Simulating the human brain involves using advanced computing power to model billions of neurons, aiming to replicate the intricacies of real brain function. Researchers aspire to enhance brain simulations, uncovering secrets of cognition with enhanced understanding of neuronal wiring.

Historically, researchers have focused on isolating specific brain regions for simulations to elucidate particular functions. However, a comprehensive model encompassing the entire brain has yet to be achieved. As Markus Diesmann from the Jülich Research Center in Germany notes, “This is now changing.”

This shift is largely due to the emergence of state-of-the-art supercomputers, nearing exascale capabilities—performing billions of operations per second. Currently, only four such machines exist, according to the Top 500 list. Diesmann’s team is set to execute extensive brain simulations on one such supercomputer, named JUPITER (Joint Venture Pioneer for Innovative Exascale Research in Germany).

Recently, Diesmann and colleagues demonstrated that a simple model of brain neurons and their synapses, known as a spiking neural network, can be configured to leverage JUPITER’s thousands of GPUs. This scaling can achieve 20 billion neurons and 100 trillion connections, effectively mimicking the human cerebral cortex, the hub of higher brain functions.

These simulations promise more impactful outcomes than previous models of smaller brains such as fruit flies. Recent insights from large language models reveal that larger systems exhibit behaviors unattainable in their smaller counterparts. “We recognize that expansive networks demonstrate qualitatively different capabilities than their reduced size equivalents,” asserts Diesmann. “It’s evident that larger networks offer unique functionalities.”

Thomas Novotny from the University of Sussex emphasizes that downscaling risks omitting crucial characteristics entirely. “Conducting full-scale simulations is vital; without it, we can’t truly replicate reality,” Novotny states.

The model in development at JUPITER is founded on empirical data from limited neuron and synapse experiments in humans. As Johanna Cenk, a collaborator with Diesmann at Sussex, explains, “We have anatomical data constraints coupled with substantial computational power.”

Comprehensive brain simulations could facilitate tests of foundational theories regarding memory formation—an endeavor impractical with miniature models or actual brains. Testing such theories might involve inputting images to observe neural responses and analyze alterations in memory formation with varying brain sizes. Furthermore, this approach could aid in drug testing, such as assessing impacts on a model of epilepsy characterized by abnormal brain activity.

The enhanced computational capabilities enable rapid brain simulations, thereby assisting researchers in understanding gradual processes such as learning, as noted by Senk. Additionally, researchers can devise more intricate biological models detailing neuronal changes and firings.

Nonetheless, despite the ability to simulate vast brain networks, Novotny acknowledges considerable gaps in knowledge. Even simplified whole-brain models for organisms like fruit flies fail to replicate authentic animal behavior.

Simulations run on supercomputers are fundamentally limited, lacking essential features inherent to real brains, such as real-world environmental inputs. “While we can simulate brain size, we cannot fully replicate a functional brain,” warns Novotny.