation (importance 2): Transition from goal-directed to habitual control with overtraining. Model-based to model-free. Basal ganglia striatal learning.. Source: (from training memory of book).
• Bennett's semantic memory (factual knowledge) (importance 2): General knowledge abstracted from episodes. Facts without context. Neocortex-based.. Source: (from training memory of book).
• Bennett's intentionality levels (Dennett's hierarchy) (importance 2): Orders of mental state attribution ('I think that you think that...'). Humans reach 5-6 levels. Other primates ~2.. Source: (from training memory of book).
• Bennett's goal-directed behavior (importance 2): Actions selected to achieve desired outcomes. Model-based control. Contrasts with habitual responding.. Source: (from training memory of book).
• Bennett's tool use (hierarchical motor planning) (importance 2): Using objects to achieve goals. Requires world modeling and prospection. Present in some mammals/birds.. Source: (from training memory of book).
• Bennett's latent learning (cognitive map) (importance 2): Learning without immediate reinforcement. Tolman's rats. Evidence for cognitive maps beyond RL.. Source: (from training memory of book).
• Bennett's eligibility trace (credit assignment) (importance 2): Temporary marker of recent neural activity. Allows delayed rewards to reinforce earlier actions.. Source: (from training memory of book).
• Bennett's memory consolidation (importance 2): Transfer from hippocampus to neocortex over time. Systems consolidation during sleep. Creates semantic memories.. Source: (from training memory of book).
• Bennett's attention schema (consciousness theory) (importance 2): Brain's model of its own attention. Graziano's proposal. May explain aspects of consciousness.. Source: (from training memory of book).
• Bennett's chunking (motor + cognitive) (importance 2): Grouping elements into units. Overcomes working memory limits. Basis of expertise and skill learning.. Source: (from training memory of book).
• Bennett's language instinct (Pinker term) (importance 2): Universal grammar hypothesis. Language as biological adaptation. Contrasts with cultural learning view.. Source: (from training memory of book).
• Bennett's catastrophic forgetting (neural nets) (importance 2): New learning overwrites old in connectionist networks. Solved in brain via complementary learning systems.. Source: (from training memory of book).
• Bennett's rapid variable binding (hippocampal) (importance 2): Quickly linking arbitrary elements into episodic memories. Hippocampus pattern separation enables this.. Source: (from training memory of book).
• Bennett's embodied cognition critique (importance 2): Claim that thought requires body/environment interaction. Bennett argues symbolic thought transcends this.. Source: (from training memory of book).
• Bennett's Moravec's paradox (importance 2): High-level reasoning is easier to automate than sensorimotor skills. Relates to evolutionary recency.. Source: (from training memory of book).
• Bennett's scaling laws (LLM capabilities) (importance 2): Performance improves predictably with model size and data. Kaplan 2020. Drives AI investment.. Source: (from training memory of book).
• Bennett's alignment problem (AI safety) (importance 2): Ensuring AI systems pursue intended goals. Orthogonality thesis. RLHF as partial solution.. Source: (from training memory of book).
• Bennett's multimodal learning (future AI) (importance 2): Combining vision, language, action. Potential path to grounded AI. CLIP, DALL-E examples.. Source: (from training memory of book).
• Bennett's narrow AI (current state) (importance 2): AI excelling at specific tasks but lacking general intelligence. All current systems including LLMs.. Source: (from training memory of book).
• Bennett's convergent evolution (intelligence) (importance 2): Similar problems produce similar solutions independently. Eyes, flight. But complex intelligence rare.. Source: (from training memory of book).
• Bennett's brain-as-computer metaphor (limits) (importance 2): Useful analogy but misleading. Brain evolved, not designed. No clean separation of hardware/software.. Source: (from training memory of book).

Claims

• Bennett's LLM thesis: GPT has Breakthrough 5 only (importance 5): Language models like GPT demonstrate language-level intelligence but lack steering, RL, simulation, and mentalizing. Bennett's central AI claim.. Source: (from training memory of book).
• Bennett's five-breakthrough completeness thesis (importance 5): These five innovations are necessary and sufficient for human-level intelligence. Core organizing claim of the book.. Source: (from training memory of book).
• Bennett's world models gap (LLMs lack Breakthrough 3) (importance 4): LLMs don't build causal models of the world. They pattern-match but don't simulate. Missing neocortex analogue.. Source: (from training memory of book).
• Bennett's RL gap (LLMs lack Breakthrough 2) (importance 4): LLMs don't learn from trial-and-error interaction with environment. No basal ganglia analogue. Static after training.. Source: (from training memory of book).
• Bennett's embodiment gap (LLMs lack Breakthrough 1) (importance 4): LLMs don't navigate or have sensorimotor steering. No body in environment. Missing foundational layer.. Source: (from training memory of book).
• Bennett's mentalizing gap (LLMs lack Breakthrough 4) (importance 4): LLMs don't model other minds or have genuine theory of mind. They simulate conversation, not mental states.. Source: (from training memory of book).
• Bennett's neuroscience → AI bridge thesis (importance 4): Understanding biological intelligence informs AI architectures. Book's method: evolution → mechanisms → algorithms → AI.. Source: (from training memory of book).
• Bennett's cooperative eye hypothesis (importance 1): Human sclera visibility evolved for social communication. Tracking others' gaze direction.. Source: (from training memory of book).
• Bennett's gestural origin hypothesis (language) (importance 1): Language may have evolved from manual gestures before vocalization. Mirror neuron connection.. Source: (from training memory of book).

Methods

• Bennett's model-free RL (Breakthrough 2 mechanism) (importance 3): Learning system that maps states to actions via cached values. Fast but inflexible. Implemented by basal ganglia.. Source: (from training memory of book).
• Bennett's model-based planning (Breakthrough 3 mechanism) (importance 3): Using internal world models to mentally simulate outcomes before acting. Slower but flexible. Neocortex function.. Source: (from training memory of book).
• Bennett's offline replay (neocortical mechanism) (importance 3): Spontaneous reactivation of neural patterns during rest/sleep. Consolidates memories, discovers abstract structure.. Source: (from training memory of book).
• Bennett's counterfactual reasoning (importance 3): Ability to mentally simulate alternative scenarios ('what if'). Key function of neocortex and hippocampus.. Source: (from training memory of book).
• Bennett's reward prediction (dopamine role) (importance 3): Dopamine neurons signal when outcomes differ from expectations. Drives learning in basal ganglia circuits.. Source: (from training memory of book).
• Bennett's predictive coding (importance 3): Hierarchical prediction of sensory input. Errors propagate up cortical hierarchy. Efficient encoding scheme.. Source: (from training memory of book).
• Bennett's temporal difference learning (TD) (importance 3): RL algorithm updating predictions based on errors. Sutton & Barto. Maps to dopamine signaling.. Source: (from training memory of book).
• Bennett's complementary learning systems (importance 3): Hippocampus (fast) + neocortex (slow) prevents catastrophic forgetting. McClelland & Goddard model.. Source: (from training memory of book).
• Bennett's steering circuits (worm-level) (importance 2): Hardwired sensorimotor reflexes. Approach/avoidance without learning. Implemented via simple neural networks.. Source: (from training memory of book).
• Bennett's sensorimotor prediction (importance 2): Predicting sensory input from motor commands. Basis of world modeling. Implemented via cortical forward models.. Source: (from training memory of book).
• Bennett's attention mechanism (importance 2): Selection of relevant information for processing. Frontal-parietal networks. Modulates cortical gain.. Source: (from training memory of book).
• Bennett's false belief task (ToM test) (importance 2): Experimental paradigm testing theory of mind. Passed by 4-year-old humans and great apes.. Source: (from training memory of book).
• Bennett's imitation learning (importance 2): Learning by observing others. Critical for cultural transmission. Enhanced by mirror neurons and mentalizing.. Source: (from training memory of book).
• Bennett's Pavlovian conditioning (pre-RL) (importance 2): Associative learning via stimulus pairing. Simpler than instrumental RL. Present even in invertebrates.. Source: (from training memory of book).
• Bennett's instrumental conditioning (RL proper) (importance 2): Learning via action-outcome contingencies. Requires basal ganglia. Emergence in vertebrates.. Source: (from training memory of book).
• Bennett's forward model (motor control) (importance 2): Predicting sensory consequences of motor commands. Cerebellum and cortex. Basis of internal simulation.. Source: (from training memory of book).
• Bennett's actor-critic architecture (importance 2): RL system with separate policy (actor) and value (critic) learners. Maps to basal ganglia structure.. Source: (from training memory of book).
• Bennett's pattern completion (hippocampal) (importance 2): Reconstructing full memory from partial cue. Attractor dynamics. Hippocampus CA3 recurrent connections.. Source: (from training memory of book).
• Bennett's pattern separation (hippocampal) (importance 2): Orthogonalizing similar inputs to reduce interference. Dentate gyrus function. Prevents memory confusion.. Source: (from training memory of book).
• Bennett's backpropagation algorithm (importance 2): Training method propagating errors through network layers. Biologically implausible but effective.. Source: (from training memory of book).
• Bennett's self-attention mechanism (importance 2): Relating different positions in a sequence to compute representation. Key innovation of transformers.. Source: (from training memory of book).
• Bennett's pretraining + finetuning paradigm (importance 2): Train on massive unlabeled data, then adapt to tasks. Transfer learning. Foundation of modern NLP.. Source: (from training memory of book).
• Bennett's RLHF (reinforcement learning from human feedback) (importance 2): Fine-tuning LLMs using human preference rankings. Makes models helpful, honest, harmless.. Source: (from training memory of book).
• Bennett's inverse model (motor control) (importance 1): Computing motor commands needed to achieve sensory goals. Motor cortex. Complements forward models.. Source: (from training memory of book).

Entities

• Bennett's basal ganglia (vertebrate innovation) (importance 4): Subcortical structure enabling action selection via dopamine-modulated reinforcement learning. First appeared in early vertebrates.. Source: (from training memory of book).
• Bennett's neocortex (mammalian innovation) (importance 4): Six-layered cortical sheet unique to mammals. Enables world-modeling, prediction, and mental simulation.. Source: (from training memory of book).
• Bennett's prefrontal cortex (primate expansion) (importance 4): Frontal lobe region massively expanded in primates. Critical for mentalizing, executive function, social reasoning.. Source: (from training memory of book).
• C. elegans (302-neuron model organism) (importance 3): Nematode worm with fully mapped connectome. Example of steering-only intelligence without learning.. Source: (from training memory of book).
• Bennett's hippocampus (spatial + episodic memory) (importance 3): Brain structure for spatial navigation and episodic memory encoding. Plays role in mental simulation.. Source: (from training memory of book).
• Bennett's deep learning (modern AI) (importance 3): Multi-layer neural networks trained via backpropagation. 2012+ revolution. Powers modern LLMs.. Source: (from training memory of book).
• Bennett's transformer (Attention Is All You Need) (importance 3): Neural architecture using self-attention. Vaswani 2017. Basis of GPT and modern LLMs.. Source: (from training memory of book).
• Bennett's vertebrate emergence (~500 MYA) (importance 2): Cambrian explosion leads to vertebrates with centralized brains, eyes, and basal ganglia.. Source: (from training memory of book).
• Bennett's mammalian emergence (~300 MYA) (importance 2): Synapsids diverge from reptiles. First neocortex appears. Enables endothermy, parental care, complex behavior.. Source: (from training memory of book).
• Bennett's primate emergence (~40 MYA) (importance 2): Early primates evolve from tree shrews. Social complexity drives prefrontal expansion and mentalizing.. Source: (from training memory of book).
• Bennett's human emergence (~100 KYA) (importance 2): Anatomically modern humans with full language capacity. Cognitive modernity, symbolic culture, rapid innovation.. Source: (from training memory of book).
• Bennett's mirror neurons (primate discovery) (importance 2): Neurons that fire both when acting and observing actions. Proposed role in imitation and theory of mind.. Source: (from training memory of book).
• Bennett's Wernicke's area (language comprehension) (importance 2): Posterior temporal cortex region for language understanding. Damage causes comprehension deficits.. Source: (from training memory of book).
• Bennett's Broca's area (language production) (importance 2): Frontal cortex region for speech production and syntax. Damage causes production deficits.. Source: (from training memory of book).
• Bennett's place cells (O'Keefe discovery) (importance 2): Hippocampal neurons firing at specific locations. Evidence for cognitive maps. Nobel Prize 2014.. Source: (from training memory of book).
• Bennett's grid cells (Mosers discovery) (importance 2): Entorhinal neurons with hexagonal firing patterns. Metric for spatial navigation. Nobel Prize 2014.. Source: (from training memory of book).
• Bennett's cortical columns (Mountcastle's discovery) (importance 2): Repeating vertical structure of neocortex. ~0.5mm wide. Proposed computational unit.. Source: (from training memory of book).
• Bennett's default mode network (DMN) (importance 2): Brain network active during rest and introspection. Mentalizing, prospection, self-referential thought.. Source: (from training memory of book).
• Bennett's symbolic AI (GOFAI) (importance 2): Classical AI approach using explicit rules and logic. 'Good Old-Fashioned AI'. Failed on common sense.. Source: (from training memory of book).
• Bennett's connectionism (neural net revival) (importance 2): 1980s AI paradigm shift to neural networks. Parallel distributed processing. Rumelhart & McClelland.. Source: (from training memory of book).
• Bennett's Chinese Room (Searle argument) (importance 2): Thought experiment arguing syntax ≠ semantics. Person following rules doesn't understand Chinese.. Source: (from training memory of book).
• Bennett's AI winters (historical periods) (importance 1): Periods of reduced AI funding and interest. Caused by overpromising early symbolic AI systems.. Source: (from training memory of book).
• Bennett's Turing Test (importance 1): Conversational imitation game as intelligence test. Turing 1950. LLMs increasingly pass.. Source: (from training memory of book).

Relations

• Bennett's Breakthrough #1: Steering exemplifies C. elegans (302-neuron model organism)
• Bennett's Breakthrough #1: Steering enables Bennett's steering circuits (worm-level)
• Bennett's Breakthrough #1: Steering precedes Bennett's Breakthrough #2: Reinforcement Learning
• Bennett's Breakthrough #2: Reinforcement Learning enables Bennett's basal ganglia (vertebrate innovation)
• Bennett's Breakthrough #2: Reinforcement Learning enables Bennett's dopamine prediction error
• Bennett's Breakthrough #2: Reinforcement Learning enables Bennett's model-free RL (Breakthrough 2 mechanism)
• Bennett's Breakthrough #2: Reinforcement Learning exemplifies Bennett's vertebrate emergence (~500 MYA)
• Bennett's Breakthrough #2: Reinforcement Learning precedes Bennett's Breakthrough #3: Simulating
• Bennett's Breakthrough #3: Simulating enables Bennett's neocortex (mammalian innovation)
• Bennett's Breakthrough #3: Simulating enables Bennett's model-based planning (Breakthrough 3 mechanism)
• Bennett's Breakthrough #3: Simulating enables Bennett's hippocampus (spatial + episodic memory)
• Bennett's Breakthrough #3: Simulating exemplifies Bennett's mammalian emergence (~300 MYA)
• Bennett's Breakthrough #3: Simulating precedes Bennett's Breakthrough #4: Mentalizing
• Bennett's Breakthrough #4: Mentalizing enables Bennett's prefrontal cortex (primate expansion)
• Bennett's Breakthrough #4: Mentalizing enables Bennett's theory of mind (ToM)
• Bennett's Breakthrough #4: Mentalizing exemplifies Bennett's primate emergence (~40 MYA)
• Bennett's Breakthrough #4: Mentalizing precedes Bennett's Breakthrough #5: Language
• Bennett's Breakthrough #5: Language enables Bennett's language compositionality
• Bennett's Breakthrough #5: Language enables Bennett's symbolic representation
• Bennett's Breakthrough #5: Language enables Bennett's cultural ratchet
• Bennett's Breakthrough #5: Language exemplifies Bennett's human emergence (~100 KYA)
• Bennett's basal ganglia (vertebrate innovation) enables Bennett's instrumental conditioning (RL proper)
• Bennett's basal ganglia (vertebrate innovation) enables Bennett's habit formation
• Bennett's basal ganglia (vertebrate innovation) enables Bennett's value function (RL core)
• Bennett's basal ganglia (vertebrate innovation) exemplifies Bennett's actor-critic architecture
• Bennett's dopamine prediction error enables Bennett's reward prediction (dopamine role)
• Bennett's dopamine prediction error exemplifies Bennett's temporal difference learning (TD)
• Bennett's dopamine prediction error supports Bennett's credit assignment problem
• Bennett's neocortex (mammalian innovation) enables Bennett's world model (neocortical construct)
• Bennett's neocortex (mammalian innovation) enables Bennett's offline replay (neocortical mechanism)
• Bennett's neocortex (mammalian innovation) exemplifies Bennett's cortical columns (Mountcastle's discovery)
• Bennett's neocortex (mammalian innovation) exemplifies Bennett's cortical hierarchy
• Bennett's neocortex (mammalian innovation) enables Bennett's predictive coding
• Bennett's hippocampus (spatial + episodic memory) enables Bennett's episodic memory (hippocampal)
• Bennett's hippocampus (spatial + episodic memory) enables Bennett's cognitive map (hippocampal)
• Bennett's hippocampus (spatial + episodic memory) exemplifies Bennett's place cells (O'Keefe discovery)
• Bennett's hippocampus (spatial + episodic memory) enables Bennett's pattern completion (hippocampal)
• Bennett's hippocampus (spatial + episodic memory) enables Bennett's pattern separation (hippocampal)
• Bennett's hippocampus (spatial + episodic memory) exemplifies Bennett's complementary learning systems
• Bennett's prefrontal cortex (primate expansion) enables Bennett's working memory (PFC function)
• Bennett's prefrontal cortex (primate expansion) enables Bennett's executive function (PFC role)
• Bennett's prefrontal cortex (primate expansion) exemplifies Bennett's default mode network (DMN)
• Bennett's theory of mind (ToM) exemplifies Bennett's false belief task (ToM test)
• Bennett's theory of mind (ToM) supports Bennett's mirror neurons (primate discovery)
• Bennett's theory of mind (ToM) enables Bennett's intentionality levels (Dennett's hierarchy)
• Bennett's theory of mind (ToM) enables Bennett's social intelligence (primate specialization)
• Bennett's model-free RL (Breakthrough 2 mechanism) builds-on Bennett's Pavlovian conditioning (pre-RL)
• Bennett's model-free RL (Breakthrough 2 mechanism) exemplifies Bennett's instrumental conditioning (RL proper)
• Bennett's model-free RL (Breakthrough 2 mechanism) requires Bennett's exploration-exploitation tradeoff
• Bennett's model-based planning (Breakthrough 3 mechanism) enables Bennett's counterfactual reasoning
• Bennett's model-based planning (Breakthrough 3 mechanism) enables Bennett's goal-directed behavior
• Bennett's model-based planning (Breakthrough 3 mechanism) evidences Bennett's latent learning (cognitive map)
• Bennett's world model (neocortical construct) enables Bennett's sensorimotor prediction
• Bennett's world model (neocortical construct) exemplifies Bennett's forward model (motor control)
• Bennett's cognitive map (hippocampal) evidences Bennett's latent learning (cognitive map)
• Bennett's cognitive map (hippocampal) enables Bennett's place cells (O'Keefe discovery)
• Bennett's cognitive map (hippocampal) enables Bennett's grid cells (Mosers discovery)
• Bennett's episodic memory (hippocampal) enables Bennett's mental time travel
• Bennett's episodic memory (hippocampal) requires Bennett's memory consolidation
• Bennett's episodic memory (hippocampal) enables Bennett's rapid variable binding (hippocampal)
• Bennett's semantic memory (factual knowledge) requires Bennett's memory consolidation
• Bennett's language compositionality exemplifies Bennett's recursion (language property)
• Bennett's symbolic representation enables Bennett's inner speech
• Bennett's symbolic representation requires Bennett's Wernicke's area (language comprehension)
• Bennett's symbolic representation requires Bennett's Broca's area (language production)
• Bennett's cultural ratchet requires Bennett's imitation learning
• Bennett's LLM thesis: GPT has Breakthrough 5 only supports Bennett's Breakthrough #5: Language
• Bennett's LLM thesis: GPT has Breakthrough 5 only supports Bennett's embodiment gap (LLMs lack Breakthrough 1)
• Bennett's LLM thesis: GPT has Breakthrough 5 only supports Bennett's RL gap (LLMs lack Breakthrough 2)
• Bennett's LLM thesis: GPT has Breakthrough 5 only supports Bennett's world models gap (LLMs lack Breakthrough 3)
• Bennett's LLM thesis: GPT has Breakthrough 5 only supports Bennett's mentalizing gap (LLMs lack Breakthrough 4)
• Bennett's embodiment gap (LLMs lack Breakthrough 1) requires Bennett's Breakthrough #1: Steering
• Bennett's RL gap (LLMs lack Breakthrough 2) requires Bennett's Breakthrough #2: Reinforcement Learning
• Bennett's world models gap (LLMs lack Breakthrough 3) requires Bennett's Breakthrough #3: Simulating
• Bennett's mentalizing gap (LLMs lack Breakthrough 4) requires Bennett's Breakthrough #4: Mentalizing
• Bennett's deep learning (modern AI) builds-on Bennett's connectionism (neural net revival)
• Bennett's deep learning (modern AI) requires Bennett's backpropagation algorithm
• Bennett's deep learning (modern AI) enables Bennett's transformer (Attention Is All You Need)
• Bennett's transformer (Attention Is All You Need) requires Bennett's self-attention mechanism
• Bennett's transformer (Attention Is All You Need) enables Bennett's pretraining + finetuning paradigm
• Bennett's pretraining + finetuning paradigm evidences Bennett's scaling laws (LLM capabilities)
• Bennett's scaling laws (LLM capabilities) evidences Bennett's emergent abilities (LLM phenomenon)
• Bennett's distributional semantics (LLM basis) enables Bennett's deep learning (modern AI)
• Bennett's symbol grounding problem exemplifies Bennett's Chinese Room (Searle argument)
• Bennett's symbol grounding problem supports Bennett's embodied cognition critique
• Bennett's symbol grounding problem supports Bennett's world models gap (LLMs lack Breakthrough 3)
• Bennett's complementary learning systems supports Bennett's catastrophic forgetting (neural nets)
• Bennett's RLHF (reinforcement learning from human feedback) supports Bennett's alignment problem (AI safety)
• Bennett's multimodal learning (future AI) supports Bennett's symbol grounding problem
• Bennett's five-breakthrough completeness thesis requires Bennett's Breakthrough #1: Steering
• Bennett's five-breakthrough completeness thesis requires Bennett's Breakthrough #2: Reinforcement Learning
• Bennett's five-breakthrough completeness thesis requires Bennett's Breakthrough #3: Simulating
• Bennett's five-breakthrough completeness thesis requires Bennett's Breakthrough #4: Mentalizing
• Bennett's five-breakthrough completeness thesis requires Bennett's Breakthrough #5: Language
• Bennett's five-breakthrough completeness thesis supports Bennett's evolutionary accumulation principle
• Bennett's evolutionary accumulation principle supports Bennett's convergent evolution (intelligence)
• Bennett's neuroscience → AI bridge thesis supports Bennett's LLM thesis: GPT has Breakthrough 5 only
• Bennett's neuroscience → AI bridge thesis supports Bennett's five-breakthrough completeness thesis
• Bennett's AGI (artificial general intelligence) requires Bennett's five-breakthrough completeness thesis
• Bennett's AGI (artificial general intelligence) contradicts Bennett's narrow AI (current state)
• Bennett's narrow AI (current state) exemplifies Bennett's deep learning (modern AI)
• Bennett's credit assignment problem requires Bennett's eligibility trace (credit assignment)
• Bennett's temporal difference learning (TD) exemplifies Bennett's dopamine prediction error
• Bennett's offline replay (neocortical mechanism) enables Bennett's memory consolidation
• Bennett's abstraction hierarchy (cortical) exemplifies Bennett's cortical hierarchy
• Bennett's tool use (hierarchical motor planning) requires Bennett's model-based planning (Breakthrough 3 mechanism)
• Bennett's social intelligence (primate specialization) requires Bennett's theory of mind (ToM)
• Bennett's social intelligence (primate specialization) requires Bennett's imitation learning
• Bennett's forward model (motor control) supports Bennett's inverse model (motor control)
• Bennett's attention mechanism supports Bennett's working memory (PFC function)
• Bennett's attention mechanism supports Bennett's self-attention mechanism
• Bennett's predictive coding exemplifies Bennett's sensorimotor prediction
• Bennett's inner speech supports Bennett's working memory (PFC function)
• Bennett's chunking (motor + cognitive) supports Bennett's working memory (PFC function)
• Bennett's mirror neurons (primate discovery) supports Bennett's imitation learning
• Bennett's mirror neurons (primate discovery) supports Bennett's gestural origin hypothesis (language)
• Bennett's language instinct (Pinker term) supports Bennett's symbolic representation
• Bennett's Moravec's paradox supports Bennett's Breakthrough #1: Steering
• Bennett's symbolic AI (GOFAI) motivates Bennett's AI winters (historical periods)
• Bennett's connectionism (neural net revival) contradicts Bennett's symbolic AI (GOFAI)
• Bennett's Turing Test contradicts Bennett's Chinese Room (Searle argument)
• Bennett's default mode network (DMN) enables Bennett's mental time travel
• Bennett's default mode network (DMN) enables Bennett's theory of mind (ToM)
• Bennett's habit formation contradicts Bennett's goal-directed behavior
• Bennett's value function (RL core) requires Bennett's temporal difference learning (TD)
• Bennett's actor-critic architecture requires Bennett's temporal difference learning (TD)
• Bennett's emergent abilities (LLM phenomenon) evidences Bennett's LLM thesis: GPT has Breakthrough 5 only

© 2026 LOOMUS. Curation, importance scoring, edge typing, and visualization. Source paper © respective authors, quoted under fair use.