It may be hard to imagine that a Swiss company has invented a computer made from living human brain cells that sends and receives signals and processes data through neurons in the human brain. These brain organoids can survive for 100 days and consume 1 million times less energy than traditional processors, which is expected to completely solve the potential energy crisis caused by the explosion of artificial intelligence large language models.
In recent years, with the rapid development of artificial intelligence (AI) technology, scientists have been trying to mimic the human brain, trying to enhance computing power by building artificial neural networks. However, as artificial neural networks become more complex and require more energy, claiming to consume 3.5% of the world's electricity by 2030, Nvidia founder Jensen Huang even claims that AI will need to burn 14 Earths of energy.
Scientists have been looking for ways to reduce the energy consumption of AI, and it turns out that our brains already have a ready-made answer – to build artificial neural networks from real human brain cells. After all, it takes 10 gigawatt-hours of energy to train a language model like GPT-3, while the human brain only needs 0.3 kilowatt-hours a day to do the work of 86 billion neurons for 24 hours.
Inspired by this, FinalSpark, a Swiss start-up, has created a new way to combine biology and computing technology by using living brain cells as computing arrays and connecting them to computers: the Neuroplatform system, a biocomputer platform. Such systems use far less energy than traditional bit computers.
FinalSpark's biocomputer platform connects 16 lab-grown spheroids of human brain cells, called organoids, each housed in an array connected to an external system by eight electrodes, and a microfluidic system provides water and nutrients to these organoids to ensure their vital activities. The method, known as wetware computing, harnesses the power of organoids grown in the lab, and is a new technology developed in recent years that allows scientists to study micro-replicas of individual organs.
To put it simply, nerve cells are placed in a petri dish, cultivated in a closed-loop microfluidic system, cultured with neuronal culture medium, grown into a 2.5 mm diameter brainball (forebrain organoid, FO), placed on a multi-electrode array (MEA) composed of 8 electrodes, and then connected to a remotely accessible electrophysiology assay system, and interactively controlled through a custom user interface or Python script.
So how does it calculate? The key lies in the 8 electrodes under each brainball, and scientists apply electrical stimulation of different intensities and frequencies to induce neuronal activity in organoids, reward them with dopamine, and then record the electrical activity signals in the neurons and transmit them to the computer for analysis and processing, which is very similar to the data processing process in artificial neural networks.
FinalSpark's biocomputer platform has four multi-electrode arrays connected to 16 brain balls, which can survive for 100 days at best, and over the past three years of experiments, scientists have cultivated and replaced more than 1,000 brain balls and collected more than 18 terabytes of data.
Of course, FinalSpark is not the first team to try to connect probes to biological systems, and in 2023, researchers in the United States developed a bioprocessor that connects computer hardware to brain organoids and has it learn to recognize speech patterns.
FinalSpark's biocomputer platform is currently mainly used for research purposes and is freely available to external academic groups, and four projects have achieved certain results. But FinalSpark plans to expand the platform's capabilities to handle a wider range of protocols for wet-matter computing experiments, such as injecting molecules and drugs into organoids for testing. This means that the potential applications of this technology will not only include more energy-efficient computing methods, but also advance organoid research.
Biocomputers built from human brain tissue demonstrate the superiority of natural design in some ways, especially 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 combination of biology and technology. Despite ethical considerations (the study used a commercially available established cell line and did not require ethical approval), the potential benefits of this biocomputer technology in terms of energy efficiency and computing power are undeniable. Future research will reveal more applications of this technology, which may play an important role in solving the global energy crisis.
I feel that this kind of computer may not be so simple, it is more likely to make the human brain and the computer more closely combined, or even completely seamless, to create a new type of cyborg that is far ahead, so that all unmodified humans are far behind, and the evolution of human beings has reached a critical crossroads? What the future holds, no one knows—probably not people.
The study was published May 2 in the journal Frontiers in Artificial Intelligence.
Thesis:
https://www.frontiersin.org/articles/10.3389/frai.2024.1376042/full