Pith. sign in

REVIEW 4 major objections 5 minor 60 references

Model-Free RL Agents Demonstrate System 1-Like Intentionality

T0 review · 4 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read This paper argues that model-free reinforcement-learning agents can act with intent even though they never plan or foresee consequences, and that this intent must be judged in the context of the environment in which the policy was learned.

desk verdict A useful interdisciplinary argument about context-dependent imputation of intent, overpackaged as a demonstration of agent-internal intentionality. read the letter →

arxiv 2501.18299 v1 pith:VFYMBYO6 submitted 2025-01-30 cs.AI

Hal Ashton , Matija Franklin This is my paper · ORCID
classification cs.AI
keywords reinforcementlearningmodel-freeagentsintentionalitySystem1and2folkpsychologyofintentcriminallawAIsafetypolicyfunction
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

This paper argues that a reinforcement-learning agent with no model of its world can still act with intent, as intent is understood in folk psychology and criminal law. It identifies model-free reinforcement learning with fast, reactive System 1 thinking and model-based reinforcement learning with slower, deliberative System 2 thinking, then shows that the usual ingredients of intent—desire, belief, and acting because of that belief—can be grounded in a reward function and a trained policy. The paper concludes that intent in such an agent is real but relational: it cannot be read off the agent alone, only from the agent together with the environment in which its policy was learned. This matters because intent-based legal offences such as deception and competition-law violations become applicable to reactive AI systems, while safety mechanisms need to control intent from outside.

What carries the argument

The load-bearing mechanism is the trained policy function of a model-free RL agent, understood as situated action: the current state cues stored experience, which produces an action without retracing any reasoning. Around this mechanism the paper places three definitional tools: a folk-legal definition of intent as desire plus belief plus action because of belief; the failure test as a fallback for the belief condition; and the shield from safe reinforcement learning, reinterpreted as an external System 2-like monitor that restricts the fast reactive policy. The policy does the acting, the reward function fixes the desire, the training environment fixes the belief, and the shield supplies the control the agent cannot exercise itself.

What would settle it

Retrain the same model-free agent on the same task with the reward for outcome X removed while keeping everything else fixed; if its probability of choosing the action that previously led to X does not drop, then the desire for X was not grounded in the reward function, and the paper's central claim loses its footing for that agent.

Watch

Extended reading notes

Core claim

The paper's central claim is that psychological and legal thinking does not exclude intent in a model-free agent, provided intent is considered in the wider context of the agent's policy and the environment in which it was learned. At deployment a model-free agent is a pure reactive actor: its policy is a static mapping from states to actions, with no introspection, no internal model, and no foresight. Nevertheless, the reward function that shaped that policy can supply desire for an outcome, the statistical association learned during training can supply belief that an action tends to bring that outcome about, and the policy's selection of the action can supply acting because of that belief. Where the belief condition strains, the paper falls back on the failure test: an agent intends the effects whose non-occurrence would make its action a failure, which for a reward-driven agent means the reward-bearing outcome. The conclusion is that intent is present but not self-contained—it exists only in the pairing of the policy with its training environment.

Load-bearing premise

The paper's argument depends on treating a reward function plus the statistical patterns in a trained policy as sufficient to ground the desire and belief that legal and folk concepts of intent require, even though the agent has no representation of the outcome and cannot foresee it.

Editorial extensions

If this is right

  • Model-free agents can no longer be dismissed as intent-free merely because they do not plan or foresee consequences.
  • Intent-defined offences such as deception could in principle attach to model-free systems, so regulators need a procedure for locating intent in the policy-plus-environment pair.
  • Explainability tools for such agents will need access to the training environment, not just the deployed policy, to attribute intent accurately.
  • Safety mechanisms like shields become an external way to control intent when the agent itself cannot be asked to intend differently.
  • Attributing intent requires an external causal analysis that separates intended ends from side-effects, because the model-free agent cannot supply that analysis itself.

Reading between the lines

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

  • A testable extension would be to measure whether laypeople's intent ascriptions to a model-free agent shift when they are shown its training environment; the paper's context thesis predicts they should shift.
  • The same reward-and-policy machinery could be applied to other reactive controllers, which raises the question of where learned intent ends and mere causal efficacy begins.
  • If intent is relational in this way, legal standards based on a notional reasonable actor may need to be supplemented by standards based on an agent's actual training distribution.
  • The shield analogy suggests a design principle: intent control can be externalized as a separate monitor rather than embedded in the agent's policy.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper argues that model-free reinforcement-learning agents can be described as System 1-like intentional actors. It maps the model-free versus model-based RL distinction onto Kahneman's System 1/System 2 dichotomy, reviews the contested status of dual-process theory, surveys legal definitions of intent (Duff's desire-belief account and failure test, and Ashton's goal/choice/foresight definition), and concludes that psychological and legal thinking does not exclude intent in a model-free agent when intent is considered in the wider context of the policy and its training environment. It further argues that explaining and controlling such intent requires an external model, drawing on the safe-RL shield as a System 2-like mechanism.

Significance. If the thesis were established, the paper would make a useful interdisciplinary contribution by challenging the assumption that intentionality requires planning, connecting RL architecture debates to legal responsibility, and distinguishing the question of whether intent exists from the question of how to explain it. The paper is strongest in its honest treatment of the contested dual-process literature, its use of legal definitions, and its observation that a model is needed for explanation even if not for action. The manuscript contains no experiments, proofs, or code; the only empirical reference to the authors' own studies is an unpublished forthcoming citation. As it stands, the core inference from reward function to desire is under-argued, and the title overstates the paper's own hedged conclusion.

major comments (4)
  1. [A (legal) definition of intent; Model-free does not imply intent-free] The central inference that having a reward function is sufficient for having desires or goals is not established. On the paper's own description, a deployed model-free agent selects actions by a static lookup in a policy function, which need not represent either the reward function or any particular outcome X; the reward signal operates during training, not at the moment of action. The paper concedes that 'Perhaps we are stretching belief too far' and then switches to Duff's failure test, but that test requires that non-occurrence of X 'in her eyes at the time of action' render the action a failure. A frozen policy has no internal evaluation of success or failure at decision time, so the test has no referent. Without a separate argument that the training objective is part of the agent's mental state at deployment, the desire and belief conditions of the adopted definition are not satisfied.
  2. [A (legal) definition of intent; Model-free does not imply intent-free] The paper adopts Ashton's (2022b) definition of intent, which includes a goal, a choice not to perform the action, and foreseeable causation, but it does not show that a model-free policy satisfies the choice or foresight conditions. A deterministic policy assigns exactly one action per state, so 'choice' requires either a counterfactual or stochastic account that is not supplied, and the agent's own inability to predict what its action will change contradicts the foreseeability condition. If these conditions are to be read from the designer's or observer's perspective, the paper needs to say so explicitly, because that reading changes the claim from one about the agent's mental state to one about external ascription.
  3. [Conclusion; Abstract] The paper's own conclusion is that psychological and legal thinking does not exclude intent in a model-free agent when intent is considered in the wider context of the policy and its training environment. The title and abstract claim something stronger: that such agents 'demonstrate System 1-Like Intentionality' and that 'intentionality can manifest' in their behaviour. Non-exclusion under an ascriptive, context-dependent definition is not a demonstration of an agent-level property. Moreover, the paper cites 'our own experiments (forthcoming!)' in the section on the role of intent in legal behaviour without reporting any data, so the word 'demonstrate' is not supported by the evidence presented. The authors should align the title and abstract with the hedged claim, or supply the missing empirical evidence.
  4. [Criticisms of dual-process accounts; Conclusion] After reviewing unresolved criticisms of the dual-process framework, including the absence of a unified account and the possibility that System 1 and System 2 differences are quantitative rather than qualitative, the paper concludes unqualifiedly that 'Model-free RL agents are purely System 1 thinkers.' The System 1/System 2 mapping is a useful analogy, but the paper does not argue that the contested features of System 1 are exactly the features of a policy lookup. The conclusion should be qualified to match the 'System 1-Like' language of the title.
minor comments (5)
  1. [Model and Model Free Reinforcement Learning] There are several typos and grammatical slips, for example 'different types of RL algorithms in existence but, they generally have a few things in common'; the manuscript would benefit from a careful proofread.
  2. [References] The reference list contains incomplete or nonstandard entries, such as Evans (2008) appearing as 'AR Further' and Shen et al. (2022) being described as a 'Technical Report' while cited in the text as a conference-style reference; please verify all entries against the journal style.
  3. [Model and Model Free Reinforcement Learning] The paper should define 'model-free' more precisely, since many modern agents combine model-free and model-based components, and the paper itself notes that the barrier is blurring; the argument may only apply to a restricted class of purely model-free tabular agents.
  4. [Model-free does not imply intent-free] The paper introduces Quillien and German's (2021) and Ashton's (2022b) definitions of intent but does not compare their conditions or state which definition the central argument adopts; this should be made explicit.
  5. [A (legal) definition of intent] The passage quoting Duff's failure test does not say whether the test is endorsed or merely mentioned as an alternative to the desire-belief account; the paper's subsequent argument would be clearer if it stated which test it takes as authoritative.

Circularity Check

2 steps flagged · score 6.0 of 10

The central claim is partly constructed from the paper's own definitions: desire/goal is stipulated as the reward function, and the operative algorithmic definition of intent is the first author's prior proposal.

  1. self definitional [Section 'Model-free does not imply intent-free' (paragraph beginning 'Quillien and German test...')]
    "Both definitions refer to the actor’s aims or desires in some sense. We argue these exist in RL agents because they necessarily have a reward function."

    'These' refers to aims or desires, and the only support offered is that RL agents 'necessarily have a reward function.' This makes desire, and hence the goal condition in the quoted intent definitions, equivalent to reward maximization by stipulation. Under that stipulation it is true by construction that every model-free agent has a goal, so the paper's conclusion that intent is not excluded is guaranteed by the definition it chose rather than by evidence about the agent. The paper itself hedges the same move earlier ('could be said to desire X'), confirming that the later unhedged assertion is a definitional choice promoted to a finding.

  2. self citation load bearing [Section 'Model-free does not imply intent-free' (paragraph on Ashton 2022b)]
    "Inspired from a legal perspective, (Ashton 2022b) suggests an agent intends X through actions a if it is their goal, they had a choice to not a, and a foreseeably caused X."

    The positive claim that model-free agents can intend depends on this definition of algorithmic intent, which is taken from the first author's own prior work and cited rather than derived or independently validated here. The paper concedes that Duff's belief condition is 'stretching belief too far,' so the conclusion rests on this self-authored definition together with the reward-function stipulation from the previous step. The definition's 'foreseeably caused' condition is also left unaddressed for an agent with no model. The central claim therefore reduces substantially to a self-citation chain: the author proposes the definition, the present paper applies it, and no external source fixes the content of 'goal' for a policy lookup.

full rationale

The paper contains genuine independent content: a survey of dual-process controversies, the shield mechanism for safe RL, and a modest final conclusion that psychological and legal thinking 'does not exclude' intent when the agent is considered together with its training environment. However, the key inference from reward function to desire/goal is stipulative, and the operative algorithmic definition of intent is the first author's own prior proposal. The title's claim to 'demonstrate' intentionality also outruns the hedged conclusion. No machine-checked or otherwise independent validation of the cited definition is offered, so the self-citation is load-bearing. Score 6 reflects partial circularity: the central claim is not an external empirical discovery but is substantially constituted by the definitions adopted and by a self-citation chain.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The paper contributes a conceptual argument, not an empirical or formal result, so there are no fitted parameters. The central claim rests on philosophical assumptions about the equivalence of reward functions to desires, the applicability of human intent definitions to algorithms, and the meaningfulness of the System 1/System 2 analogy. No new entities are postulated; the 'shield' is an existing RL safety technique, not an invented entity.

assumptions (4)
  • domain assumption A model-free RL agent's action selection via a static policy is equivalent to System 1 intuitive thinking.
    Invoked in Section 'Model and Model Free Reinforcement Learning' where the paper equates policy lookup with 'system 1 thinking (or thinking fast)' and uses this equivalence throughout the argument.
  • ad hoc to paper Possessing a reward function is sufficient for the agent to have desires or goals in the sense required by intent definitions.
    Stated in Section 'Model-free does not imply intent-free': 'We argue these exist in RL agents because they necessarily have a reward function.' This premise is load-bearing and unargued.
  • domain assumption Folk and legal definitions of intent (Duff's failure test, Quillien and German's causal attitude account) can be applied to non-conscious, non-biological agents.
    The paper applies these definitions to RL agents without arguing for the extension; adopted in Sections 'A (legal) definition of intent' and 'Model-free does not imply intent-free'.
  • domain assumption The dual-process distinction, despite its contested status in psychology, is a legitimate and useful analogy for describing RL algorithms.
    The paper acknowledges criticisms in Section 'Criticisms of dual-process accounts' but proceeds to use the analogy, stating the distinction 'is only as useful as the resulting advantages such a design might lend the decision maker.'

how reviews work

0 comments
Cite this review

Pith. "Pith review of Model-Free RL Agents Demonstrate System 1-Like Intentionality." pith.science (2026). https://pith.science/paper/VFYMBYO6

@misc{pith2026250118299,
  author       = {Pith},
  title        = {Pith review of: Model-Free RL Agents Demonstrate System 1-Like Intentionality},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VFYMBYO6}},
  note         = {Machine review of arXiv:2501.18299}
}
read the original abstract

This paper argues that model-free reinforcement learning (RL) agents, while lacking explicit planning mechanisms, exhibit behaviours that can be analogised to System 1 ("thinking fast") processes in human cognition. Unlike model-based RL agents, which operate akin to System 2 ("thinking slow") reasoning by leveraging internal representations for planning, model-free agents react to environmental stimuli without anticipatory modelling. We propose a novel framework linking the dichotomy of System 1 and System 2 to the distinction between model-free and model-based RL. This framing challenges the prevailing assumption that intentionality and purposeful behaviour require planning, suggesting instead that intentionality can manifest in the structured, reactive behaviours of model-free agents. By drawing on interdisciplinary insights from cognitive psychology, legal theory, and experimental jurisprudence, we explore the implications of this perspective for attributing responsibility and ensuring AI safety. These insights advocate for a broader, contextually informed interpretation of intentionality in RL systems, with implications for their ethical deployment and regulation.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 55 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Abbott, R. 2020. The Reasonable Robot : Artificial Intelligence and the Law . Cambridge University Press, 1 edition. ISBN 978-1-108-63176-1 978-1-108-47212-8 978-1-108-45902-0

    work page 2020

  4. [4]

    Abbott, R.; and Sarch, A. 2019. Punishing Artificial Intelligence : Legal Fiction or Science Fiction . UC Davis law review, 53: 323--384

    work page 2019

  5. [5]

    Alshiekh, M.; Bloem, R.; Ehlers, R.; Koenighofer, B.; Niekum, S.; and Topcu, U. 2017. Safe Reinforcement Learning via Shielding . 10

    work page 2017

  6. [6]

    Ashton, H. 2022 a . Defining and Identifying the Legal Culpability of Side Effects Using Causal Graphs . In Workshop on AI Safety 2022 co-located with the Thirty Sixth AAAI Conference on Artificial Intellige , 10

    work page 2022

  7. [7]

    Ashton, H. 2022 b . Definitions of intent suitable for algorithms. Artificial Intelligence and Law

    work page 2022

  8. [8]

    Ashton, H.; and Franklin, M. 2022. The problem of behaviour and preference manipulation in AI systems. In The AAAI -22 Workshop on Artificial Intelligence Safety ( SafeAI 2022)

    work page 2022

Show all 60 references
  1. [9]

    Balakrishnan, A.; Bouneffouf, D.; Mattei, N.; and Rossi, F. 2019. Incorporating Behavioral Constraints in Online AI Systems . Proceedings of the AAAI Conference on Artificial Intelligence, 33: 3--11

    2019

  2. [10]

    Bratman, M. 2009. Intention, Practical Rationality , and Self ‐ Governance . Ethics, 119(3): 411--443

    2009

  3. [11]

    Brooks, R. A. 1991. Intelligence without representation. Artificial Intelligence, 47(1-3): 139--159

    1991

  4. [12]

    Cialdini, R. B. 2001. The science of persuasion. Scientific American, 284(2): 76--81

    2001

  5. [13]

    B.; and Cialdini, R

    Cialdini, R. B.; and Cialdini, R. B. 2007. Influence: The psychology of persuasion, volume 55. Collins New York

    2007

  6. [14]

    B.; and Goldstein, N

    Cialdini, R. B.; and Goldstein, N. J. 2004. Social influence: Compliance and conformity. Annual review of psychology, 55(1): 591--621

    2004

  7. [15]

    Collins, A. G. E.; and Cockburn, J. 2020. Beyond dichotomies in reinforcement learning. Nature Reviews Neuroscience, 21(10): 576--586

    2020

  8. [16]

    Davidson, D. 2001. Essays on Actions and Events: Philosophical Essays Volume 1. Clarendon Press

    2001

  9. [17]

    De Neys, W.; and Glumicic, T. 2008. Conflict monitoring in dual process

    2008

  10. [18]

    Dill, B.; and Holton, R. 2014. The addict in us all. Frontiers in Psychiatry, 5: 139

    2014

  11. [19]

    Duff, R. A. 1990. Intention, Agency and Criminal Liability . Blackwell

    1990

  12. [20]

    Evans, J. S. B. 2008. AR Further. Annu. Rev. Psychol, 59: 255--78

    2008

  13. [21]

    Feltz, A. 2007. The Knobe Effect : A Brief Overview . The Journal of Mind and Behavior, 28: 265--278

    2007

  14. [22]

    T.; and Taylor, S

    Fiske, S. T.; and Taylor, S. E. 1991. Social cognition. Mcgraw-Hill Book Company

    1991

  15. [23]

    Franklin, M.; Ashton, H.; Gorman, R.; and Armstrong, S. 2022. Missing Mechanisms of Manipulation in the EU AI Act . The International FLAIRS Conference Proceedings, 35

    2022

  16. [24]

    Friston, K. 2010. The free-energy principle: a unified brain theory? Nature reviews neuroscience, 11(2): 127--138

    2010

  17. [25]

    Garcıa, J.; and Fernandez, F. 2015. A Comprehensive Survey on Safe Reinforcement Learning . Journal of Machine Learning Research, (16): 1437--1480

    2015

  18. [26]

    Gigerenzer, G. 2008. Why heuristics work. Perspectives on psychological science, 3(1): 20--29

    2008

  19. [27]

    Gigerenzer, G.; and Gaissmaier, W. 2011. Heuristic decision making. Annual review of psychology, 62(1): 451--482

    2011

  20. [28]

    M.; and Woloshin, S

    Gigerenzer, G.; Gaissmaier, W.; Kurz-Milcke, E.; Schwartz, L. M.; and Woloshin, S. 2007. Helping doctors and patients make sense of health statistics. Psychological science in the public interest, 8(2): 53--96

    2007

  21. [29]

    Gigerenzer, G.; and Selten, R. 2002. Bounded rationality: The adaptive toolbox. MIT press

    2002

  22. [30]

    Gigerenzer, G.; and Todd, P. M. 1999. Simple heuristics that make us smart. Oxford University Press, USA

    1999

  23. [31]

    Ha, D.; and Schmidhuber, J. 2018. World Models . ArXiv:1803.10122 [cs, stat]

    2018 arXiv

  24. [32]

    Hafner, D.; Lillicrap, T.; Fischer, I.; Villegas, R.; Ha, D.; Lee, H.; and Davidson, J. 2019. Learning Latent Dynamics for Planning from Pixels . Proceedings of machine learning research, 97: 2555--2565

    2019

  25. [33]

    Hasanbeig, M.; Abate, A.; and Kroening, D. 2019. Logically- Constrained Reinforcement Learning . ArXiv:1801.08099 [cs]

    2019 arXiv

  26. [34]

    C.; and Bloem, R

    Jansen, N.; Könighofer, B.; Junges, S.; Serban, A. C.; and Bloem, R. 2019. Safe Reinforcement Learning via Probabilistic Shields . arXiv:1807.06096 [cs]. ArXiv: 1807.06096

    2019 arXiv

  27. [35]

    Kahneman, D. 2003. A perspective on judgment and choice: mapping bounded rationality. American psychologist, 58(9): 697

    2003

  28. [36]

    Kahneman, D. 2011. Thinking, fast and slow. Macmillan

    2011

  29. [37]

    Kahneman, D.; and Tversky, A. 1979. Prospect theory: An analysis of decision under risk, econometrica, vol. 47

    1979

  30. [38]

    Keren, G. 2013. A tale of two systems: A scientific advance or a theoretical stone soup? Commentary on Evans & Stanovich (2013). Perspectives on Psychological Science, 8(3): 257--262

    2013

  31. [39]

    Kneer, M. 2021. Can a Robot Lie ? Exploring the Folk Concept of Lying as Applied to Artificial Agents . Cognitive Science, 45(10)

    2021

  32. [40]

    Lieberman, M. D. 2013. Social: Why our brains are wired to connect. OUP Oxford

    2013

  33. [41]

    F.; and Knobe, J

    Malle, B. F.; and Knobe, J. 1997. The Folk Concept of Intentionality . Journal of Experimental Social Psychology, 33(2): 101--121

    1997

  34. [42]

    Mishra, S. 2014. Decision-making under risk: Integrating perspectives from biology, economics, and psychology. Personality and Social Psychology Review, 18(3): 280--307

    2014

  35. [43]

    R.; Lagnado, D

    Newell, B. R.; Lagnado, D. A.; and Shanks, D. R. 2015. Straight choices: The psychology of decision making. Psychology Press

    2015

  36. [44]

    Osman, M. 2013. A case study: Dual-process theories of higher cognition—Commentary on Evans & Stanovich (2013). Perspectives on Psychological Science, 8(3): 248--252

    2013

  37. [45]

    Pachur, T.; Hertwig, R.; Gigerenzer, G.; and Brandst \"a tter, E. 2013. Testing process predictions of models of risky choice: A quantitative model comparison approach. Frontiers in Psychology, 4: 646

    2013

  38. [46]

    Quillien, T.; and German, T. C. 2021. A simple definition of ‘intentionally’. Cognition, 214: 104806

    2021

  39. [47]

    Savage, L. 1954. The Foundations of Statistics. New York, NY: Jon Wiley and Sons

    1954

  40. [48]

    Schlosser, M. 2019. Dual-system theory and the role of consciousness in intentional action. In Free will, causality, and neuroscience, 35--56. Brill

    2019

  41. [49]

    Shen, H.; Zhang, K.; Hong, M.; and Chen, T. 2022. Towards Understanding Asynchronous Advantage Actor -critic: Convergence and Linear Speedup . Technical Report arXiv:2012.15511, arXiv. ArXiv:2012.15511 [cs, math] type: article

    2022 arXiv

  42. [50]

    W.; Gawronski, B.; and Trope, Y

    Sherman, J. W.; Gawronski, B.; and Trope, Y. 2014. Dual-process theories of the social mind. Guilford Publications

    2014

  43. [51]

    Silver, D.; Schrittwieser, J.; Simonyan, K.; Antonoglou, I.; Huang, A.; Guez, A.; Hubert, T.; Baker, L.; Lai, M.; Bolton, A.; Chen, Y.; Lillicrap, T.; Hui, F.; Sifre, L.; van den Driessche, G.; Graepel, T.; and Hassabis, D. 2017. Mastering the game of Go without human knowledg...

    2017

  44. [52]

    Simester, A. P. 2021. Five functions, and Two kinds of Mens Rea . In Fundamentals of criminal law: responsibility, culpability, and wrongdoing. Oxford, United Kingdom: Oxford University Press, first edition edition. ISBN 978-0-19-885314-5. OCLC: on1242932280

    2021

  45. [53]

    Simon, H. A. 1957. A behavioral model of rational choice. Models of man, social and rational: Mathematical essays on rational human behavior in a social setting, 241--260

    1957

  46. [54]

    Simon, H. A. 1979. Rational decision making in business organizations. The American economic review, 69(4): 493--513

    1979

  47. [55]

    Explanation

    Simon, H. A. 1992. What is an “ Explanation ” of Behavior ? Psychological Science, 3(3): 150--161

    1992

  48. [56]

    Suhonen, N. 2007. Normative and descriptive theories of decision making under risk: A short review

    2007

  49. [57]

    Sutton, R. S. 1990. Integrated architectures for learning, planning, and reacting based on approximating dynamic programming. Machine learning proceedings, 216--224

    1990

  50. [58]

    Tversky, A.; and Kahneman, D. 1992. Advances in prospect theory: Cumulative representation of uncertainty. Journal of Risk and uncertainty, 5(4): 297--323

    1992

  51. [59]

    Von Neumann, J.; and Morgenstern, O. 1994. Theory of games and economic behavior. In Theory of games and economic behavior. Princeton university press

    1994

  52. [60]

    Wallis, P. 2004. Intention without representation. Philosophical Psychology, 17(2): 209--224

    2004

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.