Lab-Grown Brain Organoids Successfully Process Braille and Control Robot Dogs

Researchers at the University of Bristol have made significant strides in biohybrid computing by demonstrating that lab-grown models of the human forebrain, known as organoids, can process tactile information represented in Braille and guide robot dogs.

Biohybrid computing merges living neural tissue with conventional electronics and software, utilizing actual neurons instead of silicon chips for computational tasks. This approach allows biological neural networks to adapt quickly and process information in parallel while consuming minimal energy.

In a study published in _Frontiers in Artificial Intelligence_, the researchers used millimeter-wide organoids containing approximately 10,000 nerves and supporting cells, placed on microelectrode arrays. These arrays enabled stimulation and recording of the organoids’ activity.

A tactile sensor scanned Braille characters and converted them into electrical stimulation patterns, which were sent to the organoids. The researchers then analyzed the organoids’ responses using a conventional machine-learning algorithm, achieving a 61% accuracy rate with a single organoid. When three organoids were combined, the accuracy improved to 83%, demonstrating robustness against artificially introduced noise.

The study indicates that living tissue can effectively transform tactile information while preserving useful patterns. Although the organoids were not trained through experience, they utilized a fixed method for encoding letters and a computer algorithm for classifying neural responses.

Other research, such as that conducted by Melbourne-based Cortical Labs in collaboration with Honda Research Institute Europe, has shown that cultured neurons can enhance their performance through interactions with simulated environments.

However, biological computing faces challenges, as neurons do not respond predictably to electrical inputs like conventional digital hardware. The Embodied Neurocomputation Framework aims to address this issue by systematically optimizing the information entering the neural network.

At the upcoming NeurIPS conference in Sydney, a study will present findings on the Embodied Neurocomputation Framework’s effectiveness. Using Cortical Labs’ CL1 biological computers, researchers connected living neural networks to a virtual agent navigating a grid-like environment, where the agent learned to move toward a target based on simulated odor gradients.

Despite the complexity of biological systems, 12 configurations of electrical signals consistently demonstrated a suitable level of learning across multiple episodes, outperforming traditional reinforcement-learning benchmarks in navigation.

The advantages of biohybrid computing include the brain’s remarkable energy efficiency, enabling complex information processing with minimal energy consumption. Neurons exhibit plasticity, adapting their connections and activity in response to experiences, providing a unique learning pathway.

In a separate demonstration, Cortical Labs showcased how the firing frequency of cells in a biological neural network could direct a robot dog’s movements in a confined space.

Brett Kagan, co-founder and CEO of Cortical Labs, noted that while robots excel at physical control, they still struggle with higher-order control needed for rapid learning and adaptation to changing conditions.

Although these advancements are promising, biological computers are not yet ready for widespread application in environments like Amazon warehouses or autonomous vehicles. Maintaining the health of cultured brain cells requires controlled conditions and a continuous supply of nutrients, which presents additional energy costs.

Despite these challenges, the field is progressing beyond simple demonstrations of organoids in gaming. Companies are developing frameworks to measure performance, optimize interfaces, and assess the consistency of biological computation across tasks.

The ultimate goal is not to create computers entirely from brain tissue but to explore whether living neural networks can effectively complement silicon-based systems. If successful, biohybrid computing may enable the development of systems that learn and adapt in environments where human brains cannot operate.

Source: New Atlas