Image
By Cade Metz
REDWOOD CITY, Calif. — In the global race to build artificial intelligence, it was a missed opportunity.
Jeff Hawkins, a Silicon Valley veteran who spent the last decade exploring the mysteries of the human brain, arranged a meeting with DeepMind, the world’s leading A.I. lab.
Scientists at DeepMind, which is owned by Google’s parent company, Alphabet, want to build machines that can do anything the brain can do. Mr. Hawkins runs a little company with one goal: figure out how the brain works and then reverse engineer it.
The meeting, set for April at DeepMind’s offices in London, never happened. DeepMind employs hundreds of A.I. researchers along with a team of seasoned neuroscientists. But when Mr. Hawkins chatted with Demis Hassabis, one of the founders of DeepMind, before his visit, they agreed that almost no one at the London lab would understand his work.
Mr. Hawkins says that before the world can build artificial intelligence, it must explain human intelligence so it can create machines that genuinely work like the brain. “You do not have to emulate the entire brain,” he said. “But you do have to understand how the brain works and emulate the important parts.”
At his company, called Numenta, that is what he hopes to do. Mr. Hawkins, 61, began his career as an engineer, created two classic mobile computer companies, Palm and Handspring, and taught himself neuroscience along the way.
Now, after more than a decade of quiet work at Numenta, he thinks he and a handful of researchers working with him are well on their way to cracking the problem.On Monday, at a conference in the Netherlands, he is expected to unveil their latest research, which he says explains the inner workings of cortical columns, a basic building block of brain function.
How a larger community of researchers react to Mr. Hawkins’s work is hard to predict: Will they decide his research is worth exploring? Or will they write him off as too unorthodox in his methods and much too sure of himself?
Mr. Hawkins has been following his own, all-encompassing idea for how the brain works. It is a step beyond the projects of most neuroscientists, like understanding the brain of a fruit fly or exploring the particulars of human sight.
His theory starts with cortical columns. Cortical columns are a crucial part of the neocortex, the part of the brain that handles sight, hearing, language and reason. Neuroscientists don’t agree on how the neocortex works.
Mr. Hawkins says cortical columns handle every task in the same way, a sort of computer algorithm that is repeated over and over again. It is a logical approach to the brain for a man who spent decades building new kinds of computing devices.
All he has to do is figure out the algorithm.
A number of neuroscientists like the idea, and some are pursuing similar ideas. They also praise Mr. Hawkins for his willingness to think so broadly. Being a maverick is not easily done in academia and the world of traditional research. But it’s a little easier when you can fund your own work, as Mr. Hawkins has.
Still, some wonder if his self-funded operation, isolated from the rigors of academic interaction, is a quixotic adventure. They have been researching the brain one little piece at a time for a good reason: Piecing how it all works together is a monumental, hard-to-fathom task.
“It is clear we need a better understanding of intelligence,” said Tomaso Poggio, a neuroscientist at the Massachusetts Institute of Technology who introduced Mr. Hawkins and Mr. Hassabis. “But Jeff is doing this the hard way.”
If Mr. Hawkins’s work should pan out, it could help A.I. researchers leapfrog over what exists today. In recent years, the likes of Google, Apple and Amazon have built cars that drive on their own, gadgets that answer questions from across the room and smartphone apps that instantly translate languages.
They relied on “neural networks,” which are mathematical systems modeled after the web of neurons in the brain — to a point. Scientists cannot recreate the brain because they understand only pieces of how it works. And they certainly can’t duplicate its capabilities.
“The brain is by far the most complex piece of highly excitable matter in the known universe by any measure,” said Christof Koch, the chief scientist and president of the Allen Institute for Brain Science. “We don’t even understand the brain of a worm.”
A call to explain the brain
In 1979, with an article in Scientific American, Francis Crick, a Nobel Prize winner for his DNA research, called for an all-encompassing theory of the brain, something that could explain this “profoundly mysterious” organ.
Mr. Hawkins graduated from Cornell in 1979 with a degree in electrical engineering. Over the next several years, he worked at Intel, the computer chip giant, and Grid Systems, an early laptop company. But after reading that magazine article, he decided the brain would be his life’s work.
He proposed a neuroscience lab inside Intel. After the idea was rejected, he enrolled at the University of California, Berkeley. His doctoral thesis proposal was rejected, too. He was, suffice to say, an outlier.
In 1992, Mr. Hawkins founded Palm Computing. A decade and a half before the iPhone, he had created a hand-held computer for the masses. When he hired the company’s chief executive, Donna Dubinsky, he warned that whenever possible, he would drop his work with Palm and return to neuroscience. “That was always there, simmering in the background,” Ms. Dubinsky said.
U.S. Robotics acquired Palm in 1996 for $44 million. About two years later, Mr. Hawkins and Ms. Dubinksy left to start Handspring. Palm, which became an independent company again in 2000, acquired Handspring for $192 million in stock in 2003.
Around the time of the second sale, Mr. Hawkins built his own neuroscience lab. But it was short-lived. He could not get a lab full of academics focused on his neocortical theory. So, along with Ms. Dubinsky and an A.I. researcher named Dileep George, he founded Numenta.
The company spent years trying to build and sell software, but eventually, after Mr. George left, it settled into a single project. Funded mostly by Mr. Hawkins — he won’t say how much he has spent on it — the company’s sole purpose has been explaining the neocortex and then reverse engineering it.
Image
A coffee cup of clarity
Inside Numenta, Mr. Hawkins sits in a small office. Five other neuroscientists, mostly self-taught, work in a single room outside his door.
Mr. Hawkins said a moment of clarity came about two and a half years ago, while he was sitting in his office, staring at a coffee cup.
He touched the cup and dragged his finger across the rim. Then he leapt to his feet and ran through the door.
He ran headlong into his wife, who had stopped by for lunch, and stumbled toward his closest collaborator, Subutai Ahmad, the vice president of research. “The cortex knows the location of everything,” Mr. Hawkins said. Mr. Ahmad had no idea what he was talking about.
As Mr. Hawkins looked at that cup, he decided that cortical columns did not just capture sensations. They captured the location of those sensations. They captured the world in three dimensions rather than two. Everything was seen in relation to what was around it.
If cortical columns handle sight and touch in this way, Mr. Hawkins thought, they handle hearing, language and even math in similar ways. He’s been working on proving that ever since.
“When the brain builds a model of the world, everything has a location relative to everything else,” Mr. Hawkins said. “That is how it understands everything.”
The source of tension between Mr. Hawkins and other brain and A.I. researchers is not that they necessarily think he is wrong. It’s that they simply don’t know because what he has been trying to do has been so different. And so wildly ambitious.
For the science to advance, what Mr. Hawkins has been working on can’t stay in a silo. His ideas could benefit from extensive experimentation with other neuroscientists, said Nelson Spruston, a senior director at the Janelia Research Campus, a research lab in Virginia that focuses on neuroscience. “A continuous cycle of testing and revising biologically inspired models of neural computation is the key to developing insightful theories of the brain,” he said.
Translation: Mr. Hawkins will have to open his work to rigorous scrutiny and find a way to interact with researchers who most likely have never looked at the brain the way he does.
NYT
Follow Cade Metz on Twitter: @CadeMetz
Posts
No comments:
Post a Comment