arxiv: 2604.03201 · v1 · submitted 2026-04-03 · 💻 cs.AI

Recognition: 2 theorem links

· Lean Theorem

Coupled Control, Structured Memory, and Verifiable Action in Agentic AI (SCRAT -- Stochastic Control with Retrieval and Auditable Trajectories): A Comparative Perspective from Squirrel Locomotion and Scatter-Hoarding

Authors on Pith no claims yet

Pith reviewed 2026-05-13 19:51 UTC · model grok-4.3

classification 💻 cs.AI

keywords agentic AIpartially observed controlstructured episodic memoryverifiable actionscatter-hoardinginference ladderaudience-sensitive caching

0 comments

The pith

Squirrel locomotion and scatter-hoarding supply a model for coupling fast control, structured memory, and verifiable action inside one agentic AI system under partial observability.

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

The paper establishes that existing AI research treats control, memory, and verification as separate problems, yet real agents must solve them together under delay and hidden information. It uses fox, eastern gray, and red squirrel behaviors as a single-organism case where arboreal movement, food caching, and audience-sensitive hiding already integrate these demands. From that comparison the authors derive a minimal hierarchical partially observed control model that includes latent dynamics, episodic memory organized for future use, observer belief states, and delayed verifier signals. The model yields three concrete hypotheses on robustness, retrieval under conflict, and reduced silent failure when verification sits inside the control loop. The work frames this as a falsifiable benchmark agenda rather than a completed architecture.

Core claim

A minimal hierarchical partially observed control model with latent dynamics, structured episodic memory, observer-belief state, option-level actions, and delayed verifier signals can be inferred from squirrel arboreal locomotion, scatter-hoarding, and audience-sensitive caching, and this coupling produces testable improvements in robustness, delayed retrieval, and verification under asymmetric information.

What carries the argument

The explicit inference ladder (empirical observation to minimal computational inference to AI design conjecture) together with the hierarchical partially observed control model that places structured episodic memory and observer-belief states inside the action-verifier loop.

If this is right

Fast local feedback plus predictive compensation improves robustness when hidden dynamics shift.
Memory organized for future control improves delayed retrieval when cues conflict or load increases.
Placing verifiers and observer models inside the action-memory loop reduces silent failure and information leakage.
Role-differentiated proposer, executor, checker, and adversary components can lower correlated error under asymmetric information.

Where Pith is reading between the lines

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

The same coupling pattern could be tested directly in simulated environments that combine locomotion, caching, and observation tasks to generate quantitative benchmarks.
Extending the observer-belief state to multi-agent settings may address verification problems that arise when several AI systems interact under partial information.
The inference ladder itself could be applied to other biological systems that solve similar triads of control, memory, and verification, such as certain birds or primates, to generate further design conjectures.

Load-bearing premise

Biological mechanisms observed in squirrels can be mapped onto AI architectures through an explicit inference ladder without quantitative validation or loss of fidelity.

What would settle it

Build two agent systems, one using the proposed integrated model and one using separate control and memory modules, and measure whether the integrated version shows measurably lower silent failure rates and better delayed retrieval on a task with hidden dynamics shifts and strategic observers; if the difference disappears or reverses, the coupling claim does not hold.

Figures

Figures reproduced from arXiv: 2604.03201 by Christophe Kolb, Maximiliano Armesto.

**Figure 1.** Figure 1: Conceptual overview of the coupled control–memory–verification problem studied in this paper. Squirrel behavior illustrates three tightly linked components: control under uncertainty, episodic memory for future action, and embedded verification under observation. The resulting loop motivates the hypothesis that agentic systems must integrate fast feedback control, structured memory, and in-loop verificatio… view at source ↗

**Figure 2.** Figure 2: Comparative inference ladder. Each upward step from observation to engineering claim increases abstraction and therefore strengthens the need for explicit benchmarking and ablation [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: Minimal architecture implied by the comparative thesis. The proposal is not a claim of literal squirrel mechanism; it is an engineering decomposition that makes H1-H3 and C1 benchmarkable. 5. Testable design hypotheses for agentic AI The formalization earns its place only if it yields benchmark-relevant distinctions. We therefore state the main claims as hypotheses rather than findings. The first three are… view at source ↗

**Figure 4.** Figure 4: First-release coverage and staged issue load in the isolated-agent baseline and the memory-augmented review-integrated configuration from the companion software-delivery benchmark [31]. The middle bar in the issue-load panel reports issue burden before the pull-request boundary, so it isolates the structured-memory effect before the additional defect containment introduced by review; the rightmost bar repo… view at source ↗

read the original abstract

Agentic AI is increasingly judged not by fluent output alone but by whether it can act, remember, and verify under partial observability, delay, and strategic observation. Existing research often studies these demands separately: robotics emphasizes control, retrieval systems emphasize memory, and alignment or assurance work emphasizes checking and oversight. This article argues that squirrel ecology offers a sharp comparative case because arboreal locomotion, scatter-hoarding, and audience-sensitive caching couple all three demands in one organism. We synthesize evidence from fox, eastern gray, and, in one field comparison, red squirrels, and impose an explicit inference ladder: empirical observation, minimal computational inference, and AI design conjecture. We introduce a minimal hierarchical partially observed control model with latent dynamics, structured episodic memory, observer-belief state, option-level actions, and delayed verifier signals. This motivates three hypotheses: (H1) fast local feedback plus predictive compensation improves robustness under hidden dynamics shifts; (H2) memory organized for future control improves delayed retrieval under cue conflict and load; and (H3) verifiers and observer models inside the action-memory loop reduce silent failure and information leakage while remaining vulnerable to misspecification. A downstream conjecture is that role-differentiated proposer/executor/checker/adversary systems may reduce correlated error under asymmetric information and verification burden. The contribution is a comparative perspective and benchmark agenda: a disciplined program of falsifiable claims about the coupling of control, memory, and verifiable action.

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 uses squirrel behaviors to frame integrated control, memory, and verification in AI agents but provides only qualitative hypotheses without formal models or tests.

read the letter

The one thing to know is that this paper uses squirrel ecology as a comparative case for coupling control, structured memory, and verifiable action in agentic AI, but it stays at the level of untested hypotheses and does not deliver a working model or new results. The SCRAT framing is presented as a minimal hierarchical POMDP-like structure with latent dynamics, episodic memory, observer-belief states, and delayed verifiers, motivated by an inference ladder from field observations on locomotion and caching to AI design conjectures. Three hypotheses follow on robustness under hidden shifts, memory organized for future control, and internal verifiers reducing silent failures. A downstream idea about role-differentiated proposer/executor/checker systems is also sketched. This synthesis of existing demands under one biological example is the main contribution, and the structure of the argument is clear enough to follow. The hypotheses are stated plainly and could serve as prompts for later work. The soft spots are in the execution. No equations, parameterizations, or even a simple simulation appear that would take concrete squirrel observations and produce the listed AI components, so the mapping cannot be checked for fidelity or missing constraints. The circularity risk is present because the coupling is assumed in the starting analogy rather than derived or measured separately. Without any comparison to prior RL or memory-augmented agent literature, it is hard to judge what is gained. This paper is for readers who work on agent architectures and are open to bio-inspired framing or workshop-style position pieces. It might prompt discussion in that niche but will not supply ready methods or benchmarks. I would send it to peer review as a perspective paper because the core analogy is coherent and the hypotheses are falsifiable in principle, though referees will need to see more concrete modeling or initial tests before it stands alone.

Referee Report

3 major / 2 minor

Summary. The manuscript presents a comparative perspective arguing that squirrel behaviors involving arboreal locomotion, scatter-hoarding, and audience-sensitive caching provide an integrated model for coupling control, memory, and verifiable action in agentic AI systems. It synthesizes biological evidence and proposes the SCRAT (Stochastic Control with Retrieval and Auditable Trajectories) model as a minimal hierarchical partially observed Markov decision process (POMDP) incorporating latent dynamics, structured episodic memory, observer-belief states, option-level actions, and delayed verifier signals. This framework motivates three specific hypotheses (H1-H3) on robustness under hidden dynamics, memory for delayed retrieval, and verification to reduce failures, along with a conjecture on role-differentiated agent systems. The main contribution is framed as establishing a benchmark agenda with falsifiable claims.

Significance. Should the inference from squirrel ecology to AI architectures be successfully quantified and tested, this work could significantly influence the design of agentic AI by providing a biologically inspired, integrated approach to handling partial observability, delays, and strategic interactions. The focus on verifiable action and structured memory addresses key challenges in current AI research. The proposal of falsifiable hypotheses and a benchmark agenda represents a constructive step toward empirical validation in this interdisciplinary space.

major comments (3)

[Section introducing the SCRAT model and hypotheses] The central hypotheses (H1: fast local feedback plus predictive compensation; H2: memory organized for future control; H3: verifiers and observer models) are presented without any supporting quantitative data, derivations, or error analysis from the squirrel observations, making the claims rest on qualitative analogy alone.
[Description of the inference ladder] The 'explicit inference ladder' from empirical observation to minimal computational inference to AI design conjecture is described in the abstract and introduction but the computational inference step lacks any equations, mappings, or intermediate models that would connect specific squirrel field data (e.g., caching rates) to the POMDP components.
[Model formalization] Although the SCRAT model is introduced as a minimal hierarchical POMDP-like structure with components such as latent dynamics and episodic memory, no formal equations, state definitions, or transition functions are provided, preventing assessment of how the coupling is achieved.

minor comments (2)

[Abstract] The abstract is quite long and dense; consider condensing the description of the model components for better readability.
[Terminology] The acronym SCRAT is introduced but its expansion (Stochastic Control with Retrieval and Auditable Trajectories) could be clarified earlier in the text for readers unfamiliar with the domain.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments and for recognizing the potential significance of this interdisciplinary perspective. The manuscript is framed as a comparative benchmark agenda synthesizing existing evidence to motivate falsifiable hypotheses, rather than a data-driven empirical study or fully formalized technical model. We address each major comment below and indicate planned revisions.

read point-by-point responses

Referee: [Section introducing the SCRAT model and hypotheses] The central hypotheses (H1: fast local feedback plus predictive compensation; H2: memory organized for future control; H3: verifiers and observer models) are presented without any supporting quantitative data, derivations, or error analysis from the squirrel observations, making the claims rest on qualitative analogy alone.

Authors: The hypotheses are motivated by qualitative synthesis of published squirrel ecology literature rather than new quantitative data or derivations from field observations. This is intentional for a perspective paper whose primary contribution is establishing a benchmark agenda with falsifiable claims for future work. We will add a new subsection in the discussion that outlines concrete experimental designs and metrics for testing each hypothesis in AI systems (e.g., robustness under dynamics shifts for H1, retrieval accuracy under cue conflict for H2, and failure/leakage rates for H3). revision: partial
Referee: [Description of the inference ladder] The 'explicit inference ladder' from empirical observation to minimal computational inference to AI design conjecture is described in the abstract and introduction but the computational inference step lacks any equations, mappings, or intermediate models that would connect specific squirrel field data (e.g., caching rates) to the POMDP components.

Authors: We agree the computational inference step can be clarified. While the paper analyzes no new field data, we will revise the introduction to include a table of high-level mappings from key biological observations (e.g., audience-sensitive caching rates) to POMDP elements such as observer-belief states and delayed verifier signals. This will make the ladder more explicit without adding unsubstantiated equations or quantitative derivations. revision: yes
Referee: [Model formalization] Although the SCRAT model is introduced as a minimal hierarchical POMDP-like structure with components such as latent dynamics and episodic memory, no formal equations, state definitions, or transition functions are provided, preventing assessment of how the coupling is achieved.

Authors: The SCRAT model is intentionally described at a minimal conceptual level to emphasize integration across control, memory, and verification. To address the concern, we will add a structured outline (with pseudocode for the overall loop) defining the components and their high-level interactions in a new subsection, while preserving the perspective framing and avoiding full formalization that would exceed the manuscript's scope. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is analogical perspective without self-referential reduction

full rationale

The paper states an inference ladder from squirrel observations to a hierarchical POMDP-like model and three hypotheses, but supplies no equations, parameter fits, or derivations that would allow any component to reduce to the biological inputs by construction. The model is introduced as motivated by the observations rather than derived from them mathematically, and the hypotheses are framed as downstream conjectures rather than predictions forced by fitted inputs or self-definitions. No self-citations, uniqueness theorems, or ansatzes are invoked in a load-bearing way within the provided text. The contribution is explicitly a comparative perspective and benchmark agenda, which remains self-contained against external benchmarks without circular closure.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the untested premise that squirrel behaviors supply a faithful template for AI coupling; no free parameters are fitted, but the model itself is an invented construct whose validity is assumed rather than derived.

axioms (1)

domain assumption Squirrel locomotion, caching, and audience sensitivity can be faithfully abstracted into a single hierarchical partially observed control model with latent dynamics, episodic memory, and delayed verifiers
Invoked when moving from empirical observation to AI design conjecture in the inference ladder

invented entities (1)

SCRAT model no independent evidence
purpose: To couple control, memory, and verifiable action in one architecture
Postulated as a minimal hierarchical controller without external falsification criteria or independent evidence

pith-pipeline@v0.9.0 · 5584 in / 1341 out tokens · 36471 ms · 2026-05-13T19:51:28.273266+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

minimal hierarchical partially observed control model with latent dynamics, structured episodic memory, observer-belief state, option-level actions, and delayed verifier signals... SCRAT (Stochastic Control with Retrieval and Auditable Trajectories)
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

fast local feedback plus predictive compensation... memory organized for future control... verifiers and observer models inside the action-memory loop

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Toward a Science of Intent: Closure Gaps and Delegation Envelopes for Open-World AI Agents
cs.AI 2026-04 unverdicted novelty 5.0

Intent compilation turns vague human goals into verifiable artifacts, using closure-gap vectors and delegation envelopes to separate open-world agent challenges from closed-world solvers and to benchmark closure fixes...

Reference graph

Works this paper leans on

31 extracted references · 31 canonical work pages · cited by 1 Pith paper · 2 internal anchors

[1]

Acrobatic squirrels learn to leap and land on tree branches without falling,

N. H. Hunt, J. Jinn, L. F. Jacobs, and R. J. Full, “Acrobatic squirrels learn to leap and land on tree branches without falling,”Science, vol. 373, no. 6555, pp. 697–700, 2021, doi: 10.1126/science.abe5753

work page doi:10.1126/science.abe5753 2021
[2]

Grey squirrels remember the locations of buried nuts,

L. F. Jacobs and E. R. Liman, “Grey squirrels remember the locations of buried nuts,”Animal Behaviour, vol. 41, no. 1, pp. 103–110, 1991, doi: 10.1016/S0003-3472(05)80506-8

work page doi:10.1016/s0003-3472(05)80506-8 1991
[3]

Field experiments on duration and precision of grey and red squirrel spatial memory,

I. M. V . Macdonald, “Field experiments on duration and precision of grey and red squirrel spatial memory,” Animal Behaviour, vol. 54, no. 4, pp. 879–891, 1997, doi: 10.1006/anbe.1996.0528

work page doi:10.1006/anbe.1996.0528 1997
[4]

Caching for where and what: evidence for a mnemonic strategy in a scatter- hoarder,

M. M. Delgado and L. F. Jacobs, “Caching for where and what: evidence for a mnemonic strategy in a scatter- hoarder,”Royal Society Open Science, vol. 4, no. 9, Art. no. 170958, 2017, doi: 10.1098/rsos.170958. 13

work page doi:10.1098/rsos.170958 2017
[5]

Fox squirrels match food assessment and cache effort to value and scarcity,

M. M. Delgado, M. Nicholas, D. J. Petrie, and L. F. Jacobs, “Fox squirrels match food assessment and cache effort to value and scarcity,”PLOS ONE, vol. 9, no. 3, Art. no. e92892, 2014, doi: 10.1371/journal.pone.0092892

work page doi:10.1371/journal.pone.0092892 2014
[6]

Audience effects on food caching in grey squirrels (Sciurus carolinensis): evidence for pilferage avoidance strategies,

L. A. Leaver, L. Hopewell, C. Caldwell, and L. Mallarky, “Audience effects on food caching in grey squirrels (Sciurus carolinensis): evidence for pilferage avoidance strategies,”Animal Cognition, vol. 10, no. 1, pp. 23–27, 2007, doi: 10.1007/s10071-006-0026-7

work page doi:10.1007/s10071-006-0026-7 2007
[7]

The socioeconomics of food hoarding in wild squirrels,

A. N. Robin and L. F. Jacobs, “The socioeconomics of food hoarding in wild squirrels,”Current Opinion in Behavioral Sciences, vol. 45, Art. no. 101139, 2022, doi: 10.1016/j.cobeha.2022.101139

work page doi:10.1016/j.cobeha.2022.101139 2022
[8]

The functional organization and cortical connections of motor cortex in squirrels,

D. F. Cooke, J. Padberg, T. Zahner, and L. Krubitzer, “The functional organization and cortical connections of motor cortex in squirrels,”Cerebral Cortex, vol. 22, no. 9, pp. 1959–1978, 2012, doi: 10.1093/cercor/bhr228

work page doi:10.1093/cercor/bhr228 1959
[9]

Sex differences, but no seasonal variations in the hippocampus of food-caching squirrels: a stereological study,

P. Lavenex, M. A. Steele, and L. F. Jacobs, “Sex differences, but no seasonal variations in the hippocampus of food-caching squirrels: a stereological study,”Journal of Comparative Neurology, vol. 425, no. 1, pp. 152–166, 2000

work page 2000
[10]

Optimal feedback control as a theory of motor coordination,

E. Todorov and M. I. Jordan, “Optimal feedback control as a theory of motor coordination,”Nature Neuroscience, vol. 5, no. 11, pp. 1226–1235, 2002, doi: 10.1038/nn963

work page doi:10.1038/nn963 2002
[11]

Internal models in the cerebellum,

D. M. Wolpert, R. C. Miall, and M. Kawato, “Internal models in the cerebellum,”Trends in Cognitive Sciences, vol. 2, no. 9, pp. 338–347, 1998, doi: 10.1016/S1364-6613(98)01221-2

work page doi:10.1016/s1364-6613(98)01221-2 1998
[12]

The hippocampus as a predictive map,

K. L. Stachenfeld, M. M. Botvinick, and S. J. Gershman, “The hippocampus as a predictive map,”Nature Neuroscience, vol. 20, no. 11, pp. 1643–1653, 2017, doi: 10.1038/nn.4650

work page doi:10.1038/nn.4650 2017
[13]

Littman, and Anthony R

L. P. Kaelbling, M. L. Littman, and A. R. Cassandra, “Planning and acting in partially observable stochastic domains,”Artificial Intelligence, vol. 101, nos. 1–2, pp. 99–134, 1998, doi: 10.1016/S0004-3702(98)00023-X

work page doi:10.1016/s0004-3702(98)00023-x 1998
[14]

Between MDPs and semi-MDPs: a framework for temporal abstraction in reinforcement learning,

R. S. Sutton, D. Precup, and S. Singh, “Between MDPs and semi-MDPs: a framework for temporal abstraction in reinforcement learning,”Artificial Intelligence, vol. 112, nos. 1–2, pp. 181–211, 1999, doi: 10.1016/S0004- 3702(99)00052-1

work page doi:10.1016/s0004- 1999
[15]

Dyna, an integrated architecture for learning, planning, and reacting,

R. S. Sutton, “Dyna, an integrated architecture for learning, planning, and reacting,”SIGART Bulletin, vol. 2, no. 4, pp. 160–163, 1991, doi: 10.1145/122344.122377

work page doi:10.1145/122344.122377 1991
[16]

Neural episodic control,

A. Pritzelet al., “Neural episodic control,” inProceedings of the 34th International Conference on Machine Learning, PMLR 70, 2017, pp. 2827–2836

work page 2017
[17]

A brief account of runtime verification,

M. Leucker and C. Schallhart, “A brief account of runtime verification,”Journal of Logic and Algebraic Program- ming, vol. 78, no. 5, pp. 293–303, 2009, doi: 10.1016/j.jlap.2008.08.004

work page doi:10.1016/j.jlap.2008.08.004 2009
[18]

Toward Verified Artificial Intelligence,

S. A. Seshia, D. Sadigh, and S. S. Sastry, “Toward Verified Artificial Intelligence,”Communications of the ACM, vol. 65, no. 7, pp. 46–55, 2022, doi: 10.1145/3503914

work page doi:10.1145/3503914 2022
[19]

Nature , year=

D. Hafner, J. Pasukonis, J. Ba, and T. Lillicrap, “Mastering diverse control tasks through world models,”Nature, vol. 640, pp. 647–653, 2025, doi: 10.1038/s41586-025-08744-2

work page doi:10.1038/s41586-025-08744-2 2025
[20]

World Models

D. Ha and J. Schmidhuber, “World Models,” arXiv:1803.10122, 2018

work page internal anchor Pith review Pith/arXiv arXiv 2018
[21]

AI safety via debate

G. Irving, P. Christiano, and D. Amodei, “AI safety via debate,” arXiv:1805.00899, 2018

work page internal anchor Pith review arXiv 2018
[22]

Toward trustworthy ai development: mechanisms for supporting verifiable claims.arXiv preprint arXiv:2004.07213, 2020

M. Brundageet al., “Toward trustworthy AI development: mechanisms for supporting verifiable claims,” arXiv:2004.07213, 2020

work page arXiv 2004
[23]

Minsky,The Society of Mind

M. Minsky,The Society of Mind. New York, NY , USA: Simon and Schuster, 1986

work page 1986
[24]

M. E. Hasselmo,How We Remember: Brain Mechanisms of Episodic Memory. Cambridge, MA, USA: The MIT Press, 2011

work page 2011
[25]

AI Agents as Universal Task Solvers: It’s All About Time,

A. Achille and S. Soatto, “AI Agents as Universal Task Solvers: It’s All About Time,” arXiv:2510.12066, 2026

work page arXiv 2026
[26]

Discovering neural nets with low Kolmogorov complexity and high generalization capability,

J. Schmidhuber, “Discovering neural nets with low Kolmogorov complexity and high generalization capability,” Neural Networks, vol. 10, no. 5, pp. 857–873, 1997. 14

work page 1997
[27]

POWERPLAY: Training an increasingly general problem solver by continually searching for the simplest still unsolvable problem,

J. Schmidhuber, “POWERPLAY: Training an increasingly general problem solver by continually searching for the simplest still unsolvable problem,” arXiv preprint arXiv:1112.5309, 2013

work page arXiv 2013
[28]

Shifting inductive bias with success-story algorithm, adaptive Levin search, and incremental self-improvement,

J. Schmidhuber, J. Zhao, and M. Wiering, “Shifting inductive bias with success-story algorithm, adaptive Levin search, and incremental self-improvement,”Machine Learning, vol. 28, pp. 105–130, 1997

work page 1997
[29]

The speed prior: a new simplicity measure yielding near-optimal computable predictions,

J. Schmidhuber, “The speed prior: a new simplicity measure yielding near-optimal computable predictions,” in Proceedings of the 15th Annual Conference on Computational Learning Theory, 2002, pp. 216–228

work page 2002
[30]

On Learning to Think: Algorithmic Information Theory for Novel Combinations of Reinforcement Learning Controllers and Recurrent Neural World Models

J. Schmidhuber, “On learning to think: Algorithmic information theory for novel combinations of reinforcement learning controllers and recurrent neural world models,” arXiv preprint arXiv:1511.09249, 2015

work page Pith review arXiv 2015
[31]

Orchestrating Human-AI Software Delivery: A Retrospective Longitudinal Field Study of Three Software Modernization Programs,

M. Armesto, and C. Kolb, “Orchestrating Human-AI Software Delivery: A Retrospective Longitudinal Field Study of Three Software Modernization Programs,” manuscript, 2026. 15

work page 2026