It might be hard to imagine, but a Swiss company has invented a computer made from living human brain cells, using human brain neurons to send, receive signals, and process data. These brain-like organs can survive for 100 days and consume energy 1 million times less than traditional processors, potentially solving the energy crisis brought by the explosion of artificial intelligence language models. It seems that artificial neural networks are still not as efficient as biological neural networks.

In recent years, with the rapid development of artificial intelligence (AI) technology, scientists have been trying to mimic the human brain by constructing artificial neural networks to enhance computing capabilities. However, as artificial neural networks become increasingly complex, the energy they require is also growing. It is estimated that by 2030, they will consume 3.5% of global electricity. Huang Renxun, the founder of NVIDIA, even suggested that AI will need the energy equivalent to burning 14 Earths.

Scientists have been searching for ways to reduce AI's energy consumption and found that our brains already have a ready-made solution: using real human brain cells to construct artificial neural networks. After all, training language models like GPT-3 requires 10 gigawatt-hours of energy, while the human brain only needs 0.3 kilowatt-hours per day to power 86 billion neurons working 24 hours.

Inspired by this, a Swiss startup company called FinalSpark used living brain cells as a computing array, connecting them to computers and creating a new method that combines biology and computing technology: the biological computer platform Neuroplatform system. This system uses far less energy than traditional bit-based computers.

FinalSpark's biological computer platform connects 16 laboratory-grown human brain cell clusters, known as organoids, each housed in an array and interfaced with external systems via eight electrodes. A microfluidic system supplies these organoids with water and nutrients to sustain their biological functions. This approach, known as wetware computing, leverages the capabilities of laboratory-cultivated organoids—a technology that has emerged in recent years allowing scientists to study miniature replicas of individual organs.

Simply put, neural cells are placed in culture dishes and nurtured using neuron growth medium, growing into brain balls (frontal organoids, FO) with a diameter of 2.5 millimeters. These are then placed onto a multi-electrode array (MEA) composed of eight electrodes, which are connected to a remotely accessible electrophysiological testing system. Interactions and controls are managed through a customized user interface or Python scripts.

But how does it compute? The key lies in the eight electrodes beneath each brain ball. Scientists induce neuron activity within the organoids by applying electrical stimuli of varying intensity and frequency, rewarding with dopamine, and then record the electrical signals from neurons. These signals are transmitted to computers for analysis and processing, similar to the data processing in artificial neural networks.

FinalSpark's biological computer platform features four multi-electrode arrays connected to 16 brain organoids. Due to the limited lifespan of these organoids, which can last up to 100 days under optimal conditions, scientists have cultured and replaced over 1,000 brain organoids over the past three years, collecting more than 18TB of data.

Of course, FinalSpark is not the first team to attempt connecting probes to biological systems. In 2023, researchers in the United States developed a bioprocessor that connects computer hardware with brain organoids and taught it to recognize speech patterns.

Currently, FinalSpark's biological computer platform is primarily used for research purposes and is available free of charge to external academic groups. Four projects have already achieved significant results using the platform. However, FinalSpark plans to expand the platform's capabilities to handle a broader range of wetware computing experimental protocols, such as testing by injecting molecules and drugs into organoids. This implies that the potential applications of this technology extend beyond more energy-efficient computing methods and could also advance organoid research.

Biocomputers made from human brain tissue exhibit certain superiorities inherent in natural design, particularly in terms of energy efficiency. This research not only challenges our traditional understanding of computing and artificial intelligence but also opens up new possibilities for the integration of biology and technology. Despite ethical considerations (the study uses commercially available established cell lines, thus no ethical approval is required), the potential advantages of this biocomputer technology in energy efficiency and computational capacity are undeniable. Future research will reveal more applications of this technology, which might play a significant role in addressing the global energy crisis.

I feel that this type of computer might be more complex than it seems. It could potentially enable a closer, if not seamless, integration between the human brain and computers, creating a new type of advanced biohybrid humans, leaving unmodified humans far behind. Has human evolution reached a critical crossroads? What the future holds is unknown, and those who know may no longer be considered human.

This research was published on May 2nd in the journal Frontiers in Artificial Intelligence.