Artificial intelligence research and brain science are inseparable. Deep learning, which is currently applied in diverse areas such as autonomous driving and new drug development, is also widely known to be based on a technology called neural networks that mimic the functions of the human brain on computers. Through a mimicking approach, the deep learning neural network demonstrates high performance exceeding other machine learning methods for various problems related to speech, images, and natural language. Over the last decade, the areas of application of machine learning have expanded dramatically with the emergence of deep learning, allowing us to enjoy the benefits of artificial intelligence (AI) in our daily lives.
However, the neural networks make up only a small portion of our brain’s activities, mimicking the mechanism of signal processing of cranial nerves and nerve cells. The actual mechanism of the brain is much more complicated. Computers have yet to be able to reproduce the sophisticated functions of the human brain such as intelligence. Studies on combining brain science and AI are once again gaining spotlight in the areas of novel methods for dramatically improving AI performance and methods for computers to acquire human-like intelligence. Unraveling the mechanism of the human brain is thus expected to hold the key to evolution and the next breakthrough in AI research.

“If we create brain-like neural circuits outside the body, could they have the sophisticated functions of the brain?”
– If we can answer this question, we will be able to understand the factors and conditions necessary for the brain to function and obtain clues on how to develop new forms of AI. As a novel approach to understand the brain, a technique to create brain-like tissues (brain “organoids”) from human stem cells recently attracts broad attention. Current brain-like tissues could not mimic long-range inter-regional connections which are esential for brain functions. Another problem of current brain-like tissues are that they do not show complex activities. Solving these issues and making the tissues acquire functions would deepen the understanding of how the brain works, which will inspire development of AI that surpasses our brain. We can also investigate how disruption of neural circuits causes complex brain illnesses such as psychiatric diseases.
Our project aims to create brain-like tissue from human induced pluripotent stem (iPS) cells and let them acquire functions. After human iPS cells are cultured as a sphere, brain-like tissues (cerebral organoids) start to develop spontaneously. Our project team has developed a unique method to create neural circuit tissues by connecting organoids through numerous axons extended from neurons in organoids. Interestingly, activity of connected organoids is more intense and complex than previously reported. In this project, we will build an AI system that optimizes external stimuli patterns and efficiently trains the connected organoid tissues. Our ultimate goal is to develop tissues with spontaneous higher-order functions.

This project aims to understand the brain by testing the hypothesis that connections between brain regions fascilitate brain activity and development and exploring the conditions necessary for tissues to become intelligent. Unraveling the mechanism by which the brain functions holds not only the key to the further evolution of AI research, but is also expected to contribute enormously to R&D and technological progress in various arenas such as application to brain-machine interfaces, creation of bio-AI, and the development of drugs for psychiatric diseases.

In this research project, we aimed to elucidate artificial intelligence and brain function by developing a tissue called "connectoid" that mimics neural circuits by connecting neural organoids made from human iPS cells and functionalizing them. To achieve this, we modeled the complexity of brain activity beyond ordinary neural organoids by imitating the "connections" between regions in the brain, assigning different roles to each organoid. We aimed to construct a circuit that performs motor processing by stimulating an organoid expressing channelrhodopsin with a specific shape (spatial pattern) of light to provide a sensory input role and making the activity of the connected organoid a motor output. To accomplish this, we applied multiple patterns of light stimulation to the sensory organoid and obtained neural activity patterns on the motor output side using a multielectrode array. We amplified and digitized the obtained signals and examined the relationship between the given signal patterns and the output activity patterns.
Upon repeated analyses using deep learning, we observed a trend of increasing input-output relationship strength with each test. This is thought to be due to the primitive circuit enhancement and learning-like changes caused by the strengthened circuit and increased pattern discrimination ability through repeated stimulation. Furthermore, we developed a schizophrenia model and analyzed changes in neural activity and gene expression. We are also developing a technique to cut neural organoids gently and quickly. While the low efficiency of recording neural organoid activity on multielectrode arrays is recognized as an issue, we have been developing a simple method to improve the recording efficiency of neural organoid activity.

The future possibilities of this research project are expected to contribute significantly to various fields, not only the advancement of artificial intelligence research but also applications to brain-machine interfaces, the creation of bio-AI, and the development of therapeutic drugs for mental disorders. We will continue to develop and pursue the elucidation of the human brain's mechanisms.