pith. machine review for the scientific record. sign in

arxiv: 2601.15599 · v3 · submitted 2026-01-22 · 💻 cs.AI

Recognition: no theorem link

Autonomous Business System via Neuro-symbolic AI

Cecil Pang , Hiroki Sayama
Authors on Pith no claims yet

Pith reviewed 2026-05-16 12:42 UTC · model grok-4.3

classification 💻 cs.AI
keywords autonomous business systemneuro-symbolic architectureLLM agentspredicate logic programmingknowledge graphbusiness process automationenterprise data
0
0 comments X

The pith

A neuro-symbolic architecture lets AI agents and a logic engine execute complete business initiatives from natural language instructions.

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

The paper presents AUTOBUS as a way to handle changing business processes that span multiple departments by merging AI language models with strict logic rules. Enterprise information is stored as a knowledge graph that converts into logical facts and rules for consistency. AI agents create custom logic programs for each task based on instructions and tools, and a logic engine runs them to get predictable results. Humans stay in charge of setting rules and checking important choices. This setup promises faster adaptation of business operations without losing control or auditability.

Core claim

AUTOBUS models a business initiative as a network of interrelated tasks with explicit pre- and post-conditions, required data, evaluation rules, and API-level actions. Enterprise data is organized as a knowledge graph translated into logic facts and foundational rules. Core AI agents synthesize task instructions, enterprise semantics, and available tools into task-specific logic programs executed by a logic engine that enforces constraints and produces deterministic outcomes, with humans providing high-level oversight.

What carries the argument

The neuro-symbolic architecture that grounds LLM-generated logic programs in a business knowledge graph converted to predicate logic facts and rules, enabling deterministic execution by a logic engine.

If this is right

  • Business initiatives can be executed end-to-end with explicit constraints and data requirements enforced automatically.
  • Process reconfiguration becomes possible through updated natural language instructions and maintained semantics rather than code changes.
  • Outcomes are deterministic and auditable due to the logic engine's enforcement of rules.
  • Human supervision is focused on policy definition and ambiguous decisions, improving accountability.

Where Pith is reading between the lines

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

  • Extension to non-business domains such as regulatory compliance workflows could benefit from the same mix of flexibility and determinism.
  • Performance would depend on how well the knowledge graph captures all relevant enterprise constraints.
  • Testing across multiple business cases would reveal if the logic engine scales to complex, interdependent task networks.

Load-bearing premise

Large language models can reliably synthesize correct and complete logic programs from task instructions, semantics, and tools without errors or hallucinations.

What would settle it

Running a known business task where the generated logic program violates a pre-defined constraint or leads to an incorrect business outcome when executed.

Figures

Figures reproduced from arXiv: 2601.15599 by Cecil Pang, Hiroki Sayama.

Figure 1
Figure 1. Figure 1: Fig.1. Overview of Autonomous Business System (AUTO [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Business initiative model: inputs traverse [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Fig.3. Principal components of AUTOBUS. Neuro- [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Anatomy of an AUTOBUS logic program. Each t [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Entity–relationship diagram for the case st [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The initiative targets subscribers who (i) subscribe to Product 1, (ii) have churn-risk level 4, and (iii) have household income above median of the city. It is organized into three tasks: retrieving eligible subscribers from enterprise data, fetching median incomes of cities from the web, and sending promotions to the resulting target subscribers through marketing APIs [PITH_FULL_IMAGE:figures/full_fig_p… view at source ↗
Figure 8
Figure 8. Figure 8: Structure of the logic program for Task 3, generated by the AUTOBUS core AI agent from task-specific instruction. The program consists of three sections—facts, task-specific rules, and actions—where facts are derived from enterprise data and the outputs of Tasks 1 and 2; task rules encode the targeting logic; and actions persist the results and invoke marketing campaign APIs [PITH_FULL_IMAGE:figures/full_… view at source ↗
read the original abstract

Modern business environments demand continuous reconfiguration of cross-functional processes, yet most enterprise systems remain organized around siloed departments, rigid workflows, and hard-coded automation. Meanwhile, large language models (LLMs) demonstrate strong capabilities in interpreting natural language and synthesizing unstructured information, but they lack deterministic, auditable execution of complex business logic. We introduce Autonomous Business System (AUTOBUS), a system that integrates LLM-based AI agents, predicate-logic programming, and business-semantics-centric enterprise data into a unified neuro-symbolic architecture for executing end-to-end business initiatives. AUTOBUS models a business initiative as a network of interrelated tasks with explicit pre- and post-conditions, required data, evaluation rules, and API-level actions. Enterprise data is organized as a knowledge graph, whose entities, relationships, and constraints are translated into logic facts and foundational rules that ground reasoning and ensure semantic consistency. Core AI agents synthesize task instructions, enterprise semantics, and available tools into task-specific logic programs, which are executed by a logic engine that enforces constraints, coordinates auxiliary tools, and produces deterministic outcomes. Humans specify task instructions, define and maintain business semantics and policies, curate tools, and supervise high-impact or ambiguous decisions, ensuring accountability and adaptability. We detail the AUTOBUS architecture, the structure of AI-generated logic programs, and the human-AI collaboration model and present a case study that demonstrates accelerated time to market in a data-rich organization. A reference implementation of the case study is available at https://github.com/cecilpang/autobus-paper.

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.

Referee Report

2 major / 2 minor

Summary. The paper introduces AUTOBUS, a neuro-symbolic architecture that integrates LLM-based AI agents, predicate-logic programming, and business-semantics-centric enterprise data (modeled as a knowledge graph) to execute end-to-end business initiatives. Tasks are modeled as networks with explicit pre/post-conditions, data requirements, and rules; core agents synthesize natural-language instructions plus semantics into executable logic programs; a logic engine runs these deterministically while coordinating tools; humans handle oversight for high-impact decisions. A case study is presented claiming accelerated time-to-market, with a GitHub reference implementation.

Significance. If the reliability of LLM-driven logic-program synthesis can be established, the work would offer a practical bridge between flexible LLM reasoning and auditable, constraint-enforcing logic execution, potentially enabling faster reconfiguration of cross-functional business processes than siloed or hard-coded systems.

major comments (2)
  1. [Architecture description and core AI agents] The central mechanism (core AI agents synthesizing task instructions and enterprise semantics into correct, complete predicate-logic programs) is presented without any quantitative evaluation of synthesis accuracy, hallucination rate, or completeness. The logic engine only enforces constraints on the generated program; it has no independent check for semantic fidelity to the intended business logic. This assumption is load-bearing for the deterministic-outcomes claim yet receives no empirical support beyond a high-level case-study reference.
  2. [Case study section] The case study is invoked to demonstrate accelerated time to market, but the manuscript reports no metrics (e.g., generation success rate, error-recovery frequency, comparison against manually authored programs, or failure modes), leaving the central empirical claim unsupported.
minor comments (2)
  1. [Abstract] The abstract states that a reference implementation is available at the cited GitHub link; the paper should explicitly note which components (e.g., example logic programs, knowledge-graph schemas) are included and how they map to the described architecture.
  2. [Structure of AI-generated logic programs] Notation for the structure of AI-generated logic programs (pre/post-conditions, evaluation rules, API actions) could be formalized with a small example in pseudocode or Prolog-like syntax to improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and insightful comments. We agree that the current manuscript would benefit from quantitative evaluation of the core synthesis mechanism and expanded metrics in the case study to strengthen the empirical claims. We will revise the paper to address these points.

read point-by-point responses

  1. Referee: The central mechanism (core AI agents synthesizing task instructions and enterprise semantics into correct, complete predicate-logic programs) is presented without any quantitative evaluation of synthesis accuracy, hallucination rate, or completeness. The logic engine only enforces constraints on the generated program; it has no independent check for semantic fidelity to the intended business logic. This assumption is load-bearing for the deterministic-outcomes claim yet receives no empirical support beyond a high-level case-study reference.

    Authors: We acknowledge that the manuscript presents the architecture at a conceptual level without quantitative metrics on synthesis accuracy, hallucination rates, or completeness. The logic engine enforces syntactic and constraint-based correctness but does not independently verify semantic fidelity to business intent. In the revised manuscript, we will add a new evaluation section with controlled experiments measuring synthesis success rates, hallucination frequency, program completeness, and semantic alignment (via expert review or automated checks where feasible). We will also explicitly discuss the limitations of relying on the logic engine for deterministic outcomes and any planned enhancements for semantic verification. revision: yes

  2. Referee: The case study is invoked to demonstrate accelerated time to market, but the manuscript reports no metrics (e.g., generation success rate, error-recovery frequency, comparison against manually authored programs, or failure modes), leaving the central empirical claim unsupported.

    Authors: We agree that the case study section provides only a high-level illustration without supporting quantitative metrics, which leaves the time-to-market claim unsubstantiated. The reference implementation on GitHub was intended to allow readers to inspect the workflow, but no performance data is reported. In the revision, we will expand the case study with specific metrics including generation success rates, error-recovery frequency, comparisons to manually authored programs (e.g., development time and correctness), and documented failure modes, along with any human oversight statistics. revision: yes

Circularity Check

0 steps flagged

No circularity: architecture integrates existing components without self-referential derivation

full rationale

The paper describes AUTOBUS as an integration of LLMs for program synthesis, predicate logic engines for constraint enforcement, and knowledge graphs for enterprise semantics. No equations, fitted parameters, or predictions are defined that reduce by construction to the inputs. The central mechanism (LLM-generated logic programs executed deterministically) is presented as an engineering composition rather than a closed derivation; reliability of synthesis is an explicit assumption, not a tautology. No self-citation chains or uniqueness theorems are invoked to force the architecture. The case study is referenced externally without reducing claimed outcomes to renamed inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The proposal rests primarily on the domain assumption that LLM agents can reliably translate natural language instructions and semantics into correct predicate logic programs; no free parameters or new invented entities beyond the system name are introduced.

axioms (1)
  • domain assumption LLM-based AI agents can synthesize accurate and complete task-specific logic programs from instructions, semantics, and tools
    This is invoked as the core mechanism for generating executable programs but lacks quantified reliability data in the abstract.
invented entities (1)
  • AUTOBUS no independent evidence
    purpose: Unified neuro-symbolic architecture for autonomous business execution
    The proposed system itself; no independent falsifiable evidence is provided beyond the case study mention.

pith-pipeline@v0.9.0 · 5568 in / 1373 out tokens · 50019 ms · 2026-05-16T12:42:26.120339+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

48 extracted references · 48 canonical work pages · 1 internal anchor

  1. [1]

    Even though LLM-based agentic systems are a very re cent development, they have already made tremendous prog ress

    surveyed the challenges and some of the open r esearch directions in multi-party systems where AI agents interact with more than one human. Even though LLM-based agentic systems are a very re cent development, they have already made tremendous prog ress. Acharya, Kuppan and Divya [17] surveyed the current and potential applications in various real-world d...

    work page

  2. [2]

    break-glass

    used LLMs to translate logical reasoning probl ems in natural language into a symbolic representations an d then used a symbolic logic solver to solve the problems deterministically. Borazjanizadeh and Piantadosi [30] showed that inte grating logic programming into the inference pipeline of LLMs yielded robust performance gains. Yang, Chen and Tam [31] us...

    work page

  3. [3]

    subscribe to a certain product, pay a monthly rate above certain dollars, have a certain churn risk level, and

    work page

  4. [4]

    The initiative is broken down into three tasks, as in Figure 7:

    have household incomes above the median of the citi es residing in. The initiative is broken down into three tasks, as in Figure 7:

    work page

  5. [5]

    Identify and retrieve the target subscribers based on Criterion 1 only

    work page

  6. [6]

    This involves calling a specialized AI agent that fetches median city house hold income data from the web

    Obtain the median household incomes for the cities where the subscribers reside. This involves calling a specialized AI agent that fetches median city house hold income data from the web

    work page

  7. [7]

    Then send them to the marketing platform

    Filter the subscriptions to include only those whos e households have income above the cities’ median. Then send them to the marketing platform. C. Logic Programs The logic programs of the three tasks are generated by the same core AI agent of AUTOBUS, each with task speci fic instructions. Each logic program comprises three sections: (1) facts, (2) task ...

    work page

  8. [8]

    Facts and foundational rules: loaded from enterpris e data, and the outcomes of Task 1 and 2

    work page

  9. [9]

    Task rules: the logic of targeting subscribers who (i) are savable churns determined by Task 1, and (ii) have household incomes above the medians of the cities o f residence, which are the outcome of Task 2

    work page

  10. [10]

    Actions: (i) persist the target subscriptions into the database, and (ii) send the target subscriptions to a marketing campaign via a tool call. D. Result and Discussion The case study was conducted by the first author at the workplace as a proof of concept to mimic a typical business initiative that involves customer segmentation and engagement. While th...

    work page

  11. [11]

    Enterprise resource pla nning systems and its implications for operations function,

    M. Gupta and A. Kohli, “Enterprise resource pla nning systems and its implications for operations function,” Technovation , vol. 26, no. 5, pp. 687–696, May 2006, doi: 10.1016/j.technovation.2004.10.005

    work page doi:10.1016/j.technovation.2004.10.005 2006

  12. [12]

    Masteri ng the Digital Transformation Process: Business Practices and Less ons Learned,

    L. Ivančić, V. Vukšić, and M. Spremić, “Masteri ng the Digital Transformation Process: Business Practices and Less ons Learned,” TABLE 1 CASE STUDY METRIC : TIME TO MARKET Task Existing Approach AUTOBUS 1 a Retrieve savable churns based on product, rate and risk level 1 week 1 day 2 a Obtain the median household income for cities from the web 1 week 1 day...

    work page doi:10.22215/timreview/1217 2019

  13. [13]

    Reengineering Work: Don’t Automate, Obliterate,

    M. Hammer, “Reengineering Work: Don’t Automate, Obliterate,” Harvard Business Review . Accessed: Oct. 15, 2025. [Online]. Available: https://hbr.org/1990/07/reengineering-work-dont-automate-obliterate

    work page 2025

  14. [14]

    T. H. Davenport, Process innovation: reengineering work through information technology . USA: Harvard Business School Press, 1993

    work page 1993

  15. [15]

    Busine ss process redesign as a basic aspect of digital business transformation,

    R. Ivanišević, D. Horvat, and M. Matić, “Busine ss process redesign as a basic aspect of digital business transformation,” Strateg. Manag. - Int. J. Strateg. Manag. Decis. Support Syst. Strateg. Manag. , vol. 30, no. 3, Sep. 2025, doi: 10.5937/StraMan2300040I

    work page doi:10.5937/straman2300040i 2025

  16. [16]

    Ma stering digital transformation through business process management: Investigating alignments, goals, orchestration, and roles,

    A.-M. Stjepić, L. Ivančić, and D. S. Vugec, “Ma stering digital transformation through business process management: Investigating alignments, goals, orchestration, and roles,” J. Entrep. Manag. Innov. , vol. 16, no. 1, pp. 41–74, 2020, doi: 10.7341/20191612

    work page doi:10.7341/20191612 2020

  17. [17]

    Dig ital transformation and the new logics of business process management,

    A. Baiyere, H. Salmela, and T. Tapanainen, “Dig ital transformation and the new logics of business process management,” Eur. J. Inf. Syst. , vol. 29, no. 3, pp. 238–259, May 2020, doi: 10.1080/0960085X.2020.1718007

    work page doi:10.1080/0960085x.2020.1718007 2020

  18. [18]

    AI Agents as Team Members: Effects on Satisfaction, Conflict, Trustwo rthiness, and Willingness to Work With,

    A. R. Dennis, A. Lakhiwal, and A. Sachdeva, “AI Agents as Team Members: Effects on Satisfaction, Conflict, Trustwo rthiness, and Willingness to Work With,” J. Manag. Inf. Syst. , vol. 40, no. 2, pp. 307– 337, Apr. 2023, doi: 10.1080/07421222.2023.2196773

    work page doi:10.1080/07421222.2023.2196773 2023

  19. [19]

    Effective Generative AI: The Human- Algorithm Centaur,

    S. Saghafian and L. Idan, “Effective Generative AI: The Human- Algorithm Centaur,” Harv. Data Sci. Rev. , no. Special Issue 5, Dec. 2024, doi: 10.1162/99608f92.19d78478

    work page doi:10.1162/99608f92.19d78478 2024

  20. [20]

    AI’s errors may be impossible to eliminate – what that means for its use in health care,

    C. Gershenson, “AI’s errors may be impossible to eliminate – what that means for its use in health care,” The Conversation . Accessed: Dec. 14,

    work page

  21. [21]

    Available: http://theconversation.com/ais-errors-may-be- impossible-to-eliminate-what-that-means-for-its-use-in-health-care- 251036

    [Online]. Available: http://theconversation.com/ais-errors-may-be- impossible-to-eliminate-what-that-means-for-its-use-in-health-care- 251036

    work page

  22. [22]

    Logic-Based Artificial Intellige nce,

    R. Thomason, “Logic-Based Artificial Intellige nce,” in The Stanford Encyclopedia of Philosophy , Spring 2024., E. N. Zalta and U. Nodelman, Eds., Metaphysics Research Lab, Stanford University , 2024. Accessed: Dec. 19, 2025. [Online]. Available: https://plato.stanford.edu/archives/spr2024/entries/logic-ai/

    work page 2024

  23. [23]

    Autonomous agents for business process management,

    N. R. Jennings, T. J. Norman, P. Faratin, P. O ’Brien, and B. Odgers, “Autonomous agents for business process management, ” Appl. Artif. Intell. , vol. 14, no. 2, pp. 145–189, Feb. 2000, doi: 10.1080/088395100117106

    work page doi:10.1080/088395100117106 2000

  24. [24]

    An analysis of the design of a pr ogrammable autonomous business,

    N. V. Flor, “An analysis of the design of a pr ogrammable autonomous business,” J. Syst. Inf. Technol. , vol. 7, no. 1–2, pp. 111–128, Jun. 2003, doi: 10.1108/13287260380000776

    work page doi:10.1108/13287260380000776 2003

  25. [25]

    AI Agents and Agentic Systems: A Multi-Expert Analysis,

    L. Hughes et al. , “AI Agents and Agentic Systems: A Multi-Expert Analysis,” J. Comput. Inf. Syst. , vol. 65, no. 4, pp. 489–517, Jul. 2025, doi: 10.1080/08874417.2025.2483832

    work page doi:10.1080/08874417.2025.2483832 2025

  26. [26]

    LLM- Assisted Optimization of Waiting Time in Business P rocesses: A Prompting Method,

    K. Lashkevich, F. Milani, M. Avramenko, and M. Dumas, “LLM- Assisted Optimization of Waiting Time in Business P rocesses: A Prompting Method,” in Business Process Management , A. Marrella, M. Resinas, M. Jans, and M. Rosemann, Eds., Cham: Spri nger Nature Switzerland, 2024, pp. 474–492. doi: 10.1007/978-3-031-70396-6_27

    work page doi:10.1007/978-3-031-70396-6_27 2024

  27. [27]

    Conversational AI for multi-agent c ommunication in Natural Language: Research directions at the Interaction Lab,

    O. Lemon, “Conversational AI for multi-agent c ommunication in Natural Language: Research directions at the Interaction Lab,” AI Commun. , vol. 35, no. 4, pp. 295–308, Sep. 2022, doi: 10.3233/AIC-220147

    work page doi:10.3233/aic-220147 2022

  28. [28]

    Agent ic AI: Autonomous Intelligence for Complex Goals—A Comprehensive Surv ey,

    D. B. Acharya, K. Kuppan, and B. Divya, “Agent ic AI: Autonomous Intelligence for Complex Goals—A Comprehensive Surv ey,” IEEE Access , vol. 13, pp. 18912–18936, 2025, doi: 10.1109/ACCESS.2025.3532853

    work page doi:10.1109/access.2025.3532853 2025

  29. [29]

    Agentic AI Systems: Architecture a nd Evaluation Using a Frictionless Parking Scenario,

    A. Khamis, “Agentic AI Systems: Architecture a nd Evaluation Using a Frictionless Parking Scenario,” IEEE Access , vol. 13, pp. 126052– 126069, 2025, doi: 10.1109/ACCESS.2025.3590264

    work page doi:10.1109/access.2025.3590264 2025

  30. [30]

    Automated Design of Agentic Systems

    S. Hu, C. Lu, and J. Clune, “Automated Design of Agentic Systems,” Mar. 02, 2025, arXiv : arXiv:2408.08435. doi: 10.48550/arXiv.2408.08435

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2408.08435 2025

  31. [31]

    C., & Regli, W

    B. C. Colelough and W. Regli, “Neuro-Symbolic AI in 2024: A Systematic Review,” Apr. 05, 2025, arXiv : arXiv:2501.05435. doi: 10.48550/arXiv.2501.05435

    work page doi:10.48550/arxiv.2501.05435 2024

  32. [32]

    Neurosymbol ic AI: the 3rd wave,

    A. d’Avila Garcez and L. C. Lamb, “Neurosymbol ic AI: the 3rd wave,” Artif. Intell. Rev. , vol. 56, no. 11, pp. 12387–12406, Nov. 2023, doi: 10.1007/s10462-023-10448-w

    work page doi:10.1007/s10462-023-10448-w 2023

  33. [33]

    A review of neuro- symbolic AI integrating reasoning and learning for advanced cognitive systems,

    U. Nawaz, M. Anees-ur-Rahaman, and Z. Saeed, “ A review of neuro- symbolic AI integrating reasoning and learning for advanced cognitive systems,” Intell. Syst. Appl. , vol. 26, p. 200541, Jun. 2025, doi: 10.1016/j.iswa.2025.200541

    work page doi:10.1016/j.iswa.2025.200541 2025

  34. [34]

    Neuro-Symbolic Predictive Process Monitor ing,

    A. Mezini, E. Umili, I. Donadello, F. M. Maggi , M. Mancanelli, and F. Patrizi, “Neuro-Symbolic Predictive Process Monitor ing,” Aug. 31, 2025, arXiv : arXiv:2509.00834. doi: 10.48550/arXiv.2509.00834

    work page doi:10.48550/arxiv.2509.00834 2025

  35. [35]

    Towards Robust Legal Reasoning: Harnessing Logical LLMs in Law,

    M. Kant, S. Nabi, M. Kant, R. Scharrer, M. Ma, and M. Nabi, “Towards Robust Legal Reasoning: Harnessing Logical LLMs in Law,” Feb. 24, 2025, arXiv : arXiv:2502.17638. doi: 10.48550/arXiv.2502.17638

    work page doi:10.48550/arxiv.2502.17638 2025

  36. [36]

    CAFE: Coarse-to-Fine Neural Symbolic Reasoning f or Explainable Recommendation,

    Y. Xian et al. , “CAFE: Coarse-to-Fine Neural Symbolic Reasoning f or Explainable Recommendation,” in Proceedings of the 29th ACM International Conference on Information & Knowledge Management , in CIKM ’20. New York, NY, USA: Association for Computing Machinery, Oct. 2020, pp. 1645–1654. doi: 10.1145/3340531.3412038

    work page doi:10.1145/3340531.3412038 2020

  37. [37]

    Neuro- symbolic recommendation model based on logic query,

    M. Wu, B. Chen, S. Zhu, B. Zheng, W. Peng, and M. Zhang, “Neuro- symbolic recommendation model based on logic query, ” Knowl.-Based Syst. , vol. 284, p. 111311, Jan. 2024, doi: 10.1016/j.knosys.2023.111311

    work page doi:10.1016/j.knosys.2023.111311 2024

  38. [38]

    Recommender systems based on neuro-symbolic knowle dge graph embeddings encoding first-order logic rules,

    G. Spillo, C. Musto, M. de Gemmis, P. Lops, an d G. Semeraro, “Recommender systems based on neuro-symbolic knowle dge graph embeddings encoding first-order logic rules,” User Model. User-Adapt. Interact. , vol. 34, no. 5, pp. 2039–2083, Nov. 2024, doi: 10.1007/s11257- 024-09417-x

    work page doi:10.1007/s11257- 2039

  39. [39]

    ProSLM: A Prolog Synergized Language Model for explainable Domain Specific Knowledge Based Question Answering,

    P. Vakharia, A. Kufeldt, M. Meyers, I. Lane, and L. H. Gilpin, “ProSLM: A Prolog Synergized Language Model for explainable Domain Specific Knowledge Based Question Answering,” in Neural-Symbolic Learning and Reasoning , T. R. Besold, A. d’Avila Garcez, E. Jimenez-Ruiz, R. Confalonieri, P. Madhyastha, and B. Wagner, Eds., C ham: Springer Nature Switzerland...

    work page doi:10.1007/978-3-031-71170- 2024

  40. [40]

    Log ic-LM: Empowering Large Language Models with Symbolic Solvers for Fai thful Logical Reasoning,

    L. Pan, A. Albalak, X. Wang, and W. Wang, “Log ic-LM: Empowering Large Language Models with Symbolic Solvers for Fai thful Logical Reasoning,” in Findings of the Association for Computational Linguistics: EMNLP 2023 , H. Bouamor, J. Pino, and K. Bali, Eds., Singapore: Association for Computational Linguistic s, Dec. 2023, pp. 3806–3824. doi: 10.18653/v1/2...

    work page doi:10.18653/v1/2023.findings-emnlp.248 2023

  41. [41]

    Reliable reasoning beyond natural language: A neurosymbolic approach.arXiv:2407.11373, 2024

    N. Borazjanizadeh and S. T. Piantadosi, “Relia ble Reasoning Beyond Natural Language,” Dec. 01, 2025, arXiv : arXiv:2407.11373. doi: 10.48550/arXiv.2407.11373

    work page doi:10.48550/arxiv.2407.11373 2025

  42. [42]

    Arithmetic R easoning with LLM: Prolog Generation & Permutation,

    X. Yang, B. Chen, and Y.-C. Tam, “Arithmetic R easoning with LLM: Prolog Generation & Permutation,” May 28, 2024, arXiv : arXiv:2405.17893. doi: 10.48550/arXiv.2405.17893

    work page doi:10.48550/arxiv.2405.17893 2024

  43. [43]

    SWI-Prolog,

    J. Wielemaker, T. Schrijvers, M. Triska, and T . Lager, “SWI-Prolog,” Theory Pract. Log. Program. , vol. 12, no. 1–2, pp. 67–96, Jan. 2012, doi: 10.1017/S1471068411000494

    work page doi:10.1017/s1471068411000494 2012

  44. [44]

    Toward Data Systems That Are Business Semantic Centric and AI Agents Assisted,

    C. Pang, “Toward Data Systems That Are Business Semantic Centric and AI Agents Assisted,” IEEE Access , vol. 13, pp. 113752–113762, 2025, doi: 10.1109/ACCESS.2025.3583260

    work page doi:10.1109/access.2025.3583260 2025

  45. [45]

    Review: survey of directly mapping sql databases to the sem antic web,

    J. f. Sequeda, S. hamid Tirmizi, O. Corcho, and D. p. Miranker, “Review: survey of directly mapping sql databases to the sem antic web,” Knowl Eng Rev , vol. 26, no. 4, pp. 445–486, Dec. 2011, doi: 10.1017/S0269888911000208

    work page doi:10.1017/s0269888911000208 2011

  46. [46]

    PyReason: Software for Open World Temporal Logic,

    D. Aditya, K. Mukherji, S. Balasubramanian, A. Chaudhary, and P. Shakarian, “PyReason: Software for Open World Temporal Logic,” Mar. 04, 2023, arXiv : arXiv:2302.13482. doi: 10.48550/arXiv.2302.13482

    work page doi:10.48550/arxiv.2302.13482 2023

  47. [47]

    Logic a: Declarative Data Science for Mere Mortals

    E. Skvortsov, Y. Xia, and B. Ludäscher, “Logic a: Declarative Data Science for Mere Mortals.” OpenProceedings.org, 202 4. doi: 10.48786/EDBT.2024.84

    work page doi:10.48786/edbt.2024.84 2024

  48. [48]

    Modeling complex systems with adaptive networks,

    H. Sayama et al. , “Modeling complex systems with adaptive networks,” Comput. Math. Appl. , vol. 65, no. 10, pp. 1645–1664, May 2013, doi: 10.1016/j.camwa.2012.12.005

    work page doi:10.1016/j.camwa.2012.12.005 2013