Brain cells learn faster than machine learning, research reveals

Researchers have demonstrated that brain cells perform more complex networks than machine learning by comparing how both synthetic biological intelligence (SBI) systems known as “dish brain” and the state-of-the-art RL (reinforced learning) algorithms respond to specific stimuli.

This study, “Plasticity and sample efficiency of dynamic networks in biological neural cultures: a comparative study with deep reinforcement learning,” Cyborg and Bionic Systems, It is the first known of this kind.

The research was led by the Cortical Institute, a Melbourne-based startup that created CL1, the world's first commercial biological computer. The research conducted CL1 fuses lab cultured neurons from hard silicon and human stem cells to create a more sophisticated, sustainable AI known as SBI.

This study investigated the integration of complex network dynamics of the raw in vitro nervous system with high density multi-electrode arrays in real-time closed-loop gaming environments.

By embedding spike activity in a low-dimensional space, this study distinguishes between “rest” and “gameplay” conditions and reveals underlying patterns important for real-time monitoring and manipulation.

The analysis highlights dynamic changes in connectivity during gameplay and highlights the highly sample-efficient plasticity of these networks in response to stimuli. To investigate whether this makes sense in a broader context, researchers compared the learning efficiency of these biological systems with state-of-the-art deep RL algorithms such as DQN, A2C, and PPO in Pon Simulation.

In doing so, researchers can introduce meaningful comparisons between biological nervous systems and deep RL, and when samples are limited to actual time courses, even these very simple biological cultures outperform deep RL algorithms across various gaming performance characteristics, meaning higher sample efficiency.

This study was conducted in conjunction with the Turner Institute for Brain and Mental Health at Monash University in Clayton, Australia. IITB-Monash Research Academy, Mumbai, India. University College London, UK Welcome Centre for Human Neuroimaging.

“While there have been substantial advances in the field of AI in recent years, we believe that actual intelligence is not artificial. We consider actual intelligence to be biological. In this study, we decided to investigate whether basic biological learning systems can achieve performance levels that can compete with state-of-the-art deep RL algorithms.”

“The results so far have been extremely encouraging. Understanding how neural activity is linked to information processing, intelligence and ultimately behavior with the central goals of neuroscience research is an important and exciting step in that journey.

“This breakthrough was important evidence that led to the final creation of CL1, the world's first biological computer. It accesses these properties. However, this is not the end, and we believe that further research in Bioengineering Intelligence (BI) will unlock capabilities that far exceed those previously shown.”

Based on the original breakthrough and the launch of CL1, Cortical Lab released a second paper Cell Biomaterials The title “Two Roads Branched: A Pathway to Exploit the Intelligence of Neuronal Culture” proposes a new approach to generating intelligent devices called BioEngineered Intelligence (BI). The paper describing the CL1 platform was also included in the “Down to Business” section of Nature reviews bioengineering.

Interest in using in vitro neuronal cultures embodied within a structured information landscape has grown rapidly. Whether it's biomedical, basic science, information processing and intelligence applications, these systems have great potential. Currently, coordinated efforts have established the field of organoid intelligence (OI) as one pathway.

However, it could potentially lead to another path, particularly by leveraging engineering neural circuits. The research paper examines opportunities and general challenges of OI and BI, and proposes a framework for conceptualizing these different approaches using in vitro neuronal cultures for information processing and intelligence.

In doing so, BI has been formalized as a clear, innovative pathway that can proceed in parallel with OI. Ultimately, important advances can be achieved in either pathway, but the juxtaposition of results from each method is proposed to maximize progress in the most exciting yet ethically sustainable direction.

“Our goal was to go beyond anecdotal demonstrations of biological learning to provide rigorous and quantitative evidence that living neural networks exhibit rapid and adaptive reorganization in response to stimuli.

“Artificial agents often require millions of training steps to show improvement, but these neural cultures adapt much faster and reorganize activity in response to feedback.

“By analyzing how those electrical signals evolve over time, we found clear patterns of learning and dynamic connectivity that reflect the key principles of actual brain function, demonstrating the possibilities of biological systems as fast and efficient learners.”

Moine Cagenejado of the Cortical Institute said, “By converting high-dimensional spike activity into interpretable and low-dimensional representations, we were able to reveal internal plasticity and network reconstruction patterns associated with learning in biological cultures.

“What makes this research truly groundbreaking is the establishment of a head-to-head benchmark between synthetic biological systems and deep RL under comparable sampling constraints. If learning opportunities are limited, and if the animals and humans are closer to the way they actually learn, these biological systems adapt faster, are more efficient, robust and robust.

Hideaki Yamamoto, an associate professor at the Telecommunications Institute at Tohoku University, said, “These synthetic biological systems certainly provide a new approach to understanding the physical substrates of brain calculations. Furthermore, they can open new classes of computing, especially in tasks that expand the brain.

“CL1 will be a powerful platform for implementing this vision. When I first met the team three years ago, they were just beginning to discuss the idea of building their own MEA system. It's very impressive that they developed CL1 and brought it to commercialization in such a short time.”

Provided by the Cortical Institute