A Computational Perspective on NeuroAI and Synthetic Biological Intelligence
Dhruvik Patel, Md Sayed Tanveer, Jesus Gonzalez-Ferrer, Alon Loeffler, Brett J. Kagan, Mohammed A. Mostajo-Radji, Ge Wang · 2025
Read A Computational Perspective on NeuroAI and Synthetic Biological Intelligence by Dhruvik Patel, Md Sayed Tanveer, Jesus Gonzalez-Ferrer, Alon Loeffler, Brett J. Kagan, Mohammed A. Mostajo-Radji, Ge Wang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-23896.txt.
Introduction
Brains grown in dishes and brains simulated in silicon are starting to compute together. Dhruvik Patel and 6 co-authors review NeuroAI and its branch of synthetic biological intelligence from a computational angle. They sort the field into hardware, software, and wetware. Synthetic biological systems join the adaptive learning of living neural networks with engineered platforms, serving as test beds for neural computation and biohybrid designs.
Recent advances in organoid intelligence, neuromorphic computing, and neuro-symbolic learning get surveyed. The authors see a new class of systems emerging, ones that compute through the interaction of living tissue and digital algorithms.
Catalog record
| CREATOR | Dhruvik Patel, Md Sayed Tanveer, Jesus Gonzalez-Ferrer, Alon Loeffler, Brett J. Kagan, Mohammed A. Mostajo-Radji, Ge Wang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 72 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-23896
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A Computational Perspective on NeuroAI and Synthetic Biological Intelligence Dhruvik Patel, Md Sayed Tanveer, Jesus Gonzalez-Ferrer, Alon Loeffler, Brett J. Kagan, Mohammed A. Mostajo-Radji, Ge Wang Abstract NeuroAI is an emerging field at the intersection of neuroscience and artificial intelligence, where insights from brain function guide the design of intelligent systems. A central area within this field is synthetic biological intelligence (SBI), which combines the adaptive learning properties of biological neural networks with engineered hardware and software. SBI systems provide a platform for modeling neural computation, developing biohybrid architectures, and enabling new forms of embodied intelligence. In this review, we organize the NeuroAI landscape into three interacting domains: hardware, software, and wetware. We outline computational frameworks that integrate biological and non-biological systems and highlight recent advances in organoid intelligence, neuromorphic computing, and neuro-symbolic learning. These developments collectively point toward a new class of systems that compute through interactions between living neural tissue and digital algorithms. keywords Neuroscience, artificial intelligence, biological computing, synthetic biological intelligence, organoid intelligence, brain-computer interface, neuroAI † † equal-contributors: These authors contributed equally to this work. † † equal-contributors: These authors contributed equally to this work. 1 Introduction The emergence of NeuroAI, particularly synthetic biological intelligence (SBI), represents a paradigm shift at the intersection of neuroscience, artificial intelligence, engineering, and biotechnology. This field leverages the inherent computational capabilities of biological neural systems for engineered intelligent behaviors. A key approach involves the development of in vitro neural cultures and neural organoids, which recapture structural and functional features of the developing brain and enable the study of biological learning, memory, and reasoning processes Kagan et al. 2023 ; Smirnova et al. 2023c . This multidisciplinary convergence enhances the potential of each contributing field while introducing new theoretical, technological, ethical, and regulatory challenges. The pursuit of NeuroAI and SBI reflects an effort to understand and harness the computational power of biological neural networks (BNNs). By integrating living neural cultures with electronic interfaces, researchers emulate and potentially extend the brain’s information-processing capabilities. This integration opens new opportunities for biomedical research, personalized medicine, and hybrid computing systems Li et al. 2020 ; Tejavibulya and Sia 2016 . These platforms may support disease modeling, therapeutic screening, and the investigation of neural mechanisms underlying cognition, decision-making, and action Haring et al. 2017 ; Wang 2018 . The milestones enabling the development of SBI span more than a century, with contributions from both biological and computational sciences. Early biological advances include the first in vitro tissue cultures in the late 19th century Pham 2018 and the demonstration that neurons can be cultured ex vivo in the early 20th century Harrison 1910 . The introduction of multielectrode arrays (MEAs) in the 1970s provided a means to record neural activity and study circuit dynamics Pine 2006 . The isolation of embryonic stem cells (ESCs) Thomson et al. 1998 and the development of induced pluripotent stem cells (iPSCs) Takahashi and Yamanaka 2006 enabled scalable, patient-specific generation of neural tissue. Recognizing the various limitations of two-dimensional cultures, researchers developed three-dimensional brain organoids in mouse and human models, leading to the onset of organoid intelligence (OI) Eiraku et al. 2008 ; Kadoshima et al. 2013 ; Lancaster et al. 2013 ; Lancaster and Knoblich 2014 . These 3D constructs can faithfully replicate various aspects of in vivo neural development, including cellular diversity and spatial organization, which are essential for the formation of functional neural networks Li et al. 2020 ; Tejavibulya and Sia 2016 . In parallel, computational science has undergone transformative progress. Theoretical foundations were laid by early work on probabilistic models such as Markov chains Chung 1967 ; Gely P. Basharin 2004 ; Krogh et al. 1994 , neural representations McCulloch and Pitts 1943 , the Turing test Turing 1950 , and the 1956 Dartmouth Conference that launched the field of artificial intelligence Frana and Klein 2021 . Reinforcement learning approaches such as Temporal Difference Learning Sutton 1988 and Q-learning Watkins and Dayan 1992 established formal methods for decision-making. The 21st century witnessed the emergence of deep learning, including convolutional networks like AlexNet Krizhevsky et al. 2012 and transformer-based architectures Vaswani et al. 2017 , culminating in today’s large language models Brown et al. 2020 . At the same time, neuroscientific theories such as Global Workspace Theory Baars 2005 , Integrated Information Theory Tononi 2012 , Predictive Coding Kilner et al. 2007 , and Active Inference Friston et al. 2016 have inspired biologically grounded strategies for developing more flexible and interpretable AI systems. This convergence is exemplified by recent NeuroAI frameworks that draw from both neuroscience and computer science, including neuro-symbolic systems that combine rule-based logic with neural learning Garcez and Lamb 2023 ; Sheth et al. 2023 , and neuromorphic hardware that emulates the spiking, event-driven architecture of biological neurons Yang et al. 2020 ; Ivanov et al. 2022 ; Abdallah and Dang 2022 . Here, we examine the emerging field of SBI and NeuroAI with a focus on its computational foundations. We review key technological and conceptual advances, algorithms, and architectures that connect biological and artificial forms of intelligence. By highlighting these integrative efforts, we provide a comprehensive framework for understanding the current state and future directions of this rapidly evolving field. 2 Biological Neural Networks (BNNs): The Foundation of Bio-intelligence and NeuroAI A central aim of NeuroAI is to design artificial or bio-engineered systems that mimic the structure and function of brains. This requires a clear understanding of the key characteristics of biological neural networks (BNNs). With the emergence of numerous recent breakthroughs in automated ”smart” systems in the field of artificial intelligence (AI), which is based on artificial neural networks (ANNs), it is essential that we also discuss how BNNs differ from ANNs. 2.1 Complexity of a Single Biological Neuron ANNs are often described as “brain-inspired,” with perceptrons serving as analogous to biological neurons Block 1962 . Yet, a real neuron performs significantly more complex computations than the simple weighted sum and threshold used in a perceptron. First, dendritic trees act not as single summing points but as distributed filter banks, as shown in Figure 1 . Each branch approximates a leaky RC cable with distinct membrane resistance ( R_{m} ), axial resistance ( R_{a} ), and capacitance ( C_{m} ). Synaptic inputs undergo location- and time-dependent attenuation, with distal signals arriving more weakly and slowly than proximal ones London and Häusser 2005 . Superimposed on this passive scaffold are voltage-gated sodium and calcium channels. When local depolarization exceeds threshold, these channels open and drive the membrane potential toward the reversal potential E_{r} , producing a regenerative dendritic spike that propagates toward the soma. Somatic action potentials can also back-propagate into the dendrites, reopening these channels in reverse and triggering calcium influx that enables synapse-specific plasticity Bicknell and Häusser 2021 ; Larkum 2022 . Figure 1: Representation of dendrite branches as electrical circuits with active elements. Among these mechanisms, N-methyl-D-aspartate (NMDA) receptors serve as time-sensitive amplifiers. Their conductance increases only when glutamate release coincides with sufficient depolarization, generating a supra-linear plateau that amplifies synchronous inputs while filtering out asynchronous or reversed spikes. In engineering terms, NMDA-rich dendrites function like temporal filters and logical AND gates. They can distinguish input sequences and timing without requiring