In a groundbreaking advancement at the intersection of neuroscience and artificial intelligence, researchers from Cortical Labs have demonstrated that biological neural cultures learn faster and more efficiently than some of the most advanced machine learning algorithms available today. This remarkable finding was revealed through a pioneering experimental comparison between in vitro neural networks known as “DishBrain” and state-of-the-art deep reinforcement learning (RL) models, marking a critical milestone in understanding intelligence itself.
DishBrain, the central component of this study, is an innovative system that merges live human neurons cultivated from stem cells with silicon-based substrates, enabling a truly hybrid platform where biological and synthetic components interact seamlessly. Utilizing high-density multi-electrode arrays (MEAs), this synthetic biological intelligence (SBI) system facilitates real-time closed-loop interactions within dynamic game environments, specifically a version of the classic Pong game. Such integration permits the monitoring and manipulation of neural activity as it dynamically evolves in response to stimuli, an experimental design that has never before been executed in this scale or precision.
The research, formally titled “Dynamic Network Plasticity and Sample Efficiency in Biological Neural Cultures: A Comparative Study with Deep Reinforcement Learning,” systematically maps how neural cultures display plasticity—adaptive changes in their connectivity and firing patterns—on a moment-to-moment basis during gameplay versus rest states. By reducing the high-dimensional spiking activity of neurons into interpretable, low-dimensional representations, the investigators could decipher the underlying network reconfiguration that signifies learning and adaptation. This level of analysis corroborates fundamental neuroscience theories linking synaptic plasticity to functional changes tied to intelligence and learning capacity.
.adsslot_uXMRYWPr6V{ width:728px !important; height:90px !important; }
@media (max-width:1199px) { .adsslot_uXMRYWPr6V{ width:468px !important; height:60px !important; } }
@media (max-width:767px) { .adsslot_uXMRYWPr6V{ width:320px !important; height:50px !important; } }
ADVERTISEMENT
One of the most profound aspects of the study lies in its quantification of sample efficiency—the number of training examples or interactions a system requires to improve its performance. Unlike artificial RL systems such as DQN (Deep Q-Network), A2C (Advantage Actor-Critic), and PPO (Proximal Policy Optimization), which often necessitate millions of iterations before demonstrating meaningful learning, the biological neural cultures exhibited rapid acquisition of new behaviors with far fewer samples. This aligns more closely with natural animal learning, where organisms adapt effectively within limited exposures to stimuli, underscoring the superior adaptability of living neural networks.
Beyond serving as a compelling proof of concept, this work paves the way for a paradigm shift in AI research by suggesting that intelligence should not be regarded solely as an artificial construct born of algorithms but as a fundamentally biological phenomenon. According to Cortical Labs’ Chief Scientific Officer, Brett Kagan, this breakthrough challenges the existing notion that intelligence can be fully replicated through silicon-based computation alone and advocates for embracing biological substrates as powerful computational entities in their own right.
The implications of harnessing “Bioengineered Intelligence” (BI), a term introduced by the team in a companion study, extend far beyond isolated learning tests. BI envisions a future where engineered neural circuits from lab-grown neurons can be precisely structured and interfaced with computational devices to perform complex processing tasks, possibly rivaling or exceeding traditional AI methods. This contrasts yet complements the emerging field of Organoid Intelligence (OI), which employs naturally grown brain organoids but without the same degree of engineered control over network architecture.
In analyzing the data, Cortical Labs researchers illustrated that the rapid reorganization of synaptic activity seen in DishBrain was not merely a statistical phenomenon but reflected genuine functional improvements in learning task performance. This was evidenced by the reconfiguration of connectivity patterns between neurons as gameplay progressed, mirroring principles that govern cognition in intact mammalian brains. Moein Khajehnejad, a co-author of the study, highlighted how extracting interpretable, low-dimensional signals from spiking patterns illuminated these internal plasticity processes more clearly than previous methodologies allowed.
The comparative benchmarking conducted, placing biological systems and deep RL methods on equal footing regarding the number of samples and real-world time available for learning, marks a pioneering approach in AI evaluation. This direct head-to-head challenge underscores the potential of synthetic biological systems not only to match but to surpass artificial agents in adaptation speed and robustness under conditions that emulate true learning scenarios. It’s a humbling insight for researchers striving to unravel the essence of cognition and intelligence.
Support for this study comes from an esteemed international consortium involving Monash University’s Turner Institute for Brain and Mental Health, the IITB-Monash Research Academy in India, and University College London’s Wellcome Centre for Human Neuroimaging. The collaborative expertise underscores the multidisciplinary complexity of the research, integrating stem cell biology, computer science, neuroscience, and bioengineering in unprecedented ways.
Experts in the field have expressed enthusiasm about the potential of the CL1 platform, the first commercial biological computer stemming from this research. Professor Mirella Dottori of the University of Wollongong remarked that such technology not only advances fundamental neuroscience but also offers novel avenues to explore neurological diseases by providing dynamic, functional measurements of neuronal network behavior. Similarly, Hideaki Yamamoto from Tohoku University praised the rapid development and commercialization of the CL1 device, recognizing its promise as a versatile tool for investigating brain computation and beyond.
This landmark study signals a fundamental shift in how scientists and engineers might approach the future of artificial intelligence. By bridging living neural tissue with computational frameworks, researchers are charting a course toward machines that do not merely mimic but embody biological intelligence. The fast and efficient learning capabilities demonstrated by these cultured neural networks challenge prevailing assumptions and open exciting possibilities for developing adaptive, resilient, and ethically sustainable AI systems.
With ongoing research and technological refinement, Bioengineered Intelligence stands poised to redefine the boundaries of computation, intelligence, and what it means for a system to “learn.” As Cortical Labs pushes forward with this nascent technology, the convergence of biology and machine heralds a new era where the secrets of the brain are not only studied but actively harnessed to shape the future of intelligent machines.
Subject of Research: Cells, Synthetic biology, Intelligence, Biomaterials, Biotechnology
Article Title: Dynamic Network Plasticity and Sample Efficiency in Biological Neural Cultures: A Comparative Study with Deep Reinforcement Learning
News Publication Date: 12 August 2025
Web References: http://dx.doi.org/10.1016/j.celbio.2025.100156
References:
Cortical Labs et al., Dynamic Network Plasticity and Sample Efficiency in Biological Neural Cultures: A Comparative Study with Deep Reinforcement Learning, Cyborg and Bionic System: A Science Partner Journal, 2025.
Image Credits: Cortical Labs
Tags: artificial intelligence advancementsbiological neural networksdeep reinforcement learning comparisonDishBrain research findingsdynamic game environments in researchexperimental neuroscience breakthroughshybrid biological intelligence systemsmachine learning vs neuroscienceneural cultures learning efficiencyneuroplasticity in biological systemsreal-time neural activity monitoringstem cell-derived neurons
