All filters off — toggle a chip or lower the importance slider to see nodes.
Top hubs · by degree
Legend
concept
claim
result
method
entity
MAP
Interactive version —
how to use this graph
✓
fast mental map
Click ▶ Guided tour for a 60-second walk through the editor's pick. Or hover any node to focus; click for source; ★ nodes you want to come back to; ⌘+click two nodes to compare.
✓
share a specific view
Select any node, copy URL — the link encodes selection, zoom, and filters. Save it as a named view (⌘ views). Annotations save locally per paper. </> embed generates an iframe.
✗
not a citable source
Do not quote the graph as an authority. Edge labels and importance scores are interpretive judgments by the generating agent. Any claim worth citing must be traced back to the original paper.
reliability noteHeadline structure and importance-5 nodes are stable across runs. Mid-tier nodes (importance 2–3) and edge type distinctions are interpretive and may differ between runs. Click any node to see its source citation — nodes marked "training memory" or "inferred" were not directly verified against the source document.
LOOMUS™ and the Knowledge-Loom methodology are proprietary. Visual system is original to LOOMUS.
Knowledge Graph: The Myth of Artificial Intelligence (Erik Larson, 2021)
Editorial spotlight: ↑ the abductive gap — what narrow AI cannot cross
Concepts
Peirce's abduction (importance 5): Inference to the best explanation. The creative leap that generates novel hypotheses to explain observations. Distinct from deduction and induction.. Source: (from training memory of book).
Larson's AGI myth (importance 5): The unfounded belief that scaling narrow AI techniques will eventually produce artificial general intelligence. Central thesis: this is false.. Source: (from training memory of book).
Larson's narrow AI (importance 4): AI systems that solve well-defined problems through pattern matching and statistical inference. All current AI is narrow AI.. Source: (from training memory of book).
statistical AI revolution (importance 4): The shift from symbolic to statistical methods (machine learning, deep learning). Enabled by big data and compute. Still narrow AI.. Source: (from training memory of book).
Larson's common sense barrier (importance 4): Human ability to flexibly apply background knowledge to novel situations. Requires abduction. No AI system has it.. Source: (from training memory of book).
Larson's hypothesis space (importance 4): The space of possible explanations for observations. Humans navigate this through abduction. AI systems cannot generate novel hypotheses.. Source: (from training memory of book).
Larson's explanation capacity (importance 4): Intelligence involves generating explanations for observations. This is abduction. AI systems cannot explain—they classify.. Source: (from training memory of book).
Peirce's three inferences (importance 4): Deduction (logical consequence), induction (pattern from instances), abduction (hypothesis to explain). Only humans do abduction.. Source: (from training memory of book).
Larson's hypothesis generation (importance 4): Creating novel explanatory hypotheses for observations. The essence of abduction. Uniquely human so far.. Source: (from training memory of book).
symbolic AI (GOFAI) (importance 3): Good Old-Fashioned AI. Rule-based systems that manipulate symbols. Dominated 1950s-1980s. Failed to scale.. Source: (from training memory of book).
McCarthy-Hayes frame problem (importance 3): The difficulty of representing which aspects of a situation remain unchanged by an action. A classic AGI barrier.. Source: (from training memory of book).
Moravec's paradox (importance 3): Hard problems for humans (chess, math) are easy for computers. Easy problems for humans (perception, manipulation) are hard for computers.. Source: (from training memory of book).
deep learning limits (importance 3): Neural networks with many layers. Effective for pattern recognition but still performing inductive inference, not abduction.. Source: (from training memory of book).
transfer learning failure (importance 3): Limited ability of AI systems to apply learned patterns to new domains. Humans excel at this through abductive reasoning.. Source: (from training memory of book).
Larson's context problem (importance 3): Human intelligence is radically context-sensitive. AI systems struggle with context because they lack abductive hypothesis generation.. Source: (from training memory of book).
Larson's engineering-science gap (importance 3): AI is engineering (building things that work) not science (understanding intelligence). This is a fundamental problem.. Source: (from training memory of book).
Larson's creativity gap (importance 3): True creativity involves generating novel hypotheses, not recombining patterns. AI lacks this because it lacks abduction.. Source: (from training memory of book).
natural language barrier (importance 3): True language understanding requires modeling speaker intentions, context, and generating explanatory hypotheses. Statistical models don't do this.. Source: (from training memory of book).
Brentano's intentionality (importance 3): Mental states are 'about' things. Consciousness has directedness. AI systems lack intentionality—they process without aboutness.. Source: (from training memory of book).
Larson's representation problem (importance 3): AI 'representations' are statistical patterns, not semantic symbols. They don't represent in the way human concepts do.. Source: (from training memory of book).
behavior vs. intelligence distinction (importance 3): Intelligent behavior (narrow task performance) ≠ intelligence (general reasoning). AI achieves the former, not the latter.. Source: (from training memory of book).
Larson's novelty test (importance 3): True intelligence handles genuinely novel situations. AI fails when confronted with distribution shift or unexpected scenarios.. Source: (from training memory of book).
Larson's data-dependence limit (importance 3): ML systems require massive labeled datasets. Humans learn from few examples through abductive reasoning.. Source: (from training memory of book).
Pearl's causal reasoning (importance 3): Understanding cause-effect relationships, not just correlations. Requires abduction. AI systems lack this.. Source: (from training memory of book).
Larson's research redirection (importance 3): Should focus on understanding intelligence scientifically, not engineering AGI systems we don't understand.. Source: (from training memory of book).
Larson's problem-formulation gap (importance 3): Humans formulate problems before solving them. AI solves pre-formulated problems. Formulation requires abduction.. Source: (from training memory of book).
Larson's meaning problem (importance 3): Statistical patterns lack intrinsic meaning. Human concepts have meaning through grounding in experience and intention.. Source: (from training memory of book).
Larson's science-first approach (importance 3): Progress toward understanding intelligence requires scientific theory development, not engineering optimization.. Source: (from training memory of book).
Larson's plausibility filtering (importance 3): Humans rapidly filter implausible hypotheses. Based on background knowledge and explanatory coherence. AI lacks this.. Source: (from training memory of book).
Turing Test inadequacy (importance 2): Behavioral equivalence doesn't imply intelligence. A system can pass the test through narrow pattern matching.. Source: (from training memory of book).
Searle's Chinese Room (importance 2): Thought experiment showing syntax manipulation doesn't produce semantic understanding. Larson uses this to critique statistical AI.. Source: (from training memory of book).
AI winters (importance 2): Periods of reduced funding and interest after failed promises. Larson predicts another winter when AGI doesn't materialize.. Source: (from training memory of book).
Kurzweil's Singularity (importance 2): Predicted moment when AI surpasses human intelligence and accelerates beyond control. Larson argues this rests on false premises.. Source: (from training memory of book).
Larson's actual AI risks (importance 2): Real dangers: bias, privacy violation, job displacement, autonomous weapons. These stem from narrow AI we have today.. Source: (from training memory of book).
Harnad's symbol grounding (importance 2): How symbols connect to their meanings. AI symbols aren't grounded—they're free-floating statistical patterns.. Source: (from training memory of book).
Larson's embodiment requirement (importance 2): Human intelligence is embodied—shaped by physical interaction with the world. AI lacks this grounding.. Source: (from training memory of book).
adversarial fragility (importance 2): Tiny input perturbations cause catastrophic misclassifications. Shows AI lacks robust understanding.. Source: (from training memory of book).
inductive bias limits (importance 2): ML systems embed assumptions about problem structure. These biases enable learning but prevent general reasoning.. Source: (from training memory of book).
meta-learning limits (importance 2): Learning to learn. Larson argues this still operates within pre-defined task distributions, not open-ended reasoning.. Source: (from training memory of book).
Simon's bounded rationality (importance 2): Humans use heuristics and satisficing, not optimal reasoning. Even this limited rationality exceeds AI through abduction.. Source: (from training memory of book).
Larson's insight problem (importance 2): Sudden understanding or 'aha moments' involve abductive leaps. Cannot be engineered through incremental search.. Source: (from training memory of book).
Hofstadter's analogy-making (importance 2): Core of cognition according to Hofstadter. Requires perceiving deep structural similarities. AI does surface similarity.. Source: (from training memory of book).
consciousness question (importance 2): Whether AGI requires consciousness is debated. Larson focuses on the prior problem: systems can't even reason generally.. Source: (from training memory of book).
Larson's ethics focus (importance 2): Real AI ethics concerns are about narrow AI we have now: bias, surveillance, job loss. Not hypothetical AGI alignment.. Source: (from training memory of book).
narrow AI automation (importance 2): Narrow AI will continue automating specific tasks. This has real economic and social impacts to address.. Source: (from training memory of book).
Thagard's explanatory coherence (importance 2): Good explanations cohere with other beliefs, are simple, explain more. Humans judge this; AI cannot.. Source: (from training memory of book).
Gödelian limits (importance 1): Some have argued Gödel's incompleteness theorems show limits of AI. Larson sees this as separate from the abduction problem.. Source: (from training memory of book).
philosophical zombie AI (importance 1): Even if we could build conscious AI, it wouldn't solve the abduction problem. These are separate issues.. Source: (from training memory of book).
Claims
Larson's abductive gap (importance 5): The fundamental barrier between narrow AI (pattern matching) and general intelligence (hypothesis generation). No known path from induction to abduction.. Source: (from training memory of book).
Larson's no-scaling thesis (importance 5): More data, more compute, more parameters will not bridge the abductive gap. AGI requires a fundamentally different approach.. Source: (from training memory of book).
Larson's abduction absence (importance 5): No AI system performs abductive reasoning. This is the core barrier to AGI. No known path forward.. Source: (from training memory of book).
Larson's induction-only thesis (importance 4): All current machine learning is inductive inference from data. No system performs abductive reasoning.. Source: (from training memory of book).
Larson's understanding gap (importance 4): Statistical correlation is not semantic understanding. Deep learning systems process patterns without comprehending meaning.. Source: (from training memory of book).
Larson's theory absence (importance 4): We have no scientific theory of general intelligence. AGI research proceeds without theoretical foundation.. Source: (from training memory of book).
Larson's open-endedness requirement (importance 4): General intelligence requires open-ended reasoning about unbounded hypothesis spaces. Narrow AI operates in closed, pre-defined spaces.. Source: (from training memory of book).
Larson's no-shortcuts thesis (importance 4): There's no shortcut to AGI. Can't engineer it without understanding it. Understanding requires a scientific theory we don't have.. Source: (from training memory of book).
Larson's Dreyfus vindication (importance 3): Dreyfus was right about symbolic AI's limits. Statistical AI hasn't escaped the same fundamental barriers.. Source: (from training memory of book).
Larson's ImageNet critique (importance 3): ImageNet success demonstrates narrow pattern matching, not general visual understanding. Systems fail on out-of-distribution images.. Source: (from training memory of book).
Larson's hype cycle (importance 3): AI has experienced repeated cycles of optimism and disappointment. Current deep learning hype follows the pattern.. Source: (from training memory of book).
Larson's Singularity rejection (importance 3): The Singularity assumes AGI is achievable through scaling narrow AI. Since this is false, the Singularity won't occur.. Source: (from training memory of book).
Larson's risk-concern critique (importance 3): Worrying about superintelligent AI risk is premature when we have no path to AGI. Distracts from real near-term AI ethics issues.. Source: (from training memory of book).
Larson's GPT critique (importance 3): GPT models create illusion of understanding through statistical fluency. They don't comprehend meaning or generate explanations.. Source: (from training memory of book).
Larson's intentionality absence (importance 3): AI systems lack intentional states. Their 'representations' aren't about anything. This is a fundamental barrier to AGI.. Source: (from training memory of book).
Larson's task-performance limit (importance 3): Mastering specific tasks doesn't constitute general intelligence. Need ability to flexibly tackle novel problems.. Source: (from training memory of book).
Larson's causation requirement (importance 3): General intelligence requires causal reasoning. Deep learning finds correlations but doesn't understand causes.. Source: (from training memory of book).
Larson's well-defined limit (importance 3): All AI successes are on well-defined problems. Real intelligence tackles ill-defined, open-ended challenges.. Source: (from training memory of book).
Larson's syntax-semantics gap (importance 3): No amount of syntactic manipulation produces semantic understanding. This is the Chinese Room argument applied to deep learning.. Source: (from training memory of book).
Larson's theory-first thesis (importance 3): Cannot engineer general intelligence without a theory of what it is. Current approach is backwards.. Source: (from training memory of book).
Larson's induction achievement (importance 3): Machine learning has mastered inductive inference at scale. But this is only one mode of reasoning.. Source: (from training memory of book).
Larson's search critique (importance 3): Cannot search over infinite hypothesis space. Need principled way to generate plausible hypotheses. This is the hard problem.. Source: (from training memory of book).
Larson's background knowledge gap (importance 3): Abduction requires vast background knowledge flexibly applied. AI has databases, not understanding.. Source: (from training memory of book).
Larson's paradox persistence (importance 2): Deep learning has made progress on perception but hasn't solved the underlying problem: systems still lack general intelligence.. Source: (from training memory of book).
Larson's AlphaGo limit (importance 2): AlphaGo mastered Go but cannot transfer that intelligence to any other domain. Pure narrow AI.. Source: (from training memory of book).
Larson's prediction critique (importance 2): AGI timeline predictions have been consistently wrong for 60+ years. No reason to trust current predictions.. Source: (from training memory of book).
Larson's few-shot critique (importance 2): GPT-3 'few-shot learning' is still pattern matching from massive pre-training. Not comparable to human few-shot generalization.. Source: (from training memory of book).
Larson's interpretability gap (importance 2): Deep learning systems are black boxes. Even their creators can't explain their decisions. This is a feature, not a bug—they don't reason.. Source: (from training memory of book).
Larson's brittleness problem (importance 2): AI systems are brittle—they fail unpredictably on edge cases. Human intelligence is robust through flexible reasoning.. Source: (from training memory of book).
No Free Lunch theorem (importance 2): No learning algorithm works for all problems. Specialization for narrow tasks prevents generalization.. Source: (from training memory of book).
Larson's recursive improvement critique (importance 2): Idea that AI could recursively self-improve to superintelligence assumes it can cross the abductive gap. It cannot.. Source: (from training memory of book).
Larson's non-mysterian stance (importance 2): Intelligence isn't mysterious or non-physical. But it's not achievable through current engineering approaches.. Source: (from training memory of book).
Larson's cognitive science priority (importance 2): Need better theories of human reasoning, especially abduction, before attempting AGI engineering.. Source: (from training memory of book).
Larson's heuristic-abduction link (importance 2): Human heuristics work because they're guided by abductive reasoning about problem structure. AI heuristics are blind.. Source: (from training memory of book).
Larson's analogy-abduction link (importance 2): Good analogies involve hypothesizing shared deep structure. This is abductive reasoning.. Source: (from training memory of book).
Larson's alignment critique (importance 2): Alignment problem assumes AGI is imminent. Since it's not, this research is premature compared to near-term ethics.. Source: (from training memory of book).
Larson's present-focus imperative (importance 2): Should address actual AI impacts today rather than speculating about future AGI scenarios.. Source: (from training memory of book).
Larson's deduction achievement (importance 2): Computers excel at deductive reasoning. Theorem provers, symbolic math systems demonstrate this.. Source: (from training memory of book).
Larson's consciousness separation (importance 1): Don't need to solve consciousness to see AGI is blocked. The abductive gap is a prior, more tractable problem.. Source: (from training memory of book).
Entities
Charles Sanders Peirce (importance 4): 19th century philosopher who formalized the three modes of inference: deduction, induction, and abduction.. Source: (from training memory of book).
Hubert Dreyfus (importance 3): Philosopher who critiqued symbolic AI in 'What Computers Can't Do' (1972). Larson updates his arguments for the ML era.. Source: (from training memory of book).
GPT models (importance 3): Large language models from OpenAI. Impressive pattern matching but no semantic understanding or abductive reasoning.. Source: (from training memory of book).
John McCarthy (importance 2): AI pioneer who coined 'artificial intelligence' and identified the frame problem.. Source: (from training memory of book).
Alan Turing (importance 2): Proposed the Turing Test (1950). Larson argues this test is insufficient for measuring general intelligence.. Source: (from training memory of book).
John Searle (importance 2): Philosopher who proposed the Chinese Room argument against strong AI (1980).. Source: (from training memory of book).
ImageNet breakthrough (importance 2): 2012 deep learning victory in image classification. Larson argues this doesn't generalize beyond narrow domains.. Source: (from training memory of book).
AlphaGo (importance 2): DeepMind system that defeated Go world champion (2016). Narrow AI success, not AGI breakthrough.. Source: (from training memory of book).
1956 Dartmouth Conference (importance 2): Birth of AI as a field. McCarthy, Minsky, et al. Original optimism about AGI timeline.. Source: (from training memory of book).
Marvin Minsky (importance 2): AI pioneer who co-founded MIT AI Lab. Made optimistic predictions that didn't pan out.. Source: (from training memory of book).
Ray Kurzweil (importance 2): Futurist who predicts AGI by 2029 and the Singularity by 2045. Larson views this as unfounded optimism.. Source: (from training memory of book).
Nick Bostrom (importance 2): Philosopher who warned of existential risk from superintelligent AI in 'Superintelligence' (2014).. Source: (from training memory of book).
Judea Pearl (importance 2): Computer scientist who developed causal inference theory. Critiques current ML for ignoring causation.. Source: (from training memory of book).
Douglas Hofstadter (importance 2): Cognitive scientist who emphasized analogy-making in 'Gödel, Escher, Bach' and critiques of modern AI.. Source: (from training memory of book).
Franz Brentano (importance 1): 19th century philosopher who identified intentionality as the mark of the mental.. Source: (from training memory of book).
Stevan Harnad (importance 1): Cognitive scientist who formulated the symbol grounding problem (1990).. Source: (from training memory of book).
Roger Penrose (importance 1): Physicist who argued consciousness involves quantum effects and exceeds computation. Larson doesn't rely on this argument.. Source: (from training memory of book).
Herbert Simon (importance 1): Cognitive scientist who studied human problem-solving and decision-making. Bounded rationality concept.. Source: (from training memory of book).
Paul Thagard (importance 1): Philosopher who developed computational models of explanatory coherence and abductive reasoning.. Source: (from training memory of book).