Credit: BlackJack3D / iStock / Getty Images Plus “The human brain has 100 billion neurons, each neuron connected to 10,000 other neurons. Sitting on your shoulders is the most complicated object in the known universe.” — Michio Kaku, PhD.
Since most examples of brain-inspired silicon chips are based on digital electronic principles, their capacity to fully imitate brain function is limited. Self-organizing brain organoids connected to microelectrode arrays (MEAs) can be changed in function to create neural networks. These networks, called organoid neural networks (ONNs), show the capacity for unsupervised learning, which is what artificial intelligence (AI) is based on. These mini-organs, when connected to the right hardware, can even be trained to recognize speech.
This brain-inspired computing hardware, or “Brainoware,” could overcome existing shortcomings in AI technologies, providing natural solutions to challenges regarding time and energy consumption and heat production of current AI hardware. These ONNs may also have the necessary complexity and diversity to mimic a human brain, which could inspire the development of more sophisticated and human-like AI systems.
Brainoware was developed in the labs of Feng Guo, PhD, at Indiana University Bloomington and Mingxia Gu, MD, PhD, at Cincinnati Children’s Hospital Medical Center. The findings were published in the research article “ Brain organoid reservoir computing for artificial intelligence ” in Nature Electronics . Organoid neural networks
Hongwei Cai, the first author of Brainoware, and his colleagues wanted to find a biological way to solve the problem of reservoir computing. Reservoir computing is known for its unique way of processing and learning complex temporal and sequential data. Reservoir computing enables the extraction of intricate patterns and relationships from temporal sequences. This approach ensures fast training and emphasizes energy efficiency, making it a viable option for environmentally conscious AI solutions.
Reservoir computing has demonstrated promising outcomes across various applications, including time-series prediction, speech recognition, language modeling, and addressing complex nonlinear dynamical systems. Its unique architecture and training methodology offer an innovative alternative for effectively managing sequential and temporal data. This renders it a valuable tool within machine learning and artificial intelligence research domains.
Will the computers of the future be made of brains?
Not quite yet. While Brainoware doesn’t require much power consumption, it depends on things like incubators and some systems, such as skilled cell culture technicians or automated systems, for maintaining the cell culture.
Additionally, the generation of organoids is still a relatively uncontrolled, heterogeneous mix of cell types, both dead and alive. Recent engineering efforts to improve the conditions for organoid differentiation and growth, as well as changing their microenvironments, could make it possible to make and keep a lot of standardized organoids.
Data management and analysis is another technical challenge. Data interpretation, extraction, and processing improvements from multiple sources and modalities are still needed to optimize the encoding and decoding of temporal information to and from Brainoware. The power of Brainoware is all but useless without the development of new algorithms and methods for analyzing and visualizing the data.
Although it may be decades before general biocomputing systems are developed, this research will likely yield fundamental insights into learning mechanisms, neural development, and the cognitive implications of neurodegenerative diseases. It could also aid in developing preclinical models of cognitive impairment for testing new therapeutics.