Article by Jamila Martin & Roberta Lock
‘Sentient’ Neurons? A Step Towards Synthetic Biological Intelligence
Brett J. Kagan et al., Karl Friston Lab
Synthetic biology holds the key to unlocking the next wave of advancements in artificial intelligence (AI), offering a transformative path by integrating biological components that can enhance the computational capabilities, adaptability, and efficiency of AI systems, opening new frontiers of innovation and ushering in an era of truly intelligent machines. Once considered science fiction, by harnessing the computational power of living neurons, Cortical Labs, an Australian biotech company, has made a remarkable advancement with their product, “Dish Brain” which provided a simulated environment for neurons to play a game of pong. Previous attempts to mimic the complexity of Biological Neuronal Networks (BNNs) fell short, but now, functional in vitro BNNs cultivated from stem cells show promising signs of exhibiting true biological intelligence. Through Dish Brain’s integration of silicon and BNNs, scientists aim to unlock the secrets of intelligent behavior via closed-loop systems.
What did these researchers do?
Dr Brett Kagan and his team utilized the free energy principle to investigate the learning ability of neurons through a negative-positive feedback loop of electrical stimuli to simulate interactive play of a game of pong for cortical neurons from both mice and human stem cells.
The free energy principle states brain cells prefer a predictable environment, that is, cortical neurons will choose a predictable stimulation over an unpredictable one. Neuronal cultures were cultivated on thousands of closely spaced microelectrodes which allowed for electrophysiological activity of defined motor regions to be recorded in real-time to control the movement of the paddle. If the paddle failed to intercept the ball, an unpredictable stimulus was delivered, while a successful interception triggered a predictable stimulus across all electrodes.
Adaptive learning behavior was observed as cells began to generate patterns of electrical pulses that allowed the paddle to move in front of the ball. Initially without this feedback, the cells did not understand the game and had no reason to play. However, the provision of negative reinforcement when the neurons failed to intercept influenced successful play where the hit to miss ratio increased over time and the cells got progressively better at playing the game.
As they explored the role of information entropy in shaping BNN activity, the results suggested that stimulation alone is insufficient to drive learning, which emphasized the need for motivation or a goal that influenced the observable stimulus. Human cortical cultures (HCCs) also demonstrated significantly better gameplay characteristics than mouse cortical cultures (MCCs) in the DishBrain system, indicating that human neurons may have superior information-processing capacity due to biophysical differences between species.
Why is this important?
The neurons in this study revealed a remarkable finding not yet observed in vitro. Although the neurons never got very good at the game, this breakthrough illuminates the immense potential of DishBrain in advancing artificial intelligence and neurological disease models. The adaptive behavior demonstrated by these BNNs within virtual environments has not only deepened our understanding of neural information processing, learning, and adaptation but also opened avenues for developing highly advanced and biologically inspired AI algorithms.
By harnessing the capabilities of BNNs and their real-time response measurement, researchers can gain valuable insights to enhance machine learning models and design more flexible and efficient AI systems, addressing the limitations of rigid silicon computing. This integration of adaptable neural learning has the potential to bridge the gap between human and artificial intelligence, enabling AI systems to generalize knowledge and adapt in unfamiliar contexts, while also offering new approaches to assess and comprehend neurological diseases and their treatments through interactive simulations.
The DishBrain system also provides a unique and interactive platform for studying neurological diseases. Researchers can observe and analyze the effects of various disorders on neuronal behavior, offering unprecedented insights into the complexities of these conditions. The ability to assess the impact of neurological disorders in a controlled and dynamic environment allows for a more comprehensive understanding of disease progression and the evaluation of potential therapeutic interventions.
Finally, this research presents an interesting point for ethical and philosophical discussion; are these neurons ‘sentient’? Although met with opposition from the scientific community, Dr.Kagan defines his neurons as sentient, due to their ability to comprehend the connection between external and internal states via perception and action, as well as their capacity to deduce from sensory states when to participate in specific activities and anticipate the consequences of their actions on the environment.
How did the researchers do this?
Human iPSCs and Mouse embryonic tissue were differentiated into human cortical cultures (HCCs) and mouse cortical cultures (MCCs). These cultures were then seeded at 800,000 cells per MaxOne High-density Micro Electrode Array (HD-MEA), which contained 26,000 platinum electrodes for recording and stimulation purposes. The neurons attached to the culture surface containing the array and formed a neural network, in our case the BNN. As neurons communicate via electrochemical signals, the adjacent MEA electrodes facilitate the recording and capture of neuronal action potentials that are read as extracellular spikes and some electrodes were utilized for providing electrical pulses for our feedback system during game play. The Dish Brain program was designed to incorporate a real time control system of MEA during play based on feedback from the neuronal network and determined the subsequent response based on the neural input to a negative or positive reinforcement response.
Biological Neuronal Network cultured on High-Definition Microelectrode Arrays
What comes next?
The preliminary results from DishBrain suggests that in the future with further development, biology will aid in creating more intelligent computers by revolutionizing the way they learn. However, in an interview with BBC News Dr. Kagan envisions the potential use of his technology in testing treatments for neurodegenerative diseases like Alzheimer's. While current methods only assess activity or inactivity of cells in a dish, Dr. Kagan emphasizes the importance of tapping into the real-time information processing capabilities of brain cells.
The next step involves examining how alcohol affects the mini-brain's ability to play Pong, with the aim of demonstrating its similarity to the human brain's response. While Dr. Kagan describes the system as sentient, others prefer the term "thinking system" to describe the basic cognitive processes observed. The mini-brains are expected to become more complex over time, but ethical considerations are being addressed in collaboration with bioethicists to prevent the unintentional creation of a conscious brain. Comparing the technology's early stage to the development of transistors in the computer industry, Dr. Kagan highlights the potential for significant advancements with dedicated research. While AI has achieved remarkable feats, the mini-brain's ability to learn without explicit teaching makes it adaptable and flexible, according to Professor Karl Friston of University College London, who collaborates with Dr. Kagan.