There was just one problem: Nobody knew what the connectome looked like. MRI scans can capture the entire brain, but they can get down to a resolution of only a few cubic millimeters, not nearly fine enough. Other methods, such as staining, allow scientists to look at one neuron at a time but not to track the broader links between them. Seung needed a way to see every neuron in a given piece of brain tissue.
He found the way by teaming up with scientists who know how to slice brains very, very thinly. Jeff Lichtman, a neuroscientist at Harvard University, has built a device called an ultramicrotome that uses a diamond knife to shave brain tissue layer by layer. The ultramicrotome then affixes each slice to a tape that gets fed into a microscope. The microscope takes a high-resolution picture that Lichtman and his colleagues send to Seung’s lab.
Each image is a few hundredths of an inch across, but it’s packed with neural cross sections. Stacking hundreds of thousands of layers against one another can create a three-dimensional model of a piece of the brain. To render a final model, Berger and others inspect every image by eye, coloring each individual neuron in its own specific shade. Once finished, a layer is aligned to the next one, with the same neurons colored the same shades.
Such images contain details far beyond what had ever been seen before. But mapping a few neurons does not give Seung enough data to go after the really big questions about the brain. “There’s a critical scale we have to get through for this to be valuable at all,” he says. That scale, Seung suspects, is about one cubic millimeter: a pinch of brain tissue that could sit atop the period at the end of this sentence. A hundred thousand neurons can fit inside that volume, making about a billion connections with each other. Mapping the connectome inside a cubic millimeter of brain tissue would give Seung enough data to make statistically significant observations about how neurons are connected.
If memories really are encoded in cell assemblies the way Hebb claimed, Seung should be able to observe those assemblies (pdf). He will, in effect, be able to see memories in the brain’s cells.
Painting the neurons by hand, as Berger did, is far too time-consuming, so Seung accelerated the mapping by programming computers to scan brain layers and automatically recognize the neurons in them (pdf). Computers also make lots of mistakes, but those errors leap out when humans look at the models. Instead of painters, Seung and colleagues have become proofreaders.
The collaboration between humans and computers has sped the process along. Instead of weeks, it now takes only a matter of days to map an individual neuron. If Seung can find ways to speed things up further, he may be able to reach his 100,000-neuron mark in a few years.
It would be simple enough, he suggests, for neurosurgeons to set aside a smidgen of tissue removed during human surgery. Then Seung could see if the patterns of connections are different in the brains of healthy people and those with autism, schizophrenia, and other disorders. These connectopathies, or wiring defects, might cause the brain to malfunction as it processes information, shuttling signals from neuron to neuron.
Seung expects that when scientists scale up to a full human brain, in the next few decades, they’ll finally be able to investigate the deep workings of the mind and address questions that, for now, belong more to the realm of philosophy. “Is it true that you are your connectome?” Seung asks. One way to find out would be to model a person’s brain in a computer. If it were detailed enough, it might be able to function like the original brain. Would it contain the same memories stored in cell assemblies? Could you upload yourself to a computer to escape death?
Seung thinks these questions can be tested, but he doesn’t dwell on them. Instead he is focusing on his millimeter-map milestone. To hit that mark, he will need bigger, faster computers. Shaving a cubic millimeter of brain tissue would yield a petabyte of data. That’s a quadrillion bytes—enough to store more than 50 years’ worth of high-definition movies. If the cost of computer memory continues to fall, Seung should be able to store that much data in a few years.
But better technology alone won’t get Seung to his connectome. “The challenge is analyzing them,” he says. Toward that end, he’ll need a lot more eyeballs. The more people who can proofread his connectomes, the faster his maps will grow. As a result, he and his colleagues have set up a website where the public can pitch in.
“We’re trying to gamify it,” Seung says. He may end up being a computer game designer after all.