Pith. sign in

REVIEW 2 major objections 4 minor 169 references

Cognitive Convergence: Deep Similarities Between Large Language Models and Human Cognition

T0 review · 2 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read LLMs and human minds are cognitive cousins, not alien intelligences.

desk verdict A plausible, well-sourced synthesis arguing LLMs and humans converge on core cognitive principles, but its central mechanistic claim is asserted rather than demonstrated. read the letter →

arxiv 2607.26179 v1 pith:5FQEZDOR submitted 2026-07-28 q-bio.NC cs.AIcs.CL

Chandra Sripada , Richard Lewis This is my paper · ORCID
classification q-bio.NCcs.AIcs.CL
keywords LLMcognitioncognitiveconvergencedual-processtheoryproductionsystemsrepresentationalalignmentprediction-drivenlearningreinforcementarchitecturalcanalization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that the widespread view of LLMs as 'alien intelligences' is mistaken. Despite significant differences in physical substrate, training history, and environment, humans and contemporary LLM-based systems converge on five principles of cognitive organization with longstanding support in cognitive science: dual-process inference, production-system-like architecture, representational alignment, prediction-driven learning, and reinforcement-learning-like goal-directed control. The authors claim these are not superficial behavioral similarities but structural correspondences at the level of computational organization. To explain why such different systems converge, they propose two complementary principles: contravariance, which says harder tasks leave fewer viable solutions, and architectural canalization, which says a shared production-system-like architecture biases learning toward similar internal solutions. If the thesis is right, LLM–human parallels are evidence of genuinely shared computational principles, making humans and LLMs 'different members of a shared computational family.'

What carries the argument

The central mechanism is the transformer's residual stream: a persistent, high-dimensional vector at each token position that attention heads and MLP blocks read from and incrementally write back to, creating a graded condition-action system with modular, composable updates. The paper argues this is structurally analogous to the working memory and production rules of cognitive architectures. Two explanatory principles carry the convergence: contravariance (demanding tasks shrink the space of viable internal solutions) and architectural canalization (systems built on a shared production-system-like architecture will tend to converge on similar solutions even under different training regimes).

What would settle it

A concrete test: take a non-production-like architecture (e.g., a fully connected or convolutional network without a residual stream) and train it to match an LLM's performance on human-language tasks. If it shows equally strong alignment with human neural representations and behavioral error profiles under representational similarity analysis, then task demands alone, not shared architecture, explain the convergence, and the canalization claim is undermined. Alternatively, find a transformer variant whose residual stream is removed or radically altered but that still matches LLM behavior; if

Watch

Extended reading notes

Core claim

The paper's central claim is that LLM-based systems and human cognition are 'importantly different members of a shared computational family.' The authors ground this in five correspondences: both exhibit dual-process organization, with fast compiled responses alongside slower serial reasoning that uses externalized scratchpads; both are organized like production systems, where a persistent workspace is modified by many context-sensitive conditional operators; internal representations align, as shown by LLM internal units encoding linguistic constructs that predict human reading times and brain responses, with model depth tracking the cortical hierarchy; both learn through prediction and erro

Load-bearing premise

The argument's load-bearing premise is that transformer LLMs are genuinely production-system-like in their computational organization, so that a shared architecture—not merely shared task demands—biases both humans and LLMs toward the same internal solutions; if the residual stream is not truly analogous to human working memory plus production rules, the deep convergence thesis loses its mechanistic support.

Editorial extensions

If this is right

  • If convergence is genuine, behavioral similarities—like comparable garden-path difficulty, serial-position effects, and conjunctive-search costs—should be treated as evidence of shared mechanisms, not as products of text mimicry.
  • The dual-process mapping predicts that LLM internal 'thinking token' counts should track human reaction times across many reasoning tasks, a pattern the paper reports with strong correlations.
  • Representational alignment implies that next-word prediction performance should continue to predict human neural and reading-time measures, and that layerwise activations should map onto cortical hierarchy, as current data show near the noise ceiling.
  • The sample-efficiency gap between LLMs and children is a difference in the learning process, not necessarily the learning target; adding structural inductive biases, world models, and curiosity-like exploration may reduce the gap while preserving cognitive convergence.
  • In the RL domain, the paper's claim implies that current artificial limitations—reward hacking, sparse rewards, absent homeostatic signals—are engineering gaps rather than evidence of a fundamentally alien agency, and may be narrowed by richer training experience and reward design.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If architectural canalization holds, a testable prediction follows: two transformer-based systems trained on very different corpora (e.g., text-only vs. multimodal) should still develop similar human-aligned internal representations on shared tasks, whereas a non-production-like architecture trained to the same performance should not. This could be probed with representational similarity analysis
  • The thesis implies that the space of feasible intelligent cognition is narrower than often assumed, which bears on AI safety: if genuinely alien forms of intelligence are rare, value alignment may be more tractable than if intelligence can take radically unconstrained forms.
  • A useful extension would be to formalize 'canalization' as a quantitative bias: measure the distribution of internal solutions across random initializations and training runs, and test whether production-system-like architectures show lower variance and stronger attraction to human-like solutions than equally powerful non-residual architectures.
  • The paper's treatment of RL suggests a specific empirical target: current frontier models should be evaluated for whether longer-horizon RL training induces functionally distinct controllers and metacontrol (e.g., arbitration between habitual and deliberative policies), which would further confirm the convergence.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. This paper argues against the 'alien intelligence' framing of large language models. It claims that, despite differences in substrate, learning history, and environment, humans and contemporary LLM-based systems converge on several core principles of cognitive organization: dual-process inferential organization, production-system-like computational architecture, aligned representational structure, prediction-driven learning, and reinforcement-learning-like goal-directed control. The authors review a wide range of empirical and theoretical literature, propose contravariance and a new 'architectural canalization' mechanism to explain representational alignment, and conclude that humans and LLMs are 'cognitive cousins'—different members of a shared computational family. The paper is a literature synthesis rather than a new empirical study.

Significance. If the convergence thesis holds, the paper provides a valuable counterweight to the widely held assumption that LLM-human similarities are superficial or merely anthropomorphic. Its strengths are its breadth of relevant literature, its explicit acknowledgment of important differences (sample efficiency, limited RL infrastructure), and its framing of the debate in terms of mechanistic levels rather than behavioral mimicry. The paper also makes potentially testable claims, e.g., that architecture should modulate brain/behavioral alignment, and it honestly flags open empirical questions. However, the central explanatory mechanism—architectural canalization—is currently asserted rather than demonstrated, and it is load-bearing for the 'shared computational family' conclusion. The paper is best suited as a perspective/review that will inform future work, but the strength of its conclusions currently exceeds the support for its main mechanistic premise.

major comments (2)
  1. [§4–§5] The production-system analogy is too under-constrained to support the canalization premise. In §4, the defining features are 'a persistent representational workspace [modified] by a large population of context-sensitive operators,' with attention heads and MLPs described as 'graded, differentiable' condition matching. As stated, this description applies equally to ResNets, gated RNNs, and state-space models, so it does not distinguish transformers within the class of neural sequence models. §5 then relies on this analogy: 'two systems built around similar production-system-like architectures... will tend to converge on similar solutions.' Without a non-vacuous criterion (e.g., modularity, recombinability, compositionality) and evidence that the residual-stream architecture specifically biases learning toward human-like representations, the canalization explanation cannot carry the weight
  2. [§5 (Architectural Canalization)] The 'architectural canalization' hypothesis is asserted with citations to Waddington and Ariew, but no mechanistic account is given for how the residual-stream architecture biases gradient descent toward particular representational solutions. Biological canalization concerns developmental buffering against perturbations; its transfer to transformer training requires a model of the learning dynamics—not just a loose analogy. The claim is central: it is one of two explanations for representational alignment, and the conclusion explicitly invokes 'learning that unfolds within a shared computational architecture.' Without support, the explanatory force reduces to contravariance, which is already established and does not require shared architecture. Please either supply a mechanistic model or direct evidence (e.g., comparing brain/behavioral alignment across transformers, state-space models,
minor comments (4)
  1. [§7 heading] The section heading reads 'Reinforcement Learning and Reinforcement Learning and Agentive Control'—the duplicated phrase should be removed.
  2. [§7] Minor grammatical issue: 'the RL infrastructure current deployed in artificial systems' should be 'currently deployed.'
  3. [§6] The Ingrosso & Goldt analogy is introduced as supporting the 'same destination despite different routes' claim, but the discussion in the same paragraph (and ref 126) emphasizes sample-efficiency differences. It would help to explicitly distinguish the two uses of this example: task-driven convergence of representations vs. efficiency of the learning route.
  4. [§2] The behavioral parallels section includes vision-language models and text-to-image models, not just LLMs. A brief clarification of the intended scope of 'LLM-based systems' at first mention would avoid ambiguity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is a literature synthesis whose central claims rest on independent empirical citations; the canalization proposal is an explicitly tentative explanatory hypothesis, not a fitted prediction or derivation from its own inputs.

full rationale

This is a review/synthesis paper, not a derivation with fitted parameters or generated predictions. The central thesis—that LLMs and humans share five principles of cognitive organization—is supported by a broad set of cited empirical studies (e.g., Schrimpf et al. 2021, Mischler et al. 2024, de Varda et al. 2025), many of which are external to the authors. The authors' own self-citations (e.g., Lewis & Vasishth 2005; Ryu & Lewis 2025; Sripada 2025, 2026) are used as background or as one piece of a larger evidentiary mosaic, not as the sole or load-bearing justification for the convergence claim. The production-system analogy in §4 is admittedly loose—the authors explicitly state transformers are 'graded, differentiable' and do not 'literally implement classical production systems'—so the analogy is under-constrained as a mechanistic explanation, but under-constraint is a correctness/evidential concern, not circularity. Similarly, the 'architectural canalization' hypothesis in §5 is proposed after the fact as a complementary explanation of observed convergence ('We propose a second, complementary explanation'), and the paper itself flags open empirical questions about sample efficiency and whether richer RL infrastructures can emerge. No equation, fit, or prediction in the paper reduces by construction to its inputs. Therefore no circular step is present.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The paper's central claim rests on a chain of conceptual assumptions: the choice of comparison dimensions, the validity of contravariance, and the novel 'architectural canalization' principle. There are no fitted free parameters or invented physical entities; the load is carried by high-level explanatory principles that are plausible but not independently confirmed.

assumptions (6)
  • domain assumption Behavioral resemblance alone cannot settle the issue of shared cognitive organization (Introduction, §2).
    The paper's entire project presupposes that deeper mechanistic comparison is necessary and informative; this is a standard methodological stance in cognitive science but not proven.
  • domain assumption The five identified dimensions (inferential organization, computational architecture, representational structure, prediction-driven learning, RL-like mechanisms) are the right levels for comparing intelligent systems (§1).
    The choice of dimensions is motivated by cognitive science traditions, but it is the authors' selection and could bias the convergence finding.
  • domain assumption Contravariance: as tasks become more demanding, the space of viable solutions narrows (§5).
    Adopted from Cao & Yamins (2024); it is used as an explanation for convergence but is itself a general principle, not verified for LLMs in this paper.
  • ad hoc to paper Architectural canalization: shared production-system-like architecture biases the solutions that emerge from learning toward similar representational structures (§5).
    This is the paper's own proposed explanatory mechanism, supported only by analogy to biological canalization (Waddington, Ariew) and not independently demonstrated for transformer LLMs.
  • domain assumption Human-level cognitive skills are among the hardest optimization targets, so systems that achieve them will converge (§5).
    The paper asserts this to motivate contravariance; it is plausible but not established, and if false, the convergence would need another explanation.
  • ad hoc to paper Circuitousness in learning route (more data, less inductive bias) is not evidence of a different learning destination (§6).
    The Ingrosso & Goldt analogy supports the 'same destination' claim for one task, but extending it to LLMs and human cognition is an extrapolation the paper asserts without direct evidence.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Cognitive Convergence: Deep Similarities Between Large Language Models and Human Cognition." pith.science (2026). https://pith.science/paper/5FQEZDOR

@misc{pith2026260726179,
  author       = {Pith},
  title        = {Pith review of: Cognitive Convergence: Deep Similarities Between Large Language Models and Human Cognition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5FQEZDOR}},
  note         = {Machine review of arXiv:2607.26179}
}
read the original abstract

LLMs are widely regarded as alien intelligences, systems whose cognitive operations are fundamentally unlike our own. Apparent similarities to human cognition are therefore often seen as the result of anthropomorphic projection. We argue that this framing is mistaken. LLMs clearly differ from humans in important respects, including their physical substrate, learning history, and the environments with which they interact. These differences make it all the more striking that contemporary LLM-based systems converge with human cognition on a number of principles of cognitive organization with longstanding support in cognitive science. We identify structural correspondences across five dimensions: inferential organization, computational architecture, representational structure, prediction-driven learning, and reinforcement-learning-like mechanisms supporting goal-directed action. These correspondences support a broader model of intelligent cognition in which core principles long used to explain human intelligence also characterize contemporary LLM-based systems.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

169 extracted references · 1 linked inside Pith

  1. [1]

    Harari, Y. N. Nexus: A Brief History of Information Networks from the Stone Age to AI. (Random House, New York, 2024)

    2024

  2. [2]

    M., Gebru, T., McMillan-Major, A

    Bender, E. M., Gebru, T., McMillan-Major, A. & Shmitchell, S. On the dangers of stochastic parrots: Can language models be too big? in Proceedings of the 2021 ACM conference on fairness, accountability, and transparency 610–623 (2021)

    2021

  3. [3]

    K., Belkin, M., Bergen, L

    Chen, E. K., Belkin, M., Bergen, L. & Danks, D. Does AI already have human-level intelligence? The evidence is clear. Nature 650, 36–40 (2026)

    2026

  4. [4]

    & Norvig, P

    Agüera y Arcas, B. & Norvig, P. Artificial General Intelligence is Already Here. Noema https://www.noemamag.com/artificial-general-intelligence-is-already-here/ (2023)

    2023

  5. [5]

    Bengio, Y. et al. Managing extreme AI risks amid rapid progress. Science 384, 842–845 (2024)

    2024

  6. [6]

    Talking about large language models

    Shanahan, M. Talking about large language models. Commun. ACM 67, 68–79 (2024)

    2024

  7. [7]

    Bender, E. M. & Koller, A. Climbing towards NLU: On meaning, form, and understanding in the age of data. in Proceedings of the 58th annual meeting of the association for computational linguistics 5185–5198 (2020)

    2020

  8. [8]

    & Reynolds, L

    Shanahan, M., McDonell, K. & Reynolds, L. Role play with large language models. Nature 623, 493–498 (2023)

    2023

Show all 169 references
  1. [9]

    The superintelligent will: Motivation and instrumental rationality in advanced artificial agents

    Bostrom, N. The superintelligent will: Motivation and instrumental rationality in advanced artificial agents. Minds Mach. 22, 71–85 (2012)

    2012

  2. [10]

    & Watumull, J

    Chomsky, N., Roberts, I. & Watumull, J. The false promise of chatgpt. N. Y. Times 8, 177– 179 (2023)

    2023

  3. [11]

    No, Artificial Intelligence Is Not Conscious

    Chiang, T. No, Artificial Intelligence Is Not Conscious. The Atlantic (2026)

    2026

  4. [12]

    & Reviriego, P

    Fu, T., Ferrando, R., Conde, J., Arriaga, C. & Reviriego, P. Why Do Large Language Models (LLMs) Struggle to Count Letters? arXiv preprint arXiv:2412.18626 (2024)

    2024 arXiv

  5. [13]

    A is B” fail to learn “B is A

    Berglund, L. et al. The Reversal Curse: LLMs trained on “A is B” fail to learn “B is A”. in International Conference on Learning Representations vol. 2024 18623–18642 (2024)

    2024

  6. [14]

    Lewis, R. L. & Vasishth, S. An activation‐based model of sentence processing as skilled memory retrieval. Cogn. Sci. 29, 375–419 (2005)

    2005

  7. [15]

    Blaubergs, M. S. & Braine, M. D. Short-term memory limitations on decoding self- embedded sentences. J. Exp. Psychol. 102, 745 (1974)

    1974

  8. [16]

    C., Hendrick, R

    Gordon, P. C., Hendrick, R. & Johnson, M. Memory interference during language processing. J. Exp. Psychol. Learn. Mem. Cogn. 27, 1411 (2001)

    2001

  9. [17]

    Wason, P. C. & Reich, S. S. A verbal illusion. Q. J. Exp. Psychol. 31, 591–597 (1979)

    1979

  10. [18]

    Can language models handle recursively nested grammatical structures? A case study on comparing models and humans

    Lampinen, A. Can language models handle recursively nested grammatical structures? A case study on comparing models and humans. Comput. Linguist. 50, 1441–1476 (2024)

    2024

  11. [19]

    Li, A. et al. Incremental comprehension of garden-path sentences by large language models: Semantic interpretation, syntactic re-analysis, and attention. ArXiv Prepr. ArXiv240516042 (2024)

    2024

  12. [20]

    J., Meltzer-Asscher, A

    Amouyal, S. J., Meltzer-Asscher, A. & Berant, J. When the LM misunderstood the human chuckled: Analyzing garden path effects in humans and language models. in Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) 8235–8...

    2025

  13. [21]

    J., Meltzer-Asscher, A

    Amouyal, S. J., Meltzer-Asscher, A. & Berant, J. Comparing human and language models sentence processing difficulties on complex structures. in Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) 22704– 22722 (2026)

    2026

  14. [22]

    Murdock Jr, B. B. The serial position effect of free recall. J. Exp. Psychol. 64, 482 (1962)

    1962

  15. [23]

    Kahana, M. J. Associative retrieval processes in free recall. Mem. Cognit. 24, 103–109 (1996)

    1996

  16. [24]

    Liu, N. F. et al. Lost in the middle: How language models use long contexts. Trans. Assoc. Comput. Linguist. 12, 157–173 (2024)

    2024

  17. [25]

    Y., Benna, M

    Ji-An, L., Zhou, C. Y., Benna, M. K. & Mattar, M. G. Linking in-context learning in transformers to human episodic memory. Adv. Neural Inf. Process. Syst. 37, 6180–6212 (2024)

    2024

  18. [26]

    & Schulz, E

    Binz, M. & Schulz, E. Using cognitive psychology to understand GPT-3. Proc. Natl. Acad. Sci. 120, e2218523120 (2023)

    2023

  19. [27]

    Webb, T., Holyoak, K. J. & Lu, H. Emergent analogical reasoning in large language models. Nat. Hum. Behav. 7, 1526–1541 (2023)

    2023

  20. [28]

    W., Holyoak, K

    Webb, T. W., Holyoak, K. J. & Lu, H. Evidence from counterfactual tasks supports emergent analogical reasoning in large language models. PNAS Nexus 4, pgaf135 (2025)

    2025

  21. [29]

    Evaluating large language models in theory of mind tasks

    Kosinski, M. Evaluating large language models in theory of mind tasks. Proc. Natl. Acad. Sci. 121, e2405460121 (2024)

    2024

  22. [30]

    R., Trott, S

    Jones, C. R., Trott, S. & Bergen, B. Comparing humans and large language models on an Experimental Protocol Inventory for Theory of Mind Evaluation (EPITOME). Trans. Assoc. Comput. Linguist. 12, 803–819 (2024)

    2024

  23. [31]

    & Pickering, M

    Cai, Z., Duan, X., Haslett, D., Wang, S. & Pickering, M. Do large language models resemble humans in language use? in Proceedings of the workshop on cognitive modeling and computational linguistics 37–56 (2024)

    2024

  24. [32]

    Binz, M. et al. Centaur: a foundation model of human cognition. ArXiv Prepr. ArXiv241020268 (2024)

    2024

  25. [33]

    W., Huang, X

    Bini, P., Cong, L. W., Huang, X. & Jin, L. J. Behavioral Economics of AI: LLM Biases and Corrections. Available SSRN 5213130 (2025)

    2025

  26. [34]

    Treisman, A. M. & Gelade, G. A feature-integration theory of attention. Cognit. Psychol. 12, 97–136 (1980)

    1980

  27. [35]

    Campbell, D. et al. Understanding the limits of vision language models through the lens of the binding problem. Adv. Neural Inf. Process. Syst. 37, 113436–113460 (2024)

    2024

  28. [36]

    L., Lord, M

    Kaufman, E. L., Lord, M. W., Reese, T. W. & Volkmann, J. The discrimination of visual number. Am. J. Psychol. 62, 498–525 (1949)

    1949

  29. [37]

    & Griffiths, T

    Milli, S., Lieder, F. & Griffiths, T. L. A rational reinterpretation of dual-process theories. Cognition 217, 104881 (2021)

    2021

  30. [38]

    & Griffiths, T

    Lieder, F., Shenhav, A., Musslick, S. & Griffiths, T. L. Rational metareasoning and the plasticity of cognitive control. PLoS Comput. Biol. 14, e1006043 (2018)

    2018

  31. [39]

    Stanovich, K. E. & West, R. F. Individual differences in reasoning: Implications for the rationality debate. Behav. Brain Sci. 23, 645–665 (2000)

    2000

  32. [40]

    Evans, J. S. B. T. Dual-processing accounts of reasoning, judgment, and social cognition. Annu. Rev. Psychol. 59, 255–278 (2008)

    2008

  33. [41]

    Thinking, Fast and Slow

    Kahneman, D. Thinking, Fast and Slow. (Farrar, Straus and Giroux, 2011)

    2011

  34. [42]

    Radford, A. et al. Language models are unsupervised multitask learners. OpenAI Blog 1, 9 (2019)

    2019

  35. [43]

    Brown, T. et al. Language models are few-shot learners. Adv. Neural Inf. Process. Syst. 33, 1877–1901 (2020)

    1901

  36. [44]

    & Frank, M

    Russin, J., Pavlick, E. & Frank, M. J. Parallel trade-offs in human cognition and neural networks: The dynamic interplay between in-context and in-weight learning. Proc. Natl. Acad. Sci. 122, e2510270122 (2025)

    2025

  37. [45]

    Petroni, F. et al. Language models as knowledge bases? in Proceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing (EMNLP-IJCNLP) 2463–2473 (2019)

    2019

  38. [46]

    & Shazeer, N

    Roberts, A., Raffel, C. & Shazeer, N. How much knowledge can you pack into the parameters of a language model? in Proceedings of the 2020 conference on empirical methods in natural language processing (EMNLP) 5418–5426 (2020)

    2020

  39. [47]

    & Rush, A

    Davison, J., Feldman, J. & Rush, A. M. Commonsense knowledge mining from pretrained models. in Proceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing (EMNLP-IJCNLP) 1173–1...

    2019

  40. [48]

    M., Raghunathan, A., Liang, P

    Xie, S. M., Raghunathan, A., Liang, P. & Ma, T. An Explanation of In-context Learning as Implicit Bayesian Inference. ArXiv Prepr. ArXiv211102080 (2021)

    2021

  41. [49]

    Garg, S., Tsipras, D., Liang, P. S. & Valiant, G. What can transformers learn in-context? a case study of simple function classes. Adv. Neural Inf. Process. Syst. 35, 30583–30598 (2022)

    2022

  42. [50]

    Nye, M. et al. Show your work: Scratchpads for intermediate computation with language models. (2021)

    2021

  43. [51]

    Wei, J. et al. Chain-of-thought prompting elicits reasoning in large language models. Adv. Neural Inf. Process. Syst. 35, 24824–24837 (2022)

    2022

  44. [52]

    Guo, D. et al. DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning. Nature 645, 633–638 (2025)

    2025

  45. [53]

    Artificial intelligence learns to reason

    Mitchell, M. Artificial intelligence learns to reason. Science 387, eadw5211 (2025)

    2025

  46. [54]

    Hu, X. N. et al. Conflict and Congruency Effects in Large Language Models: In-Weight and In-Context Competition in a Verbal Conflict Task. (2026)

    2026

  47. [55]

    G., D’Elia, F

    de Varda, A. G., D’Elia, F. P., Kean, H., Lampinen, A. & Fedorenko, E. The cost of thinking is similar between large reasoning models and humans. Proc. Natl. Acad. Sci. 122, e2520077122 (2025)

    2025

  48. [56]

    Production systems: Models of control structures

    Newell, A. Production systems: Models of control structures. in Visual information processing 463–526 (Elsevier, 1973)

    1973

  49. [57]

    Unified Theories of Cognition

    Newell, A. Unified Theories of Cognition. (Harvard University Press, 1994)

    1994

  50. [58]

    & Ritter, F

    Jones, G. & Ritter, F. E. Production systems and rule-based inference. Encycl. Cogn. Sci. 3, 741–747 (2003)

    2003

  51. [59]

    Anderson, J. R. et al. An integrated theory of the mind. Psychol. Rev. 111, 1036 (2004)

    2004

  52. [60]

    & King, J

    Davis, R. & King, J. An overview of production systems. Memo AM-271 (1975)

    1975

  53. [61]

    Post, E. L. Formal reductions of the general combinatorial decision problem. Am. J. Math. 65, 197–215 (1943)

    1943

  54. [62]

    & Simon, H

    Newell, A. & Simon, H. A. Human Problem Solving. vol. 104 (Prentice-Hall Englewood Cliffs, NJ, 1972)

    1972

  55. [63]

    E., Newell, A

    Laird, J. E., Newell, A. & Rosenbloom, P. S. Soar: An architecture for general intelligence. Artif. Intell. 33, 1–64 (1987)

    1987

  56. [64]

    Anderson, J. R. Rules of the Mind. (Psychology Press, 2014)

    2014

  57. [65]

    Minsky, M. L. Computation. (Prentice-Hall Englewood Cliffs, 1967)

    1967

  58. [66]

    Anderson, J. R. & Douglass, S. Tower of Hanoi: evidence for the cost of goal retrieval. J. Exp. Psychol. Learn. Mem. Cogn. 27, 1331 (2001)

    2001

  59. [67]

    Anderson, J. R. Acquisition of cognitive skill. Psychol. Rev. 89, 369–406 (1982)

    1982

  60. [68]

    R., Bothell, D., Lebiere, C

    Anderson, J. R., Bothell, D., Lebiere, C. & Matessa, M. An integrated theory of list memory. J. Mem. Lang. 38, 341–380 (1998)

    1998

  61. [69]

    Anderson, J. R. & Matessa, M. A production system theory of serial memory. Psychol. Rev. 104, 728 (1997)

    1997

  62. [70]

    Salvucci, D. D. & Taatgen, N. A. Threaded cognition: an integrated theory of concurrent multitasking. Psychol. Rev. 115, 101 (2008)

    2008

  63. [71]

    Kieras, D. E. & Meyer, D. E. An overview of the EPIC architecture for cognition and performance with application to human-computer interaction. Human–Computer Interact. 12, 391–438 (1997)

    1997

  64. [72]

    Anderson, J. R. & Lebiere, C. J. The Atomic Components of Thought. (Psychology Press, 2014)

    2014

  65. [73]

    R., Fincham, J

    Anderson, J. R., Fincham, J. M., Qin, Y. & Stocco, A. A central circuit of the mind. Trends Cogn. Sci. 12, 136–143 (2008)

    2008

  66. [74]

    & Anderson, J

    Stocco, A., Lebiere, C. & Anderson, J. R. Conditional routing of information to the cortex: a model of the basal ganglia’s role in cognitive coordination. Psychol. Rev. 117, 541 (2010)

    2010

  67. [75]

    Anderson, J. R. How Can the Human Mind Occur in the Physical Universe? (Oxford University Press, 2009)

    2009

  68. [76]

    Byrne, M. D. Unified theories of cognition. Wiley Interdiscip. Rev. Cogn. Sci. 3, 431–438 (2012)

    2012

  69. [77]

    Vaswani, A. et al. Attention is all you need. Adv. Neural Inf. Process. Syst. 30, (2017)

    2017

  70. [78]

    Elhage, N. et al. A mathematical framework for transformer circuits. Transform. Circuits Thread 1, 12 (2021)

    2021

  71. [79]

    & Berant, J

    Dar, G., Geva, M., Gupta, A. & Berant, J. Analyzing transformers in embedding space. in Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) 16124–16170 (2023)

    2023

  72. [80]

    & Pavlick, E

    Merullo, J., Eickhoff, C. & Pavlick, E. Language models implement simple word2vec-style vector arithmetic. in Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) ...

    2024

  73. [81]

    & Levy, O

    Geva, M., Schuster, R., Berant, J. & Levy, O. Transformer feed-forward layers are key- value memories. in Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing 5484–5495 (2021)

    2021

  74. [82]

    & Goldberg, Y

    Geva, M., Caciularu, A., Wang, K. & Goldberg, Y. Transformer feed-forward layers build predictions by promoting concepts in the vocabulary space. in Proceedings of the 2022 conference on empirical methods in natural language processing 30–45 (2022)

    2022

  75. [83]

    & Sun, J

    He, K., Zhang, X., Ren, S. & Sun, J. Deep residual learning for image recognition. in Proceedings of the IEEE conference on computer vision and pattern recognition 770–778 (2016)

    2016

  76. [84]

    Smolensky, P. et al. Mechanisms of symbol processing for in-context learning in transformer networks. J. Artif. Intell. Res. 84, (2025)

    2025

  77. [85]

    Olah, C. et al. Zoom in: An introduction to circuits. Distill 5, e00024-001 (2020)

    2020

  78. [86]

    & Steinhardt, J

    Wang, K., Variengien, A., Conmy, A., Shlegeris, B. & Steinhardt, J. Interpretability in the Wild: A Circuit for Indirect Object Identification in GPT-2 Small. ArXiv Prepr. ArXiv221100593 (2022)

    2022

  79. [87]

    Olsson, C. et al. In-context learning and induction heads. ArXiv Prepr. ArXiv220911895 (2022)

    2022

  80. [88]

    & Gavves, E

    Bereska, L. & Gavves, E. Mechanistic interpretability for AI safety--a review. ArXiv Prepr. ArXiv240414082 (2024)

    2024

  81. [89]

    & Linzen, T

    Hao, S. & Linzen, T. Verb conjugation in transformers is determined by linear encodings of subject number. in Findings of the Association for Computational Linguistics: EMNLP 2023 4531–4539 (2023)

    2023

  82. [90]

    & Mueller, A

    Brinkmann, J., Wendler, C., Bartelt, C. & Mueller, A. Large language models share representations of latent grammatical concepts across typologically diverse languages. in Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computat...

    2025

  83. [91]

    & Goldberg, Y

    Ravfogel, S., Prasad, G., Linzen, T. & Goldberg, Y. Counterfactual interventions reveal the causal effect of relative clause representations on agreement prediction. in Proceedings of the 25th Conference on Computational Natural Language Learning 194–209 (2021)

    2021

  84. [92]

    Marks, S. et al. Sparse feature circuits: Discovering and editing interpretable causal graphs in language models. in International Conference on Learning Representations vol. 2025 23888–23923 (2025)

    2025

  85. [93]

    & Steinhardt, J

    Feng, J. & Steinhardt, J. How do language models bind entities in context? in International Conference on Learning Representations vol. 2024 36391–36413 (2024)

    2024

  86. [94]

    & Mahowald, K

    Boguraev, S., Potts, C. & Mahowald, K. Causal Interventions Reveal Shared Structure Across English Filler–Gap Constructions. in Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing 25032–25053 (2025)

    2025

  87. [95]

    G., Fedorenko, E

    Kryvosheieva, D., de Varda, A. G., Fedorenko, E. & Tuckute, G. Different types of syntactic agreement recruit the same units within large language models. in Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) 209–227 (2026)

    2026

  88. [96]

    Ryu, S. H. & Lewis, R. L. Memory for prediction: A Transformer-based theory of sentence processing. J. Mem. Lang. 145, 104670 (2025)

    2025

  89. [97]

    & Mahowald, K

    Futrell, R. & Mahowald, K. How linguistics learned to stop worrying and love the language models. Behav. Brain Sci. 1–98 (2025)

    2025

  90. [98]

    & Bandettini, P

    Kriegeskorte, N., Mur, M. & Bandettini, P. A. Representational similarity analysis: connecting the branches of systems neuroscience. Front. Syst. Neurosci. 2, 249 (2008)

    2008

  91. [99]

    Sucholutsky, I. et al. Getting aligned on representational alignment. ArXiv Prepr. ArXiv231013018 (2023)

    2023

  92. [100]

    Schrimpf, M. et al. The neural architecture of language: Integrative modeling converges on predictive processing. Proc. Natl. Acad. Sci. 118, e2105646118 (2021)

    2021

  93. [101]

    Goldstein, A. et al. Shared computational principles for language processing in humans and deep language models. Nat. Neurosci. 25, 369–380 (2022)

    2022

  94. [102]

    Kumar, S. et al. Shared functional specialization in transformer-based language models and the human brain. Nat. Commun. 15, 5523 (2024)

    2024

  95. [103]

    & Fedorenko, E

    Tuckute, G., Kanwisher, N. & Fedorenko, E. Language in brains, minds, and machines. Annu. Rev. Neurosci. 47, 277–301 (2024)

    2024

  96. [104]

    A., Bickel, S., Mehta, A

    Mischler, G., Li, Y. A., Bickel, S., Mehta, A. D. & Mesgarani, N. Contextual feature extraction hierarchies converge in large language models and the brain. Nat. Mach. Intell. 6, 1467–1477 (2024)

    2024

  97. [105]

    Yamins, D. L. et al. Performance-optimized hierarchical models predict neural responses in higher visual cortex. Proc. Natl. Acad. Sci. 111, 8619–8624 (2014)

    2014

  98. [106]

    J., Yamins, D

    Kell, A. J., Yamins, D. L., Shook, E. N., Norman-Haignere, S. V. & McDermott, J. H. A task-optimized neural network replicates human auditory behavior, predicts brain responses, and reveals a cortical processing hierarchy. Neuron 98, 630–644 (2018)

    2018

  99. [107]

    & Yamins, D

    Cao, R. & Yamins, D. Explanatory models in neuroscience, Part 2: Functional intelligibility and the contravariance principle. Cogn. Syst. Res. 85, 101200 (2024)

    2024

  100. [108]

    Waddington, C. H. The Evolution of an Evolutionist. (Cornell University Press, Ithaca, NY, 1975)

    1975

  101. [109]

    Innateness and canalization

    Ariew, A. Innateness and canalization. Philos. Sci. 63, S19–S27 (1996)

    1996

  102. [110]

    Rao, R. P. N. & Ballard, D. H. Predictive coding in the visual cortex: a functional interpretation of some extra-classical receptive-field effects. Nat. Neurosci. 2, 79–87 (1999)

    1999

  103. [111]

    Wolpert, D. M. & Ghahramani, Z. Computational principles of movement neuroscience. Nat. Neurosci. 3, 1212–1217 (2000)

    2000

  104. [112]

    P., Heilbron, M

    De Lange, F. P., Heilbron, M. & Kok, P. How do expectations shape perception? Trends Cogn. Sci. 22, 764–779 (2018)

    2018

  105. [113]

    Smith, N. J. & Levy, R. The effect of word predictability on reading time is logarithmic. Cognition 128, 302–319 (2013)

    2013

  106. [114]

    Pickering, M. J. & Gambi, C. Predicting while comprehending language: A theory and review. Psychol. Bull. 144, 1002 (2018)

    2018

  107. [115]

    Predictions: a universal principle in the operation of the human brain

    Bar, M. Predictions: a universal principle in the operation of the human brain. Philos. Trans. R. Soc. B Biol. Sci. 364, 1181 (2009)

    2009

  108. [116]

    The free-energy principle: a unified brain theory? Nat

    Friston, K. The free-energy principle: a unified brain theory? Nat. Rev. Neurosci. 11, 127– 138 (2010)

    2010

  109. [117]

    Whatever next? Predictive brains, situated agents, and the future of cognitive science

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

    2013

  110. [118]

    Kalman, R. E. A new approach to linear filtering and prediction problems. Trans ASME D 82, 35–44 (1960)

    1960

  111. [119]

    F., Della Pietra, V

    Brown, P. F., Della Pietra, V. J., Desouza, P. V., Lai, J. C. & Mercer, R. L. Class-based n- gram models of natural language. Comput. Linguist. 18, 467–480 (1992)

    1992

  112. [120]

    & Jauvin, C

    Bengio, Y., Ducharme, R., Vincent, P. & Jauvin, C. A neural probabilistic language model. J. Mach. Learn. Res. 3, 1137–1155 (2003)

    2003

  113. [121]

    & Sutskever, I

    Radford, A., Narasimhan, K., Salimans, T. & Sutskever, I. Improving language understanding by generative pre-training. (2018)

    2018

  114. [122]

    Warstadt, A. et al. Findings of the BabyLM challenge: Sample-efficient pretraining on developmentally plausible corpora. in Proceedings of the babylm challenge at the 27th conference on computational natural language learning 1–34 (2023)

    2023

  115. [123]

    Gilkerson, J. et al. Mapping the early language environment using all-day recordings and automated analysis. Am. J. Speech Lang. Pathol. 26, 248–265 (2017)

    2017

  116. [124]

    An Essay Concerning Human Understanding

    Locke, J. An Essay Concerning Human Understanding. (Clarendon Press, Oxford, 1690)

  117. [125]

    & Goldt, S

    Ingrosso, A. & Goldt, S. Data-driven emergence of convolutional structure in neural networks. Proc. Natl. Acad. Sci. 119, e2201854119 (2022)

    2022

  118. [126]

    & Wyart, M

    Favero, A., Cagnetta, F. & Wyart, M. Locality defeats the curse of dimensionality in convolutional teacher-student scenarios. Adv. Neural Inf. Process. Syst. 34, 9456–9467 (2021)

    2021

  119. [127]

    & Lillicrap, T

    Hafner, D., Pasukonis, J., Ba, J. & Lillicrap, T. Mastering diverse control tasks through world models. Nature 640, 647–653 (2025)

    2025

  120. [128]

    Pathak, D., Agrawal, P., Efros, A. A. & Darrell, T. Curiosity-driven exploration by self- supervised prediction. in International conference on machine learning 2778–2787 (PMLR, 2017)

    2017

  121. [129]

    K., Wang, W., Orhan, A

    Vong, W. K., Wang, W., Orhan, A. E. & Lake, B. M. Grounded language acquisition through the eyes and ears of a single child. Science 383, 504–511 (2024)

    2024

  122. [130]

    Sutton, R. S. & Barto, A. G. Reinforcement Learning: An Introduction. (A Bradford Book, 1998)

    1998

  123. [131]

    Daw, N. D. & Doya, K. The computational neurobiology of learning and reward. Curr. Opin. Neurobiol. 16, 199–204 (2006)

    2006

  124. [132]

    & Montague, P

    Rangel, A., Camerer, C. & Montague, P. R. A framework for studying the neurobiology of value-based decision making. Nat Rev Neurosci 9, 545–56 (2008)

    2008

  125. [133]

    Reinforcement learning in the brain

    Niv, Y. Reinforcement learning in the brain. J. Math. Psychol. 53, 139–154 (2009)

    2009

  126. [134]

    The Valuationist Model of Human Agent Architecture

    Sripada, C. The Valuationist Model of Human Agent Architecture. Philos. Psychol. 1–30 (2025) doi:doi.org/10.1080/09515089.2025.2485323

    2025

  127. [135]

    & Ruppin, E

    Joel, D., Niv, Y. & Ruppin, E. Actor–critic models of the basal ganglia: New anatomical and computational perspectives. Neural Netw. 15, 535–547 (2002)

    2002

  128. [136]

    Dolan, R. J. & Dayan, P. Goals and Habits in the Brain. Neuron 80, 312–325 (2013)

    2013

  129. [137]

    Shenhav, A., Botvinick, M. M. & Cohen, J. D. The expected value of control: an integrative theory of anterior cingulate cortex function. Neuron 79, 217–240 (2013)

    2013

  130. [138]

    Callaway, F. et al. Rational use of cognitive resources in human planning. Nat. Hum. Behav. 6, 1112–1125 (2022)

    2022

  131. [139]

    & Griffiths, T

    Lieder, F. & Griffiths, T. L. Strategy selection as rational metareasoning. Psychol. Rev. 124, 762 (2017)

    2017

  132. [140]

    & Gutkin, B

    Keramati, M. & Gutkin, B. Homeostatic reinforcement learning for integrating reward collection and physiological stability. Elife 3, e04811 (2014)

    2014

  133. [141]

    & Summerfield, C

    Juechems, K. & Summerfield, C. Where does value come from? Trends Cogn. Sci. 23, 836–850 (2019)

    2019

  134. [142]

    & Eldar, E

    Emanuel, A. & Eldar, E. Emotions as computations. Neurosci. Biobehav. Rev. 104977 (2022)

    2022

  135. [143]

    B., Dolan, R

    Eldar, E., Rutledge, R. B., Dolan, R. J. & Niv, Y. Mood as representation of momentum. Trends Cogn. Sci. 20, 15–24 (2016)

    2016

  136. [144]

    D., Niv, Y

    Daw, N. D., Niv, Y. & Dayan, P. Uncertainty-based competition between prefrontal and dorsolateral striatal systems for behavioral control. Nat. Neurosci. 8, 1704–1711 (2005)

    2005

  137. [145]

    & Piray, P

    Keramati, M., Dezfouli, A. & Piray, P. Speed/accuracy trade-off between the habitual and the goal-directed processes. PLoS Comput. Biol. 7, e1002055 (2011)

    2011

  138. [146]

    Kool, W., Gershman, S. J. & Cushman, F. A. Cost-benefit arbitration between multiple reinforcement-learning systems. Psychol. Sci. 28, 1321–1333 (2017)

    2017

  139. [147]

    The case for value as a common currency in decision-making and intersystem competition

    Sripada, C. The case for value as a common currency in decision-making and intersystem competition. Front. Cogn. 5, 1767189 (2026)

    2026

  140. [148]

    Ouyang, L. et al. Training language models to follow instructions with human feedback. Adv. Neural Inf. Process. Syst. 35, 27730–27744 (2022)

    2022

  141. [149]

    OpenAI https://openai.com/index/learning-to-reason-with- llms/ (2024)

    Learning to reason with LLMs. OpenAI https://openai.com/index/learning-to-reason-with- llms/ (2024)

    2024

  142. [150]

    Shao, Z. et al. DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models. ArXiv Prepr. ArXiv240203300 (2024)

    2024

  143. [151]

    Yan, S. et al. Memory-r1: Enhancing large language model agents to manage and utilize memories via reinforcement learning. in Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) 12805–12825 (2026)

    2026

  144. [152]

    Team, K. et al. Kimi K2. 5: Visual Agentic Intelligence. ArXiv Prepr. ArXiv260202276 (2026)

    2026

  145. [153]

    & Lockhart, E

    Luong, T. & Lockhart, E. Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad. Google DeepMind Blog https://deepmind.google/blog/advanced-version-of-gemini-with-deep-think-officially- achieves-gold-medal-...

    2025

  146. [154]

    & Singh, J

    Cai, K. & Singh, J. Google clinches milestone gold at global math competition, while OpenAI also claims win. Reuters (2025)

    2025

  147. [155]

    Kwa, T. et al. Measuring AI ability to complete long tasks. ArXiv Prepr. ArXiv250314499 (2025)

    2025

  148. [156]

    & Krueger, D

    Skalse, J., Howe, N., Krasheninnikov, D. & Krueger, D. Defining and characterizing reward gaming. Adv. Neural Inf. Process. Syst. 35, 9460–9471 (2022)

    2022

  149. [157]

    Baker, B. et al. Monitoring reasoning models for misbehavior and the risks of promoting obfuscation. ArXiv Prepr. ArXiv250311926 (2025)

    2025

  150. [158]

    & Sutton, R

    Silver, D. & Sutton, R. S. Welcome to the era of experience. Google AI (2025)

    2025

  151. [159]

    Chen, Z., Deng, Y., Yuan, H., Ji, K. & Gu, Q. Self-play fine-tuning converts weak language models to strong language models. ArXiv Prepr. ArXiv240101335 (2024)

    2024

  152. [160]

    Packer, C. et al. MemGPT: Towards LLMs as Operating Systems. ArXiv Prepr. ArXiv231008560 (2023)

    2023

  153. [161]

    Schick, T. et al. Toolformer: Language models can teach themselves to use tools. Adv. Neural Inf. Process. Syst. 36, 68539–68551 (2023)

    2023

  154. [162]

    R., Yao, S., Narasimhan, K

    Sumers, T. R., Yao, S., Narasimhan, K. & Griffiths, T. L. Cognitive architectures for language agents. ArXiv Prepr. ArXiv230902427 (2023)

    2023

  155. [163]

    Wang, H. et al. Emergent hierarchical reasoning in llms through reinforcement learning. ArXiv Prepr. ArXiv250903646 (2025)

    2025

  156. [164]

    He, Z. et al. Deepmath-103k: A large-scale, challenging, decontaminated, and verifiable mathematical dataset for advancing reasoning. ArXiv Prepr. ArXiv250411456 (2025)

    2025

  157. [165]

    L., Barto, A

    Singh, S., Lewis, R. L., Barto, A. G. & Sorg, J. Intrinsically motivated reinforcement learning: An evolutionary perspective. IEEE Trans. Auton. Ment. Dev. 2, 70–82 (2010)

    2010

  158. [166]

    Silver, D. et al. A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play. Science 362, 1140–1144 (2018)

    2018

  159. [167]

    Sorg, J., Lewis, R. L. & Singh, S. Reward design via online gradient ascent. Adv. Neural Inf. Process. Syst. 23, (2010)

    2010

  160. [168]

    Gunjal, A. et al. Rubrics as rewards: Reinforcement learning beyond verifiable domains. ArXiv Prepr. ArXiv250717746 (2025)

    2025

  161. [169]

    Yu, R. et al. Reward models in deep reinforcement learning: A survey. ArXiv Prepr. ArXiv250615421 (2025)

    2025

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.