I’m wearing a piece of 3-D-printed plastic headgear that looks like a bicycle helmet designed by Buckminster Fuller. Tiny metal pins inside it poke lightly into my scalp. On a screen in front of me are the electroencephalogram (EEG) readouts of signals picked up by the metal pins. And beyond the monitor is a wall of windows giving this dilapidated Brooklyn office building the sweetest view of a Manhattan sunset I have ever seen.
That lovely view makes it easier when Conor Russomanno, a self-described neurohacker, asks me to close my eyes and relax. After a few seconds, he tells me later, the screen showed a slight spike at around 10 Hz–a rise in the alpha waves that indicates a restful state. Russomanno seems as pleased with the electrical feedback as he is with my verbal feedback (when I tell him the headgear doesn’t hurt). This was his latest, but still not final, version of the Ultracortex–a low-cost, research-grade EEG headset set to hit Kickstarter in the fall. It will allow fellow hackers to start peering inside the workings of their own brains.
It’s also one of the tools for crowdsourcing EEG data to a repository called Cloudbrain, where artificial intelligence machine-learning algorithms will scour the raw data for patterns that might help explain how we speak or perceive color or allow our minds to directly control machinery like motorized prosthetics. They might even expand understanding of mental illness and cognitive impairments.
These DIY brain scientists, or neurohackers, aren’t sure how deep they will be able to go; but they are excited that they finally have cheap tools to start looking. That’s what brought about half a dozen of them to this Brooklyn lab for the NeuroTechNYC meetup. Together with groups in San Francisco, Montreal, and Toronto, they form NeuroTechX, a new international collaboration of researchers and inventors (some still in grad school) building an open-source project to investigate the mysteries of the mind. Since it was formed two months ago, NeuroTechX has already drawn about 1,000 members.
None of this would have been possible a few years ago, before the advent of low-cost, easy-to-set-up EEG gear and sophisticated cloud-based AI resources. Budget EEG headsets started emerging about six years ago with toys such as Mattel’s Mindflex in 2009. The simple headset, made by a company called NeuroSky, could measure someone’s general level of concentration, which allowed them to move a ball through a maze. NeuroSky’s device and others that followed use “dry” electrodes that don’t require the traditional spaghetti of wires and electrodes that attach to the scalp with a conductive gel, like William Hurt wore in the 1980 film Altered States.
The hottest new headset, the $300 Emotiv Insight, just started shipping to Kickstarter backers. Morgan Collino, an engineer from France interning in New York, brought his new Insight to the meetup–eliciting a jealous grumble from Russomanno, who still hasn’t gotten his. The Emotiv is amazing for what it does at that price, but it has several limitations. The Insight has just five electrodes; the Ultracortex can sport up to 16. The Insight samples data 128 times per second (per electrode), and some consumer headsets sample even less frequently. The Ultracortex samples up to 250 times per second, making it a research-grade device. The Insight automatically translates the data into a limited range of readings indicating levels of attention, focus, engagement, interest, or stress, for example. The Ultracortex outputs the full torrent of raw data gathered from the brain.
The Ultracortex’s power and flexibility comes from OpenBCI, the product of a grant program from the U.S. military’s Defense Advanced Research Projects Agency (known as DARPA, the ones who really did invent the Internet). As part of President Obama’s BRAIN Initiative for neurological research, DARPA funded the development of a low-cost, high-resolution EEG system, with the requirement that it must be open source and must be productized.
Joel Murphy, a former professor at Parsons School of Design in Manhattan, worked on the hardware. Murphy brought in two of his students, Russomanno and Leif Percifield, to help with the firmware and interface. “All of us went to art school,” said Percifield, a former EMT, firefighter, and photographer who taught himself programming and hardware development. The three self-taught technologists developed OpenBCI (brain-computer interface). The heart of the system is the $450 OpenBCI bio-sensing kit, which gathers and processes data from the electrodes.
Russomanno had always seen OpenBCI as a tool to start collecting and analyzing data. “I want to build an open repository,” says Russomanno, “kinda upend or undermine the whole privatizing and capitalizing on peoples’ user-generated data.”
He’s not the only one with that idea. Though the Brooklyn meeting was all guys, several spoke reverently about Marion Le Borgne, a senior software engineer at Numenta, a Bay Area company building machine-learning technology modeled on the human brain’s neocortex. Previously, she worked at CloudWeaver (now part of F5 Networks) developing analytics for cloud-computing infrastructure. “At some point, I had this idea,” says Le Borgne. “If we can get this sensor data from computers, maybe we can get it from humans.” She then started Cloudbrain as a side project for gathering and analyzing data from all types of wearable health sensors, but began with EEG data, in part because of the challenge. “I like to start with the hard stuff, because if you can handle this one, you can handle the rest,” she says.
Le Borgne first demoed Cloudbrain in February 2015 at the Cognitive Technology exhibit at the Exploratorium science museum in San Francisco. Visitors donned headsets that allowed them to control a robotic arm or project patterns derived from their brainwaves. “That was one of the big stress tests for Cloudbrain,” she says. The setup collected more than 15 million data points per day, and visitors who signed consent forms donated their data to NeuroTechX for further research.
Le Borgne and her team are still working on the infrastructure to handle a large amount of data, and experimenting with algorithms to analyze it. Through NeuroTechX, they now have funding to host a centralized Cloudbrain, and will start taking mass data submissions in a few months, she says. For now, they have put up a demonstration of Cloudbrain receiving data from a recorded sample.
The main task is developing classifiers. For example: What’s the EEG pattern for moving your right hand? Sometimes that requires volunteers submitting brainwave data while they are performing a particular exercise meant to elicit a particular thought or action.
Another technique–and the most intriguing–is bulk analysis of EEG streams, without knowing what the people submitting it are doing or thinking. With enough data and powerful artificial intelligence analysis, the system starts to find chunks of brainwave activity that are common to all people. Essentially, it starts to pick out the individual words and phrases used in the language of the brain.