On Monday, IBM announced an agreement to acquire Confluent, a data streaming platform, in an all-cash deal that would value the company at $11 billion. It’s the latest attempt by IBM, which risks falling behind relatively recent entrants in the A.I. race, to strengthen its hand: IBM previously acquired software maker Apptio for $4.6 billion in 2023, and cloud software company HashiCorp for $6.4 billion in 2024. Shares of Confluent surged nearly 30 percent on Monday following the announcement, getting right up to the premium ($31 per share) that IBM paid.

And now for the main event…

Everyone is entitled to their own version of happiness, and Dr. Anton Arkhipov has spent the past 12 years at the Allen Institute working to digitally replicate tens of thousands of miles of mouse brain circuitry. Last month, his team successfully created a digital simulation of a mouse brain cortex, which the Allen Institute heralded as “one of the largest and most detailed simulations” of an animal brain ever. Arkhipov described the achievement as a major milestone along a “very long and complicated path toward being able to simulate the brain in all its glorious complexity.” (The work builds on Europe’s Blue Brain Project.)

The researchers began by collecting a “huge amount” of physiological data, Arkhipov told me, and integrating it with their brain-modeling and neuron -simulating software on one of the most powerful supercomputers in the world, Japan’s Fugaku. The entire mouse brain has around 70 million neurons; the cortex that the Allen Institute is simulating has around 10 million. (Full human brains have roughly 86 billion neurons; human cortices have around 21 billion neurons.)

The point of this endeavor, Arkhipov explained, is to better understand the ways in which brain cells actively communicate with one another. He described the success as barely “scratching the surface,” noting that while it provides a proof of concept that larger-scale models are possible, there’s still a lot more work needed to get there. “None of this … is the end,” he said. “It’s more like the beginning.” The promise of brain simulations is that we can use them to conduct the kinds of research and experiments that are difficult or even impossible to conduct physically. Ultimately, scientists hope to learn what happens in the brain before and during the advancement of certain neurological disorders, so they can safely test various interventions.

But according to a paper detailing the project, the researchers also have an additional goal in mind: Better models of the brain might offer a road map to building more efficient A.I. algorithms. After all, one great distinction between artificial intelligence and the real thing is that the human brain operates on around 20 watts of power. For its part, A.I. requires gigawatts upon gigawatts to even shallowly mimic what a teenager can do with the caloric power of a Kit Kat.

Much of the promise here is fuzzy. After all, while the human brain resembles a computer in some very superficial ways—both rely on electrical impulses, and, sure, neurons have on and off states—it is also endlessly more complicated, with brain activity affected by innumerable genetic, chemical, and enzymatic factors. But even giant, full-scale brain simulations are not the end-all, be-all of computational neuroscience, according to Dr. Omar Ahmed, a professor of neuroscience at the University of Michigan.

The trouble, he told me, is that all those elements that we haven’t yet learned about the brain obviously can’t go into a brain simulation. “This is a great snapshot of what we already understand,” he said. “What is under the hood and unknown is still unknown and not being simulated.” Because of this, a simulation might produce the same recorded outputs of a real brain, but for a different reason. The scale of the unknowns at play here is such that it’s almost impossible to tell whether a simulation is offering a truly novel insight or simply approximating an expected outcome.

When I asked Arkhipov about this dynamic, he offered me a quote from the physicist Richard Feynman: “What I cannot create, I do not understand. That’s exactly what we’re trying to do,” he said. “You probably cannot claim you understand something unless you can build a relatively accurate model of it.” Still, he acknowledged, such models could themselves become so complex that they surpass the limits of the human mind—which, at least as far as understanding goes, gets us right back to where we started.

Ahmed said that the work the Allen Institute is doing, specifically in brain data collection, is incredibly important in advancing the field. But he reiterated that these simulations are just one more tool in a very large, complex toolbox—one that also includes deep learning algorithms, which are increasingly being used to simulate neural behavior and model brain activity. “I think there’s amazing progress being made, but these are very challenging questions, and there’s not one solution set,” Ahmed said. “Why is the brain the final frontier on Earth? It’s that ridiculously complicated, and we need to throw the kitchen sink and keep throwing the kitchen sink. Then one day, perhaps, we’ll be ready to actually make sense of the output of a massive-scale computational simulation. Are we there yet? No, but I think we just have to keep going along the same trajectory.”

It’s work that he expects must continue in a similar fashion for the next several decades, if not longer. “Right now, we just don’t know all the beautiful algorithms that the brain uses that are, frankly, far more optimal for speedy, flexible computation than any A.I. currently is,” he added. “We know a lot, but not everything, and these models are still incomplete.”

That’s all for today. I’ll see you on Thursday.

Ian

Need help? Review our FAQ page or contact us for assistance. For brand partnerships, email ads@puck.news.

You received this email because you signed up to receive emails from Puck, or as part of your Puck account associated with {{customer.email}}. To stop receiving this newsletter and/or manage all your email preferences, click here.

Puck is published by Heat Media LLC. 107 Greenwich St., New York, NY 10006