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

What would it mean to simulate a human brain? Today’s most powerful computer systems now contain enough computing power to run simulations of billions of neurons, comparable to the sophistication of real brains. We also increasingly understand how these neurons are connected to each other, leading to brain simulations that researchers hope will reveal secrets of brain functioning that were previously hidden.

Researchers have long tried to isolate specific parts of the brain, modeling smaller regions with a computer to explain particular brain functions. But “we’ve never managed to bring them all together in one place, in a larger brain model where we could check whether these ideas are coherent,” says Markus Diesmann of the Jülich research center in Germany. “That’s changing.”

This is largely due to the power of today’s most advanced supercomputers, which are now approaching exascale, meaning they can perform a billion trillion operations per second. Only four such machines exist, according to the Top500 list. Diesmann and his team plan to run large-scale brain simulations on one such system 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 to run on JUPITER’s thousands of graphics processing units (GPUs), giving it a size of 20 billion neurons and 100 trillion connections – the equivalent of the human cerebral cortex, where almost all higher brain functions take place.

Running such a simulation promises to produce more valuable results than simulations of smaller brains, like that of a fruit fly, that have already been performed, Diesmann says. Large language models, like the one behind ChatGPT, have shown in recent years that larger systems contain features that are simply not present in smaller ones. “We now know that large networks can do qualitatively different things than small networks,” says Diesmann. “It’s clear that the big networks are different.”

“Reducing the scale is not just about simplifying it a little or making it a little cruder, it actually means abandoning some properties altogether,” says Thomas Nowotny of the University of Sussex, UK. “It’s really important that one day we can do large-scale projects. [simulations]because otherwise we will never get the real thing.

The model tested on JUPITER will be based on real data from smaller experiments on neurons and synapses in the human brain, such as how many synapses a neuron should have or their activity levels, says Johanna Senk of the University of Sussex, who is collaborating with Diesmann. “We now have these anatomical data as constraints, but also the power of the computer,” explains Diesmann.

Large-scale brain simulations could allow researchers to test basic theories about brain functionality that are impossible on smaller models or with real brains, Nowotny says, such as how memories are formed. This could be tested by feeding images to a brain network, observing how it responds, and recording how this memory formation changes depending on brain size. It could also create a way to test drugs, Nowotny says, for example by examining how certain drugs affect models of epilepsy, characterized by seizures and bursts of abnormal brain activity.

The additional computing power also means that brain simulations can be run more quickly, which will give researchers insight into relatively slow processes, like learning, Senk says. Researchers will also be able to work out much more detailed biological details, such as more complex models of how neurons change and fire.

But even with the ability to run brain-scale simulations, there’s still a lot we don’t know, Nowotny says. And even simulations of smaller whole brains, like that of the fruit fly, cannot replicate the full behavior of real animals.

Simulations run on these supercomputers are also still very limited and lack basic features essential to real brains, such as the ability to benefit from real-world environments. “We can’t actually build brains,” says Nowotny. “Even if we can simulate the size of a brain, we can’t simulate the brain.”