arxiv: 2603.18203 · v4 · submitted 2026-03-18 · 💻 cs.CL · cs.CY

Recognition: no theorem link

How Psychological Learning Paradigms Shaped and Constrained Artificial Intelligence

Alex Anvi Eponon , Ildar Batyrshin , Christian E. Maldonado-Sifuentes , Grigori Sidorov

Authors on Pith no claims yet

Pith reviewed 2026-05-15 09:11 UTC · model grok-4.3

classification 💻 cs.CL cs.CY

keywords systematic compositional reasoningpsychological learning paradigmsAI architecturebehaviourism cognitivism constructivismReSynth frameworkauxiliary hypotheseschain-of-thought prompting

0 comments

The pith

AI's failure at systematic compositional reasoning is architectural, inherited from behaviorism, cognitivism, and constructivism rather than fixable by scale or prompting.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that current AI systems cannot reliably recombine known components into novel configurations because their core architectures were shaped by psychological learning theories that omitted necessary structural features. Behaviourism excluded internal structure, cognitivism left representations opaque, and constructivism provided no formal construction operators, so techniques such as chain-of-thought prompting and human-feedback alignment function only as auxiliary patches. The authors trace this genealogy explicitly and conclude that systematicity will remain an after-the-fact correction until architectures are redesigned. They introduce ReSynth as a trimodular framework that separates reasoning, identity, and memory so that systematic behaviour becomes a direct consequence of the design. A reader should care because the argument implies that simply enlarging models or adding data will not overcome the deficit if the underlying architectural indifference persists.

Core claim

The central claim is that the inability of AI to exhibit systematic compositional reasoning is not a matter of insufficient scale or data but an architectural consequence of the psychological learning paradigms that informed AI methodology. Behaviourism bequeathed the exclusion of internal structure, cognitivism the opacity of representations, and constructivism the absence of formal construction operators. As a result, proliferating corrective methods such as chain-of-thought prompting and alignment through human feedback operate as auxiliary hypotheses that address symptoms without altering the underlying indifference to systematicity. The paper proposes ReSynth, a trimodular conceptual框架,

What carries the argument

ReSynth, the trimodular conceptual framework that separates reasoning, identity, and memory so systematic behaviour becomes a structural consequence of the architecture rather than a post-hoc correction.

If this is right

Corrective techniques such as chain-of-thought prompting and reinforcement learning from human feedback will remain insufficient because they do not alter the architectural source of the deficit.
AI development must incorporate internal structure, transparent representations, and formal construction operators if systematicity is to arise by design.
A cross-cultural reappraisal of rote learning offers an underexploited route to supplying the missing formal operators.
Scaling models or increasing training data alone will not produce architectures in which systematic behaviour is a necessary consequence rather than an auxiliary result.
Future systems should treat the separation of reasoning, identity, and memory as a foundational design principle rather than an optional module.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

If the architectural diagnosis holds, current scaling trajectories in language models may reach a plateau on tasks requiring genuine novelty rather than pattern completion.
The argument suggests that alternative psychological or philosophical learning theories beyond the three dominant ones could supply additional structural ingredients currently missing from AI.
Implementing the ReSynth separation in an actual system would allow direct tests of whether the trimodular split is sufficient to produce systematic recombinations on held-out configurations.
The same logic could be applied to other cognitive capacities, such as causal reasoning or analogical mapping, that also depend on explicit structural recombination.

Load-bearing premise

The assumption that AI's deficit in systematic compositional reasoning is architectural in origin and traces directly to the structural limitations bequeathed by behaviourism, cognitivism, and constructivism.

What would settle it

An existing connectionist or transformer-based model, trained at current scales without ReSynth-style separation of reasoning, identity, and memory, that reliably produces correct outputs on novel recombinations of components never seen together during training would falsify the architectural-indifference claim.

read the original abstract

Current artificial intelligence systems struggle with systematic compositional reasoning: the capacity to recombine known components in novel configurations. This paper argues that the failure is architectural, not merely a matter of scale or training data, and that its origins lie in the psychological learning theories from which AI paradigms were derived. The argument proceeds in three stages. First, drawing on the systematicity debate in cognitive science and on the demonstration of Aizawa that neither connectionism nor classicism can make systematicity a structural consequence of the architecture, the paper establishes that the corrective techniques proliferating in modern AI, from chain-of-thought prompting to alignment through human feedback, function as auxiliary hypotheses that address symptoms without resolving the underlying architectural indifference to systematicity. Second, it traces the genealogy from psychological learning theory to AI methodology, showing that behaviourism, cognitivism, and constructivism each bequeathed a specific structural limitation to the AI paradigm it inspired: the exclusion of internal structure, the opacity of representation, and the absence of formal construction operators. A cross-cultural reappraisal of rote learning reveals a further underexploited pathway. Third, the paper introduces ReSynth, a trimodular conceptual framework that proposes the principled separation of reasoning, identity, and memory as a path toward architectures in which systematic behaviour is a structural consequence of design rather than a correction applied after the fact.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper traces AI's systematicity failures to inherited psychological learning theories and offers ReSynth as a conceptual fix, but the causal links stay interpretive without derivation from actual mechanisms.

read the letter

The paper's main point is that current AI systems fail at systematic compositional reasoning because of architectural constraints inherited from behaviorism, cognitivism, and constructivism, and that fixes like chain-of-thought or RLHF are just bandaids. It proposes ReSynth, a framework separating reasoning, identity, and memory, to make systematicity built-in. The work does a good job connecting the systematicity debate in cognitive science to AI development history. The section on how each psychological paradigm passed on a specific limitation is clear, and the cross-cultural take on rote learning brings in something new. It rightly notes that many modern techniques address symptoms rather than the root architectural indifference. The soft spots are in the evidence for the central claim. The genealogy is traced conceptually, but there's no step-by-step derivation showing how those theories constrain concrete elements like attention updates or embedding spaces. Other factors, such as statistical learning or hardware limits, aren't ruled out as primary causes. ReSynth itself is presented at a high level without formal specification or tests. This paper would appeal to researchers at the AI-cognitive science boundary who are open to historical and philosophical analysis. It raises worthwhile questions about design principles. I would send it to peer review. The conceptual synthesis is worth referee input, even if it needs more technical backing to hold up.

Referee Report

3 major / 2 minor

Summary. The paper claims that current AI systems' failure at systematic compositional reasoning is architectural rather than a matter of scale or data, with roots in the structural limitations bequeathed by behaviourism (exclusion of internal structure), cognitivism (representational opacity), and constructivism (absence of formal operators). It argues that techniques such as chain-of-thought prompting and RLHF function only as auxiliary hypotheses addressing symptoms, draws on the systematicity debate and Aizawa's demonstration to support this, and proposes the ReSynth trimodular framework (separating reasoning, identity, and memory) to make systematicity a structural consequence of design.

Significance. If the genealogical links and the architectural diagnosis hold, the work would reframe AI limitations as historically contingent rather than inevitable, offering a path to architectures in which systematic behaviour is designed in rather than corrected post hoc. The explicit engagement with the systematicity debate and the concrete proposal of ReSynth provide a falsifiable direction for future architectural work that could be tested against existing connectionist and symbolic baselines.

major comments (3)

[Genealogy tracing (second stage)] The second stage traces behaviourism, cognitivism, and constructivism to specific structural limitations in AI but does not derive these limitations from concrete mechanisms such as the attention update rule, the back-propagation gradient, or the geometry of the token embedding space; without this derivation the claim that the observed failures are architectural inheritances rather than statistical or hardware constraints remains unproven.
[First stage (systematicity debate and corrective techniques)] The assertion that chain-of-thought, RLHF, and related techniques are merely auxiliary hypotheses rests on the prior claim of architectural indifference; yet the manuscript provides no analysis showing why these methods cannot, in principle, induce the required structural properties through training dynamics or prompting, leaving the auxiliary-hypothesis diagnosis unsupported by examination of the model's inductive bias.
[Third stage (ReSynth proposal)] ReSynth is introduced as a trimodular conceptual framework whose separation of reasoning, identity, and memory is said to render systematicity structural, but the manuscript supplies neither a formal specification of the modules nor a proof (or even a sketch) that this separation satisfies Aizawa's criterion that systematicity be a structural consequence rather than an emergent or corrected property.

minor comments (2)

[Abstract] The abstract refers to 'Aizawa' without a full citation; the reference list should include the specific work invoked in the systematicity argument.
[Second stage] The cross-cultural reappraisal of rote learning is mentioned only briefly; if it is intended to supply an additional pathway, it should be expanded with at least one concrete example of how it differs from the three main paradigms.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the constructive and detailed comments on our manuscript. We address each major comment point by point below, noting the revisions we intend to make in response.

read point-by-point responses

Referee: [Genealogy tracing (second stage)] The second stage traces behaviourism, cognitivism, and constructivism to specific structural limitations in AI but does not derive these limitations from concrete mechanisms such as the attention update rule, the back-propagation gradient, or the geometry of the token embedding space; without this derivation the claim that the observed failures are architectural inheritances rather than statistical or hardware constraints remains unproven.

Authors: We concur that strengthening the link between the psychological paradigms and concrete AI mechanisms would make the architectural diagnosis more robust. The manuscript currently focuses on the historical and conceptual lineage, but we will revise the second stage to include explicit mappings. For example, we will derive how behaviourism's rejection of internal states leads to the non-modular attention update rules in modern transformers, and how cognitivism's opaque representations persist in the geometry of token embeddings despite gradient-based training. This addition will address the concern that the failures might be statistical rather than architectural. revision: yes
Referee: [First stage (systematicity debate and corrective techniques)] The assertion that chain-of-thought, RLHF, and related techniques are merely auxiliary hypotheses rests on the prior claim of architectural indifference; yet the manuscript provides no analysis showing why these methods cannot, in principle, induce the required structural properties through training dynamics or prompting, leaving the auxiliary-hypothesis diagnosis unsupported by examination of the model's inductive bias.

Authors: The auxiliary hypothesis characterization is based on the systematicity debate and Aizawa's demonstration that current architectures lack the structural basis for systematicity. To support this further, we will add to the first stage an examination of the inductive biases in transformer models. Specifically, we will analyze why chain-of-thought prompting and RLHF cannot fundamentally alter the architecture to make systematicity structural, citing evidence that these methods improve performance on specific tasks but do not resolve generalization failures in novel compositions. This will include discussion of how training dynamics reinforce existing representational limitations. revision: yes
Referee: [Third stage (ReSynth proposal)] ReSynth is introduced as a trimodular conceptual framework whose separation of reasoning, identity, and memory is said to render systematicity structural, but the manuscript supplies neither a formal specification of the modules nor a proof (or even a sketch) that this separation satisfies Aizawa's criterion that systematicity be a structural consequence rather than an emergent or corrected property.

Authors: As a conceptual framework, ReSynth outlines a direction for future work rather than providing an implemented system. We will revise the third stage to include a sketch of the module specifications, describing the interfaces between reasoning, identity, and memory modules and how they enforce structural systematicity. We will also provide a brief argument demonstrating alignment with Aizawa's criterion by showing that systematic recombination becomes a direct outcome of the modular separation. However, a full formal proof would require detailed implementation and testing, which we note as future work. revision: partial

standing simulated objections not resolved

A full formal proof and implementation of ReSynth to rigorously satisfy Aizawa's criterion.

Circularity Check

0 steps flagged

No significant circularity detected in derivation chain

full rationale

The paper's argument proceeds via external citation to Aizawa's demonstration on systematicity (neither connectionism nor classicism makes it structural) and a historical/genealogical tracing of behaviourism, cognitivism and constructivism to specific architectural limitations. No equations, fitted parameters, or self-citations appear in the provided text that reduce any central claim to its own inputs by construction. The establishment that corrective techniques function as auxiliary hypotheses follows directly from the cited external demonstration rather than from a self-referential loop. Introduction of the ReSynth framework is presented as a forward proposal, not a derived prediction forced by prior steps. The derivation remains interpretive and self-contained against external benchmarks, with no load-bearing self-citation chains or renamings of known results.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The paper rests on the established systematicity debate in cognitive science and historical interpretations of learning theories; it introduces ReSynth as a new conceptual entity without new empirical anchors or derivations.

axioms (1)

domain assumption Neither connectionism nor classicism can make systematicity a structural consequence of the architecture (Aizawa demonstration)
Invoked in the first stage to establish that corrective techniques are auxiliary hypotheses.

invented entities (1)

ReSynth no independent evidence
purpose: Trimodular conceptual framework that separates reasoning, identity, and memory to make systematic behaviour structural
Newly proposed architecture intended to overcome the limitations traced to psychological paradigms

pith-pipeline@v0.9.0 · 5556 in / 1387 out tokens · 51507 ms · 2026-05-15T09:11:05.276803+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

75 extracted references · 75 canonical work pages · 4 internal anchors

[1]

Springer (2009)

Michalski, R.S., Carbonell, J.G., Mitchell, T.M.: Machine Learning: An Artificial Intelligence Approach. Springer (2009)

work page 2009
[2]

Prestianni, T.: Behaviorism learning theory (2023)

work page 2023
[3]

In: Encyclopedia of the Sciences of Learning

Graham, S.: Behaviorism. In: Encyclopedia of the Sciences of Learning . Springer (2012)

work page 2012
[4]

International Publishers (1930)

Pavlov, I.: Lectures on Conditioned Reflexes. International Publishers (1930)

work page 1930
[5]

Skinner’s Verbal Behavior

Chomsky, N.: A review of B.F. Skinner’s Verbal Behavior . Language 35(1), 26–58 (1959)

work page 1959
[7]

Against Personalized AI Moral Advisors: Commentary on ‘Can AI Rely on the Systematicity of Truth?’ by Matthieu Queloz

Leuenberger, M. Against Personalized AI Moral Advisors: Commentary on ‘Can AI Rely on the Systematicity of Truth?’ by Matthieu Queloz. Philosophy & Technology. 38 (2025), https://api.semanticscholar.org/CorpusID:277685745

work page 2025
[8]

On the Fundamental Limitations of AI Moral Advisors

Queloz, M. On the Fundamental Limitations of AI Moral Advisors. Philosophy & Technology . 38, 71 (2025,5), https://doi.org/10.1007/s13347 -025-00896 -3

work page doi:10.1007/s13347 2025
[9]

Perspectives on Psycholog - ical Science 1(2), 164–180 (2006)

Bandura, A.: Toward a psychology of human agency. Perspectives on Psycholog - ical Science 1(2), 164–180 (2006)

work page 2006
[10]

Expert Systems with Applications 231 (2023)

Shakya, S., et al.: Reinforcement learning algorithms: A brief survey. Expert Systems with Applications 231 (2023)

work page 2023
[11]

MIT Press (1998)

Sutton, R.S., Barto, A.G.: Reinforcement Learning: An Introduction. MIT Press (1998)

work page 1998
[12]

Deubel, S.W.: An investigation of behaviorist and cognitive approaches to instruc - tional multimedia design. J. Educational Multimedia and Hypermedia 12(1) (2003)

work page 2003
[13]

Performance Improvement Quarterly 6(4), 50–72 (1993)

Ertmer, P.A., Newby, T.J.: Behaviorism, cognitivism, constructivism: Comparing critical features. Performance Improvement Quarterly 6(4), 50–72 (1993)

work page 1993
[14]

Pearson (2012) 23

Ormrod, J.E.: Human Learning, 6th edn. Pearson (2012) 23

work page 2012
[15]

Word embeddings: A survey,

Almeida, E., Xexeo, G.: Word embeddings: A survey. arXiv:1901.09069 (2019)

work page arXiv 1901
[16]

Neural Computation 9(8), 1735–1780 (1997)

Hochreiter, S., Schmidhuber, J.: Long short-term memory. Neural Computation 9(8), 1735–1780 (1997)

work page 1997
[17]

Neurocom - puting 452, 48–62 (2021)

Niu, Z., et al.: A review on the attention mechanism of deep learning. Neurocom - puting 452, 48–62 (2021)

work page 2021
[18]

IEEE Trans

Hospedales, T., et al.: Meta -learning in neural networks: A survey. IEEE Trans. Pattern Analysis and Machine Intelligence 44(9), 5149–5169 (2021)

work page 2021
[19]

Pearson Research Report (2011)

Lai, E.R.: Metacognition: A literature review. Pearson Research Report (2011)

work page 2011
[20]

Alahmad, B.A.: A review of cogni tivism and its relationship with e-learning (2020)

work page 2020
[21]

A Path Towards Autonomous Machine Intelligence

LeCun, Y. A Path Towards Autonomous Machine Intelligence. (2022), Open Review preprint, version 0.9.2

work page 2022
[22]

& Rosch, E

Varela, F., Thompson, E. & Rosch, E. The Embodied Mind: Cognitive Science and Human Experience. (MIT Press,1991)

work page 1991
[23]

Psychology of Learning and Motivation 24, 109–165 (1989)

McCloskey, M., Cohen, N.J.: Catastrophic interference in connectionist networks. Psychology of Learning and Motivation 24, 109–165 (1989)

work page 1989
[24]

Institute for Inquiry (2023)

Hein, G.E.: Constructivist learning theory. Institute for Inquiry (2023)

work page 2023
[25]

Advances in Social Science, Education and Humanities Research (2021)

Efgivia, T., et al.: Constructivism approach in learning. Advances in Social Science, Education and Humanities Research (2021)

work page 2021
[26]

In: Proceedings of ICML, pp

Bengio, Y., et al.: Curriculum learning. In: Proceedings of ICML, pp. 41–48 (2009)

work page 2009
[27]

Artificial Intelligence Review 18, 77–95 (2002)

Vilalta, R., Drissi, Y.: A perspective view and survey of meta -learning. Artificial Intelligence Review 18, 77–95 (2002)

work page 2002
[28]

Distilling the Knowledge in a Neural Network

Hinton, G., Vinyals, O., Dean, J.: Distilling the knowledge in a neural network. arXiv:1503.02531 (2015)

work page internal anchor Pith review Pith/arXiv arXiv 2015
[29]

Asian EFL Journal (2004)

Li, X.: An analysis of Chinese EFL learners’ beliefs about the role of rote learning in vocabulary learning strategies. Asian EFL Journal (2004)

work page 2004
[30]

Journal of AI Research (2017)

Ahmed, S., et al.: Comparative analysis of learning approaches. Journal of AI Research (2017)

work page 2017
[31]

In: The Chinese Learner (1999)

Biggs, J.: Western misperceptions of the Confucian -heritage learning culture. In: The Chinese Learner (1999)

work page 1999
[32]

Kirkpatrick, J., et al.: Overcoming catastrophic forgetting in neural networks. Proc. National Academy of Sciences 114(13), 3521–3526 (2017) 24

work page 2017
[33]

PLoS Computational Biology 11(4), e1004128 (2015)

Ellefsen, K.O., Mouret, J.B., Clune, J.: Neural modularity helps organisms evolve to learn new skills without forgetting old skills. PLoS Computational Biology 11(4), e1004128 (2015)

work page 2015
[34]

arXiv:1912.04508 (2019)

Aly, R., Dugan, T.: Reducing catastrophic forgetting in modular neural networks by dynamic information balancing. arXiv:1912.04508 (2019)

work page arXiv 1912
[35]

Not all llm reasoners are created equal.arXiv preprint arXiv:2410.01748,

Hosseini, A., Sordoni, A., Toyama, D., Courville, A. & Agarwal, R. Not All LLM Reasoners Are Created Equal. (2024), https://arxiv.org/abs/2410.01748

work page arXiv 2024
[36]

ACL Findings, pp

Li, Z., et al.: Understanding and patching compositional reasoning in LLMs. ACL Findings, pp. 9668–9688 (2024)

work page 2024
[37]

& Cioffi, J

Mohsin, M., Umer, M., Bilal, A., Memon, Z., Qadir, M., Bhattacharya, S., Rizwan, H., Gorle, A., Kazmi, M., Amir, N., Subhan, A., Rafique, M., He, Z., Mehta, P., Jamshed, M. & Cioffi, J. On the Fundamental Limits of LLMs at Scale. (2026), https://arxiv.org/abs/2511.12869

work page arXiv 2026
[38]

Cognition 28, 3–71 (1988)

Fodor, J.A., Pylyshyn, Z.W.: Connectionism and cognitive architecture: A critical analysis. Cognition 28, 3–71 (1988)

work page 1988
[39]

Mind and Language 12(2), 115–136 (1997)

Aizawa, K.: Explaining systematicity. Mind and Language 12(2), 115–136 (1997)

work page 1997
[40]

NeurIPS (2022)

Wei, J., et al.: Chain -of-thought prompting elicits reasoning in large language models. NeurIPS (2022)

work page 2022
[41]

NeurIPS (2020)

Lewis, P., et al.: Retrieval -augmented generation for knowledge -intensive NLP tasks. NeurIPS (2020)

work page 2020
[42]

NeurIPS (2022)

Ouyang, L., et al.: Training language models to follow instructions with human feedback. NeurIPS (2022)

work page 2022
[43]

On the Measure of Intelligence

Chollet, F.: On the measure of intelligence. arXiv:1911.01547 (2019)

work page internal anchor Pith review Pith/arXiv arXiv 1911
[44]

& Ponti, E

Pfeiffer, J., Ruder, S., Vuli´c, I. & Ponti, E. Modular Deep Learning. (2024), https://arxiv.org/abs/2302.11529

work page arXiv 2024
[45]

arXiv preprint arXiv:2211.08411 , year=

Kandpal, N., Deng, H., Roberts, A., Wallace, E. & Raffel, C. Large Language Models Struggle to Learn Long -Tail Knowledge. (2023), https://arxiv.org/abs/2211.08411

work page arXiv 2023
[46]

arXiv:1606

Rusu, A.A., et al.: Progressive neural networks. arXiv:1606. 04671 (2016)

work page 2016
[47]

ACL (2020)

Bender, E.M., Koller, A.: Climbing towards NLU: On meaning, form, and understanding in the age of data. ACL (2020)

work page 2020
[48]

arXiv:2002.06177 (2020) 25

Marcus, G.: The next decade in AI: Four steps towards robust artificial intelli - gence. arXiv:2002.06177 (2020) 25

work page arXiv 2002
[49]

Harvard Univer - sity Press (1996)

Dunbar, R.: Grooming, Gossip, and the Evolution of Language . Harvard Univer - sity Press (1996)

work page 1996
[50]

MIT Press (2008)

Tomasello, M.: Origins of Human Communication. MIT Press (2008)

work page 2008
[51]

Generalization without systematicity: On the compositional skills of sequence-to-sequence recurrent networks

Lake, B. & Baroni, M. Generalization without systematicity: On the compositional skills of sequence -to-sequence recurrent networks. (2018), https://arxiv.org/abs/1711.00350

work page internal anchor Pith review Pith/arXiv arXiv 2018
[52]

& Balestriero, R

Huang, H., LeCun, Y. & Balestriero, R. Semantic Tube Prediction: Beating LLM Data Efficiency with JEPA. (2026), https://arxiv.org/abs/2602.22617

work page arXiv 2026
[53]

Anderson, J.R.: How Can the Human Mind Occur in the Physical Universe? Oxford University Press (2007)

work page 2007
[54]

MIT Press (2012)

Laird, J.E.: The Soar Cognitive Architecture. MIT Press (2012)

work page 2012
[55]

PLDI, pp

Ellis, K., Wong, C., Nye, M., et al.: DreamCoder: Bootstrapping inductive program synthesis with wake -sleep library learning. PLDI, pp. 835 –850 (2021)

work page 2021
[56]

Artificial Intelligence Review 56, 12387 –12406 (2023)

Garcez, A.d., Lamb, L.C.: Neurosymbolic AI: The 3rd wave. Artificial Intelligence Review 56, 12387 –12406 (2023)

work page 2023
[57]

IJCAI Survey Track (2020)

Lamb, L.C., Garcez, A.d., Gori, M., et al.: Graph neural networks meet neural - symbolic computing: A survey and perspective. IJCAI Survey Track (2020)

work page 2020
[58]

Cambridge University Press (1988)

Baars, B.J.: A Cognitive Theory of Consciousness . Cambridge University Press (1988)

work page 1988
[59]

-P., Naccache, L.: The global neuronal workspace model of conscious access

Dehaene, S., Changeux, J. -P., Naccache, L.: The global neuronal workspace model of conscious access. Biological Bulletin 221(1), 76–93 (2011)

work page 2011
[60]

Oxford University Press (2010)

Shanahan, M.: Embodiment and the Inner Life: Cognition and Consciousness in the Space of Possible Minds. Oxford University Press (2010)

work page 2010
[61]

Goyal, A., Bengio, Y.: Inductive biases for deep learning of higher-level cognition. Proc. Royal Society A 478(2266), 20210068 (2022)

work page 2022
[62]

Behavioral and Brain Sciences 36(3), 181–204 (2013)

Clark, A.: Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behavioral and Brain Sciences 36(3), 181–204 (2013)

work page 2013
[63]

Oxford University Press (2015)

Clark, A.: Surfing Uncertainty: Prediction, Action, and the Embodied Mind. Oxford University Press (2015)

work page 2015
[64]

Routledge (2004) 26

Boden, M.A.: The Creative Mind: Myths and Mechanisms, 2nd edn. Routledge (2004) 26

work page 2004
[65]

NeurIPS (2023)

Conmy, A., Mavor-Parker, A.N., Lynch, A., et al.: Towards automated circuit discovery for mechanistic interpretability. NeurIPS (2023)

work page 2023
[66]

Transformer Circuits Thread, Anthropic (2022)

Elhage, N., et al.: Toy models of superposition. Transformer Circuits Thread, Anthropic (2022)

work page 2022
[67]

Compu- tational Linguistics 48(1), 207–219 (2022)

Belinkov, Y.: Probing classifiers: Promises, shortcomings, and advances. Compu- tational Linguistics 48(1), 207–219 (2022)

work page 2022
[68]

Springer (2013)

Mac Lane, S.: Categories for the Working Mathematician, 2nd edn. Springer (2013)

work page 2013
[69]

Cambridge University Press (2019)

Fong, B., Spivak, D.I.: An Invitation to Applied Category Theory: Seven Ske tches in Compositionality. Cambridge University Press (2019)

work page 2019
[70]

Constitutional AI: Harmlessness from AI Feedback

Bai, Y., et al.: Constitutional AI: Harmlessness from AI feedback. arXiv:2212.08073 (2022)

work page internal anchor Pith review Pith/arXiv arXiv 2022
[71]

Cognition 35, 183–204 (1990)

Fodor, J., McLaughlin, B.: Connectionism and the problem of systematicity: Why Smolensky’s solution doesn’t work. Cognition 35, 183–204 (1990)

work page 1990
[72]

NeurIPS (2017)

Rockt¨aschel, T., Riedel, S.: End-to-end differentiable proving. NeurIPS (2017)

work page 2017
[73]

NeurIPS (2018)

Manhaeve, R., Dumancic, S., Kimmig, A., et al.: DeepProbLog: Neural proba - bilistic logic programming. NeurIPS (2018)

work page 2018
[74]

PLDI (2023)

Li, Z., et al.: Scallop: A language for neurosymbolic programming. PLDI (2023)

work page 2023
[75]

Presentation at AAAI 2020 Spring Symposium (2020)

Marcus, G.: Neurosymbolic AI and common sense. Presentation at AAAI 2020 Spring Symposium (2020)

work page 2020
[76]

ICML (2017)

Finn, C., Abbeel, P., Levine, S.: Model-agnostic meta-learning for fast adaptation of deep networks. ICML (2017)

work page 2017