By Robin Manhaeve
Review Details
Reviewer has chosen not to be Anonymous
Overall Impression: Good
Content:
Technical Quality of the paper: Average
Originality of the paper: Yes, but limited
Adequacy of the bibliography: Yes
Presentation:
Adequacy of the abstract: Yes
Introduction: background and motivation: Good
Organization of the paper: Satisfactory
Level of English: Satisfactory
Overall presentation: Good
Detailed Comments:
Summary of the paper
This paper surveys neuro-symbolic AI (NSAI) architectures and examines their relationship to recent developments in generative AI. Building on Kautz’s taxonomy, the authors propose a refined classification of NSAI paradigms, including an additional “fibring” architecture, and attempt to relate a range of generative AI methods to these paradigms. The paper also introduces evaluation criteria for comparing NSAI architectures and includes a 4D printing case study as an illustrative application.
Reasons to accept
The paper addresses a timely and relevant problem for the neuro-symbolic AI community: the lack of a shared architectural vocabulary and the growing need to situate NSAI relative to modern generative models.
The revised manuscript is clearer about its scope as a conceptual and taxonomic survey, rather than an empirical contribution, and the architectural descriptions are more precise than in the original version.
The taxonomy-based organization provides a useful high-level structure that can help readers, especially newcomers, navigate a fragmented literature.
The authors have substantially engaged with reviewer feedback and corrected several overgeneralizations in the discussion of generative AI methods.
Reasons to reject
The manuscript is close to being acceptable as a survey article, but several issues should be addressed more explicitly:
While the fibring architecture is now more clearly defined (symbolic aggregation enforcing constraints across multiple neural components), its necessity as a distinct taxonomic category remains unclear. The authors should explicitly address whether fibring introduces an integration principle that cannot be subsumed under existing cooperative or nested NSAI architectures. This could be done, for example, by providing concrete examples where existing paradigms fail to capture the interaction pattern.
The revised manuscript appropriately clarifies that standard generative AI techniques (e.g., GANs, pre-training, fine-tuning, distillation) are not neuro-symbolic by default. However, the section still risks being interpreted as implying stronger connections than are currently established in practice. The authors should clearly label these mappings as conditional or hypothetical throughout the section, and distinguish between methods that are already instantiated in the literature and those that represent plausible future integrations.
The evaluation framework is now more transparent, but the resulting tables remain based largely on expert judgment. The authors should more clearly state the limitations of these evaluations as qualitative comparisons, and avoid language that could be read as definitive rankings between architectures when empirical evidence is sparse.
Further comments
Overall, the paper has improved meaningfully and is close to fulfilling the expectations of a survey article. Addressing the points above would primarily require clarifying scope, strengthening justifications, and further aligning claims with evidence, rather than adding new technical content. With these revisions, the manuscript would be well positioned as a clear and useful overview for the neuro-symbolic AI community.
Survey review criteria
(1) Suitability as an introductory text:
Suitable as an introductory overview for researchers and PhD students, provided that proposed extensions to existing taxonomies are clearly distinguished from established consensus.
(2) Comprehensiveness and balance:
Architectural coverage is broad but somewhat uneven; the generative AI discussion is illustrative rather than comprehensive and should be framed accordingly.
(3) Readability and clarity:
Generally clear and well structured, though dense in places; additional signposting of descriptive vs. proposed material would improve accessibility.
(4) Importance to the NSAI community:
The topic is of clear importance to the field. While the contribution is primarily organizational rather than foundational, it has value as a shared reference if claims are carefully delimited.
