The U.S. Government Is Betting $28 Million That We Can Replicate The Brain

The great reverse engineering expedition

First found in boxers and called "punch-drunk syndrome," CTE now shows up in the brains of former NFL players too. Bryan Christie Design

A partial digital reconstruction of the brain previously made by Harvard. Harvard

We’ve talked a lot about making a computer that works like the mammalian brain.

The U.S. government is now betting $28 million dollars that all these projects are wrong.

A series of three grants snagged by Harvard University from Intelligence Advanced Research Projects Activity (IARPA) last week has funded a “moonshot” project to throw out all of the previous attempts at understanding the brain and start fresh. (IARPA is the sibling organization to the better-known Defense Advanced Research Projects Agency, or DARPA. But while DARPA focuses on military projects, IARPA focuses on intelligence agency research.)

The 5-year project will try to build a virtual model of part of a rat brain, as the brain learns. From this, the lab hopes to be able to identify the way organic neurons work together to learn, so they can replicate it artificially. Ultimately, the hope is to replicate the human brain itself.

“It’s a reverse engineering expedition,” said David Cox, professor at Harvard and leader of the project. “If your competitor released a product and it was way better than your product, you might buy the product, open it up, and try to figure out how it works. Nature is the competing company in this scenario.”

The basic methodology of the project is to show live rats a series of images while examining their brains under a microscope. This information is later correlated to a virtual representation of the brain, made when the rats’ brains are cut into small sections and imaged under an electron microscope.

Harvard says the whole project will produce more than a petabyte of information, which will be made publicly available. (A petabyte is equivalent to an mp3 audio track 2000 years long.)

Having all this data will be great, but one thing we do know already is that rodent brains work differently than human brains.

Acknowledging that, Cox thinks that methods used to virtually replicate the brain in this project could be valuable in their later application to humans. Replicating rat brains is sort of like using training wheels, while replicating human brains someday would be the end goal. So in a sense, this project might be a first step to neural maps being used in regular human treatment, albeit far into the future.

With electron microscopy, the Harvard teams will be able to reconstruct the “wiring diagrams” of the brain, Cox says. They’ll be able to see in realtime which neurons are working with others, and how the connections are shaped.

Understanding the brain is a huge step in neuroscience, but this project’s main thrust is actually aimed at computer science. Modeling the brain is only the first step. After (and potentially during) creating that model, researchers are tasked with creating algorithms that replicate the way that the brain processes information. That is the task that artificial intelligence researchers have been grappling with for more than 50 years. Right now, most A.I. research heavily relies on statistics, rather than biology.

“We know that brains are really good at things like learning and inference, we don’t yet have algorithms that can match those abilities,” Cox said. “But we’re not completely on a limb without an idea. We know the shape of the book, but we need to know the words inside.”

However, Cox says that most researchers, like those working with DARPA’s SyNapse project and Europe’s Human Brain Project, are plunging into creating models of the brain without first understanding how they work. The SyNapse project, which is largely pioneered by IBM, has already created chips based on previous knowledge of the brain, and the Human Brain Project is also building models on “existing data,” according to their research goals.

Meanwhile, the IARPA project will span five years, over three stages. Each stage will study a larger portion of the brain, with the largest being just a one millimeter cube.