Write a comment

This paper offers a timely, high-level synthesis(review) of NeuroAI with a central focus on Synthetic Biological Intelligence (SBI); systems that fuse living neural tissue with engineered hardware and software. The authors organize the field into a three-part framework (hardware, software, wetware), survey key advances (organoid intelligence, neuromorphic platforms, neuro-symbolic methods), and argue that computation emerges through interactions between biological and digital components. The review’s strengths are its breadth across modalities, up-to-date exemplars, and practical attention to pipelines (data acquisition, interfacing, spike processing, and standards). As a field-scoping piece, it clarifies terminology, highlights integration patterns for bio-hybrid systems, and motivates evaluation along stability–plasticity and embodiment dimensions. Overall, its an excellent work that provides a useful map for researchers building hybrid systems that learn and adapt in brain-like ways. I had a lot to learn from it, its absolutely excellent.

Major issues

• Software treatment is comparatively narrow (esp. section 5.2).

  The discussion concentrates on neuro-symbolic AI, RL, and Active Inference, while giving less depth to other active neuro-inspired families that are directly relevant to SBI controllers/decoders and closed-loop learning.

• You can consider expanding section 5 with concise subsections and one alignment table that maps SBI use-cases (e.g., reservoir readout, closed-loop shaping/stimulation, decoding with memory, adaptive control on neuromorphic substrates) to software families, including:

  • Predictive coding & dendritic/compartmental models.

  • Continual/biologically plausible learning (local credit assignment, neuromodulation-inspired rules; cross-reference stability–plasticity content).

  • Reservoir/liquid computing (and software analogs to in-vitro reservoirs).

  • Memory-augmented models (NTMs, Memory Networks; working-memory controllers).

  • World-model / model-based RL approaches as engineering complements to AIF.

  • On-chip/online learning patterns (Hebbian, STDP, reward-modulated STDP) bridging neuromorphic hardware to controller design.

Minor issues

1. Terminology and scope clarity: Terms like SBI, organoid intelligence, and broader “bioengineered intelligence” occasionally blur.

Suggestion: Add a short glossary/box near the start; commit to one primary term and cross-reference synonyms when first used.

1. Evaluation criteria are implicit rather than explicit: It would beneficial to show a compact view of “what good looks like” for SBI systems.

Suggestion: You could provide a metric checklist (e.g., stability vs. plasticity, retention under task-switch, adaptation latency, energy/efficiency, robustness to perturbation, interface bandwidth) with brief justifications.

This is a great job! Thank you for publishing this. Weldone!

Competing interests

The author declares that they have no competing interests.

Use of Artificial Intelligence (AI)

The author declares that they did not use generative AI to come up with new ideas for their review.