We're about to simulate a human brain on a supercomputer

Digitally enhanced 3D magnetic resonance imaging (MRI) scan of a human brain K H FUNG/SCIENCE PHOTO LIBRARY What would it mean to simulate a human brain? Today’s most powerful computing systems now contain enough computational firepower to run simulations of billions of neurons, comparable to the sophistication of real brains. We increasingly understand how these neurons are wired together, too, leading to brain simulations that researchers hope will reveal secrets of brain function that were previously hidden. Researchers have long tried to isolate specific parts of the brain, modelling smaller regions with a computer to explain particular brain functions. But “we have never been able to bring them all together into one place, into one larger brain model where we can check whether these ideas are at all consistent”, says Markus Diesmann at the Jülich Research Centre in Germany. “This is now changing.” This is in large part because of the power of today’s most advanced supercomputers, which are now approaching exascale , meaning they can carry out a billion billion operations per second. Only four such machines exist, according to the Top500 list . Diesmann and his team are looking at running large-scale brain simulations on one of these systems called JUPITER, short for Joint Undertaking Pioneer for Innovative and Transformative Exascale Research, based in Germany. Last month, Diesmann and his colleagues showed that a simple model of the brain’s neurons and their synapses, called a spiking neural network, could be configured and scaled up to run on JUPITER’s thousands of graphical processing units (GPUs), which would give it a size of 20 billion neurons and 100 trillion connections - equivalent to the human cerebral cortex, where almost all the higher brain functions take place. Running such a simulation promises to produce more valuable results than simulations of smaller brains, such as that of a fruit fly , which have been done before, says Diesmann. Large language models, like the one behind ChatGPT, have shown in recent years that larger systems will contain features that are simply not present in smaller ones. “We know now that large networks can do qualitatively different things than small ones,” says Diesmann. “It’s clear the large networks are different.” Free newsletter Sign up to The Daily The latest on what’s new in science and why it matters each day. “Downscaling is not just simplifying it a little bit, or making it a bit coarser, it means actually giving up certain properties altogether,” says Thomas Nowotny at the University of Sussex, UK. “It’s really important that eventually we can do full-scale [simulations], because otherwise we’re never going to get the real thing.” The model being tested on JUPITER will be grounded in real data from smaller experiments on human brain neurons and synapses, such as how many synapses one neuron should have or their activity levels, says Johanna Senk at the University of Sussex, who is collaborating with Diesmann. “We now have these anatomical data as constraints, but also the computer power,” says Diesmann....

Preview: ~500 words