arxiv: 2604.09567 · v1 · submitted 2026-02-18 · 💻 cs.LO · cs.AI

Recognition: 2 theorem links

· Lean Theorem

Neuro-Symbolic Strong-AI Robots with Closed Knowledge Assumption: Learning and Deductions

Zoran Majkic

Authors on Pith no claims yet

Pith reviewed 2026-05-15 20:42 UTC · model grok-4.3

classification 💻 cs.LO cs.AI

keywords strong-AI robotsBelnap bilatticeclosed knowledge assumptionneuro-symbolic learninglogic inferenceknowledge representationparadox handlingAGI deductions

0 comments

The pith

Strong-AI robots expand knowledge over time and handle paradoxes by combining neural learning with Belnap's four-valued logic under a closed knowledge assumption.

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

The paper tries to establish that AGI robots can learn and advance through input and experiences by treating unknown facts as missing knowledge that gets filled in naturally. It integrates Belnap's bilattice, where unknown is the bottom value and inconsistent is the top value, with neural networks to support causal directionality in deductions and axioms for action security. A sympathetic reader would care because this setup aims to give robots both statistical learning and logical structure so they can emulate human intelligence while managing inconsistencies like the Liar paradox without losing control.

Core claim

The central claim is that the Closed Knowledge Assumption together with logic inference on Belnap's bilattice represents the expansion of robot knowledge through learning and experiences, while the inconsistent truth-value at the top of the knowledge ordering allows strong-AI robots to support inconsistent information and paradoxes during deduction processes, and axioms guarantee controlled security about robot actions based on those inferences.

What carries the argument

Belnap's bilattice with four truth-values under the Closed Knowledge Assumption, which supplies the knowledge ordering for unknowns and inconsistencies in a neuro-symbolic robot knowledge base.

If this is right

Robot knowledge databases grow as unknown facts become known through ongoing input and experiences.
Deductions remain stable when encountering inconsistent information or paradoxes like the Liar paradox.
Axioms enforce controlled security on all actions derived from logic inferences.
Causality is emulated by the directionality built into the logic entailments.

Where Pith is reading between the lines

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

The framework could be tested by injecting paradoxes into simulated robot environments and checking whether actions stay within safe bounds.
It suggests that pure neural approaches may need an explicit logical layer to reach AGI-level handling of uncertainty and inconsistency.
The same bilattice structure might apply to other neuro-symbolic systems that must expand knowledge bases over time.

Load-bearing premise

Belnap's bilattice integrated with neural networks will automatically deliver controlled security and causality for robot actions solely through the Closed Knowledge Assumption and axioms without needing extra mechanisms or empirical checks.

What would settle it

A robot deduction that produces an unsafe action when fed a paradox such as the Liar sentence, even though the system uses the bilattice, closed assumption, and stated axioms.

Figures

Figures reproduced from arXiv: 2604.09567 by Zoran Majkic.

**Figure 1.** Figure 1: Belnap’s bilattice two natural orders: truth order, ≤, and knowledge order, ≤k, such that f ≤ ⊤ ≤ t, f ≤ ⊥ ≤ t, ⊥ ⊲⊳t ⊤ and ⊥ ≤k f ≤k ⊤, ⊥ ≤k t ≤k ⊤, f ⊲⊳k t. That is, bottom element for ≤ ordering is f, and for ≤k ordering is ⊥, and top element for ≤ ordering is t, and for ≤k ordering is ⊤. Meet and join operators under ≤ are denoted ∧ and ∨; they are natural generalizations of the usual conjunction and d… view at source ↗

read the original abstract

Knowledge representation formalisms are aimed to represent general conceptual information and are typically used in the construction of the knowledge base of reasoning agent. A knowledge base can be thought of as representing the beliefs of such an agent. Like a child, a strong-AI (AGI) robot would have to learn through input and experiences, constantly progressing and advancing its abilities over time. Both with statistical AI generated by neural networks we need also the concept of \textsl{causality} of events traduced into directionality of logic entailments and deductions in order to give to robots the emulation of human intelligence. Moreover, by using the axioms we can guarantee the \textsl{controlled security} about robot's actions based on logic inferences. For AGI robots we consider the 4-valued Belnap's bilattice of truth-values with knowledge ordering as well, where the value "unknown" is the bottom value, the sentences with this value are indeed unknown facts, that is, the missed knowledge in the AGI robots. Thus, these unknown facts are not part of the robot's knowledge database, and by learn through input and experiences, the robot's knowledge would be naturally expanded over time. Consequently, this phenomena can be represented by the Closed Knowledge Assumption and Logic Inference provided by this paper. Moreover, the truth-value "inconsistent", which is the top value in the knowledge ordering of Belnap's bilattice, is necessary for strong-AI robots to be able to support such inconsistent information and paradoxes, like Liar paradox, during deduction processes.

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

This paper sketches a way to use Belnap's bilattice and a Closed Knowledge Assumption for safe deductions in neuro-symbolic AGI robots, but supplies no mechanism or evidence that the logic actually enforces security or causality.

read the letter

The main takeaway is that the work proposes a neuro-symbolic framework for strong-AI robots that uses Belnap's four-valued logic to manage unknown facts and inconsistencies during learning and deduction. It introduces a Closed Knowledge Assumption to represent how robots build knowledge over time from inputs and experiences, while using the inconsistent value to handle paradoxes. On the positive side, the paper correctly identifies that standard neural networks lack built-in mechanisms for causality and controlled inference, and it points to multi-valued logic as a way to address that. The emphasis on expanding the knowledge base by resolving unknowns is a reasonable way to model progressive learning in robots. It also makes sense to include the top value for inconsistencies so that the system can at least represent conflicting information without crashing. The soft spots are more substantial. The description provides no mapping from neural network outputs to the bilattice values, no explicit inference rules, and no demonstration that the axioms actually enforce directional causality or prevent unsafe actions. The security claim appears to rest on the logic system by definition rather than through any derived property or test case. Without that, it's hard to see how this differs from existing applications of Belnap logic in knowledge representation. This paper would mainly appeal to researchers already working on neuro-symbolic integration who are looking for ideas on logical safety layers. A reader focused on formal verification or empirical AI would find little to engage with here. I would not recommend sending it to peer review in its current state. It needs concrete technical development, such as a small example of deduction or a proof that the assumption yields the desired control properties, before it could usefully be evaluated by referees.

Referee Report

3 major / 3 minor

Summary. The paper claims that Belnap's 4-valued bilattice (with 'unknown' as bottom and 'inconsistent' as top in the knowledge ordering), combined with a Closed Knowledge Assumption and unspecified axioms, enables neuro-symbolic strong-AI robots to learn by expanding their knowledge base from experiences, handle inconsistencies such as the Liar paradox during deductions, and guarantee controlled security plus directional causality in logical inferences for robot actions.

Significance. If the missing formal mappings and derivations were supplied, the framework could offer a logic-based mechanism for safe integration of neural learning with symbolic reasoning in AGI systems, potentially addressing causality and security without external guardrails. As presented, however, the contribution remains at the level of conceptual description without evidence that the bilattice ordering or Closed Knowledge Assumption actually enforces the claimed properties.

major comments (3)

[Abstract] Abstract (paragraph 3): The assertion that the Closed Knowledge Assumption 'represents' learning phenomena and delivers 'controlled security' via logic inference is unsupported; no formal definition of the assumption, no mapping from neural outputs to bilattice values, and no inference rules converting knowledge ordering into action constraints are supplied.
[Abstract] Abstract (final paragraph): The claim that the 'inconsistent' top value is 'necessary' for supporting paradoxes like the Liar paradox during deduction processes lacks any derivation, resolution rule, or example showing how such values produce safe physical robot commands rather than unsafe or undefined actions.
[Abstract] Abstract (paragraph 1): The statement that axioms 'guarantee the controlled security about robot's actions based on logic inferences' is load-bearing for the central security claim yet provides neither the axioms nor a proof that they enforce causality or prevent uncontrolled inferences.

minor comments (3)

First paragraph: 'traduced' is likely intended as 'translated'; the sentence on causality and directionality of entailments is unclear.
Abstract: 'this phenomena' should read 'this phenomenon'.
The manuscript contains no sections, equations, tables, or examples despite the title promising 'Learning and Deductions'; this absence makes the claims difficult to evaluate.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the detailed and constructive report. We acknowledge that the manuscript is primarily conceptual and that the abstract makes strong claims requiring explicit formal support. We will revise the paper to supply the missing definitions, mappings, and examples while preserving the core proposal.

read point-by-point responses

Referee: [Abstract] Abstract (paragraph 3): The assertion that the Closed Knowledge Assumption 'represents' learning phenomena and delivers 'controlled security' via logic inference is unsupported; no formal definition of the assumption, no mapping from neural outputs to bilattice values, and no inference rules converting knowledge ordering into action constraints are supplied.

Authors: We agree the presentation is insufficiently formal. The Closed Knowledge Assumption is intended to capture monotonic expansion of the knowledge base by moving facts from the bottom element ('unknown') upward in the knowledge ordering through experience. In revision we will add an explicit definition, a concrete mapping (neural confidence thresholds to the four bilattice values, with conflict detection producing 'inconsistent'), and inference rules that restrict robot actions to facts that have ascended sufficiently in the ordering. revision: yes
Referee: [Abstract] Abstract (final paragraph): The claim that the 'inconsistent' top value is 'necessary' for supporting paradoxes like the Liar paradox during deduction processes lacks any derivation, resolution rule, or example showing how such values produce safe physical robot commands rather than unsafe or undefined actions.

Authors: Belnap's bilattice is chosen precisely because the top element ('inconsistent') permits continued reasoning without explosion when conflicts or paradoxes appear. We will add a short derivation using the bilattice operations and a worked example in which an inconsistent conclusion triggers a safe default (no physical action) rather than an unsafe command. revision: yes
Referee: [Abstract] Abstract (paragraph 1): The statement that axioms 'guarantee the controlled security about robot's actions based on logic inferences' is load-bearing for the central security claim yet provides neither the axioms nor a proof that they enforce causality or prevent uncontrolled inferences.

Authors: The security claim rests on the directional causality built into the knowledge ordering and the non-explosive handling of inconsistency. We will list the relevant axioms explicitly (monotonic knowledge expansion, non-propagation of inconsistency, and action justification only from sufficiently known facts) and supply a proof sketch showing that these properties block uncontrolled inferences. revision: yes

Circularity Check

1 steps flagged

Closed Knowledge Assumption introduced by the paper to represent learning, then used to guarantee security by construction

specific steps

self definitional [Abstract]
"Moreover, by using the axioms we can guarantee the controlled security about robot's actions based on logic inferences. [...] Consequently, this phenomena can be represented by the Closed Knowledge Assumption and Logic Inference provided by this paper."

The paper defines the Closed Knowledge Assumption to capture the exclusion of unknown facts and expansion via learning, then immediately claims that this assumption (plus axioms provided by the paper) guarantees controlled security. The security result is therefore equivalent to the definitional choice of the assumption rather than derived from independent principles or mappings.

full rationale

The derivation asserts that axioms and the Closed Knowledge Assumption (introduced in the paper to exclude unknowns from the knowledge base while supporting inconsistencies via Belnap bilattice) suffice to guarantee controlled security and directional causality for robot actions. This reduces the central claim to a property of the assumption's own definition rather than an independent derivation, as the text supplies no separate mapping, inference rules, or external validation showing how bilattice values enforce physical action constraints. The step is load-bearing for the strong-AI robot security result but remains internal to the paper's framing.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The framework rests on the standard Belnap bilattice for truth values and introduces the Closed Knowledge Assumption as the mechanism for knowledge expansion and security; no free parameters or invented entities are specified.

axioms (2)

standard math Belnap's 4-valued bilattice with knowledge ordering (unknown as bottom, inconsistent as top)
Provides the truth-value system for representing robot beliefs and handling missing or conflicting information.
ad hoc to paper Closed Knowledge Assumption
States that unknown facts are excluded from the knowledge base and can only be added through learning, enabling controlled deductions.

pith-pipeline@v0.9.0 · 5580 in / 1471 out tokens · 42765 ms · 2026-05-15T20:42:31.962529+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/AbsoluteFloorClosure.lean absolute_floor_iff_bare_distinguishability unclear

?

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

For AGI robots we consider the 4-valued Belnap's bilattice of truth-values with knowledge ordering as well, where the value 'unknown' is the bottom value... Closed Knowledge Assumption and Logic Inference
IndisputableMonolith/Foundation/ArithmeticFromLogic.lean LogicNat.induction unclear

?

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

the truth-value 'inconsistent', which is the top value in the knowledge ordering of Belnap's bilattice, is necessary for strong-AI robots to be able to support such inconsistent information and paradoxes, like Liar paradox

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.

Reference graph

Works this paper leans on

39 extracted references · 39 canonical work pages · 2 internal anchors

[1]

: Multivalued logics: A uniform approach to r easoning in artiﬁcial intelligence

Ginsberg, M. : Multivalued logics: A uniform approach to r easoning in artiﬁcial intelligence. Computational Intelligence, vol.4, 265–316, (1988)

work page 1988
[2]

: Billatices and modal operators

Ginsberg, M. : Billatices and modal operators. Tech.Rep. N.94305,Comp.Science Dept. Stan- ford University, California, (1990)

work page 1990
[3]

: A useful four-valued logic

Belnap, N.D. : A useful four-valued logic. In J-M.Dunn and G.Epstein, editors, Modern Uses of Multiple-V alued Logic. D.Reidel, (1977)

work page 1977
[4]

: Bilattices and the semantics of logic programming

M.C.Fitting, M.C. : Bilattices and the semantics of logic programming. Journal of Logic Programming,11, 91–116, (1991)

work page 1991
[5]

Journal of Multiple-valued Logic & Soft Computing, V ol.14, No.6, 525–564, (2008)

Majki´ c, Z.: Bilattices, Intuitionism and Truth-knowle dge Duality: Concepts and Founda- tions. Journal of Multiple-valued Logic & Soft Computing, V ol.14, No.6, 525–564, (2008)

work page 2008
[6]

Bry F.: Logic programming as constructivism: a formaliza tion and its application to databases. In Proc. of the Symposium on Principles of Databa se Systems, ACM SIGACT- SIGMOD, (1989)

work page 1989
[7]

and Bruynooghe M

Denecker M. and Bruynooghe M. and Marek V .: Logic Programm ing Revisited: Logic Pro- grams as Inductive Deﬁnitions. ACM Transactions on Computa tional Logic 2 (4), 623–654, (2001)

work page 2001
[8]

Inter- national Symposium on Logic-based Program Synthesis and Tr ansformation (LOPSTR), September 7-9, 2005, Imperial College, London,UK, (2005)

Z.Majki´ c Z.: Autoepistemic logic programming for reaso ning with inconsistency. Inter- national Symposium on Logic-based Program Synthesis and Tr ansformation (LOPSTR), September 7-9, 2005, Imperial College, London,UK, (2005)

work page 2005
[9]

: Many-valued intuitionistic implication and inference closure in a bilattice based logic

Majki´ c Z. : Many-valued intuitionistic implication and inference closure in a bilattice based logic. 35th International Symposium on Multiple-V alued Logic (ISMVL 2005), May 18-21, Calgary, Canada, (2005). 32

work page 2005
[10]

19th Workshop on (Constraint) Logic Programming (W(C)LP 20 05), February 21-25, Ulm, Germany, (2005)

Majki´ c Z.: Truth and knowledge ﬁxpoint semantics for ma ny-valued logic programming. 19th Workshop on (Constraint) Logic Programming (W(C)LP 20 05), February 21-25, Ulm, Germany, (2005)

work page 2005
[11]

and Sikorski R.: The Mathematics of Metamathe matics

Rasiowa H. and Sikorski R.: The Mathematics of Metamathe matics. PWN- Polisch Scientiﬁc Publishers, Warsaw, 3rd edition, (1970)

work page 1970
[12]

Journal of Logic Program- ming, V ol

Miller D.: A logical analysis of Modules in Logic Program ming. Journal of Logic Program- ming, V ol. 6, 79–108, (1989)

work page 1989
[13]

Foun- dations of Software Technology and Theoretical Computer Sc ience (Bangalore), 415–422, (2008)

V aliant L.G.: Knowledge Infusion: In Pursuit of Robustn ess in Artiﬁcial Intelligence. Foun- dations of Software Technology and Theoretical Computer Sc ience (Bangalore), 415–422, (2008)

work page 2008
[14]

Dimensions of Neural-symbolic Integration - A Structured Survey

Bader S. and Hitzler P .: Dimensions of Neural-symbolic I ntegration - A Structured Survey. arXiv:cs/0511042, (2005)

work page internal anchor Pith review Pith/arXiv arXiv 2005
[15]

and Zhou L

Sarker K. and Zhou L. and Eberhart A. and Hitzler P .: Neuro-Symbolic Artiﬁcial Intelligence. arXiv:2105.05330, (2021)

work page arXiv 2021
[16]

arXiv:2308.06374, (2023)

Pan Z.J at all: Large Language Models and Knowledge Graph s: Opportunities and Chal- lenges. arXiv:2308.06374, (2023)

work page arXiv 2023
[17]

and Regli W.: Neuro-Symbolic AI in 2024: A Systematic Review

Colelough B.C. and Regli W.: Neuro-Symbolic AI in 2024: A Systematic Review. Logical Foundations of Neuro-Symbolic AI 2024 workshop (IJCAI 2024 ), Jeju Island, South Korea, (2024)

work page 2024
[18]

Walter De Gruyter GmbH, Berlin/Boston, ISBN 978-3-11-099494-0, (20 22)

Majki´ c, Z.: Intensional First Order Logic: from AI to Ne w SQL Big Data. Walter De Gruyter GmbH, Berlin/Boston, ISBN 978-3-11-099494-0, (20 22)

work page
[19]

: Intensional First Order Logic for Strong-A I Generation of Robots

Majki´ c, Z. : Intensional First Order Logic for Strong-A I Generation of Robots. Journal of Advances in Machine Learning & Artiﬁcial Intelligence, V ol ume 4, Issue 1, 23–31, (2023)

work page 2023
[20]

: Strong-AI Autoepistemic Robots Build on In tensional First Order Logic

Majki´ c, Z. : Strong-AI Autoepistemic Robots Build on In tensional First Order Logic. In F.Zhao, D.Miao(eds), AI-generated Content. AIGC 2023. Communications in Computer and Information Science, vol 1946. Springer, Singapore, 33–58 , (2024)

work page 2023
[21]

Deep Learning: A Critical Appraisal

Marcus, G. : Deep learning: a critical appraisal. arXiv: 1801.00631 [cs.AI], (2018)

work page internal anchor Pith review Pith/arXiv arXiv 2018
[22]

P hD Thesys, Stranford University, (2016)

Karpathy, A.: Connecting Images and Natural Language. P hD Thesys, Stranford University, (2016)

work page 2016
[23]

and Tellex, S

Kollar, T. and Tellex, S. and Roy, D. and Roy, N. : Toward un derstanding natural language directions. In Proceedings of the 4th ACM international con ference on human robot interac- tion, (2010)

work page 2010
[24]

: Intensional FOL: Many-Sorted Extension

Majki´ c,Z. : Intensional FOL: Many-Sorted Extension. a rXiv:2409.04469 [cs.AI], 1–21, (2024)

work page arXiv 2024
[25]

: Intensional FOL over Belnap’s Billatice fo r Strong-AI Robotics

Majki´ c, Z. : Intensional FOL over Belnap’s Billatice fo r Strong-AI Robotics. arXiv:2508.02774 [cs.LO], 1–29, (2025)

work page arXiv 2025
[26]

: Der wahrheitsbegriff in den formalisierten Sprachen

Tarski, A. : Der wahrheitsbegriff in den formalisierten Sprachen. Studia Philosophica 1: 261-405. English translation, ’The concepts of truth in for malized langiuges’, appeared in A.Tarski 1956, Logic, Semantics and Metamathematics: Pape rs by Alfred Tarski from 1923 to 1938, Oxford: Clarendon Press, (1936)

work page 1956
[27]

: Deﬂationism and Tarski’s paradise

Ketland, J. : Deﬂationism and Tarski’s paradise. Mind 10 8, 69–94, (1999)

work page 1999
[28]

: Truth and proof - through thick and thin

Shapiro, S. : Truth and proof - through thick and thin. Jou rnal of Philosophy 95, 493–522, (1998)

work page 1998
[29]

and Jeffrey, R

Boolos, G. and Jeffrey, R. : Computability and Logic. Thi rd edition. Cambridge University Press, (1989)

work page 1989
[30]

Analysis 60, 1–4, (2000)

Ketland, J : A proof of the (strethened) Liar formula in a s emantic extension of Peano arith- metic. Analysis 60, 1–4, (2000)

work page 2000
[31]

and Truszczynski, M

Marek, W. and Truszczynski, M. : Autoepistemic logic. Jo urnal of the ACM 38 (3), 588–618, (1991). 33

work page 1991
[32]

: Intensional logic and epistemic independe ncy of intelligent database agents

Majki´ c, Z. : Intensional logic and epistemic independe ncy of intelligent database agents. 2nd International Workshop on Philosophy and Informatics ( WSPI 2005), Kaiserslautern, Germany, April 10-13, (2005)

work page 2005
[33]

and Y osinski, J

Nguyen, A. and Y osinski, J. and Clune, J. : Deep neural net works are easily fooled: High conﬁdence predictions for unrecognizable images. In 2015 I EEE Conference on Computer Vision and Pattern Recognition (CVPR), 427–436, (2015)

work page 2015
[34]

: The purpose put into the machine

Russell, S. : The purpose put into the machine. In Brockma n J (ed) Possible minds: 25 ways of looking at AI, Chap. 3. Penguin Press, New Y ork, 20–32, (20 20)

work page
[35]

: Minds, brains, and programs

Searle, J.R. : Minds, brains, and programs. In Behaviora l and Brain Sciences, V olume 3, Issue 3, Cambridge University Press, (1980)

work page 1980
[36]

Physica D, 335 –346, (1990)

Harnad, S.: The Symbol Grounding Problem. Physica D, 335 –346, (1990)

work page 1990
[37]

ands Floridi, L

Mariarosaria, T. ands Floridi, L. : The symbol grounding problem: A critical review of ﬁf- teen years of research. Journal of Experimental and Theoret ical Artiﬁcial Intelligence, 17(4), 419—445, (2005)

work page 2005
[38]

Cambridge, Mas- sachusetts, The MIT Press, (1999)

Brooks, R.: Cambrian Intelligence: The Early History of the New AI. Cambridge, Mas- sachusetts, The MIT Press, (1999)

work page 1999
[39]

and Aggrawal,M

Sing,H. and Aggrawal,M. and Krishnamurty, B. : Explorin g Neural Models for Parsing Nat- ural Language into First-Order Logic. arXiv:2002.06544 [c s.CL], (2020). 34

work page arXiv 2002