IBM Tests Mobile Computing Pioneer’s Controversial Brain Algorithms
IBM is testing a contentious idea for making computers more intelligent by trying to copy mechanisms from the human brain.
Computers capable of learning more like a person would have many valuable applications.
For more than a decade Jeff Hawkins, founder of mobile computing company Palm, has dedicated his time and fortune to a theory meant to explain the workings of the human brain, and provide a blueprint for a powerful new kind of artificial intelligence software. But Hawkins’s company, Numenta, has made little impact on the tech industry, even as machine learning has become central to companies such as Google.
Now, one tech giant is finally taking an interest.
IBM has established a research group to work on Numenta’s learning algorithms at its Almaden research lab in San Jose, California. The algorithms are being tested for tasks including interpreting satellite imagery, and the group is working on designs for computers that would implement Hawkins’s ideas in hardware. Hawkins says that around 100 people are working on the project, known internally as the Cortical Learning Center.
IBM would not make the project’s leader, Winfried Wilcke, available for an interview. But Wilcke described his work publicly at a conference at Sandia National Lab in February. He praised Numenta’s software for being closer to biological reality than other machine learning software, and said it can learn how to make sense of raw data more efficiently. Experts usually have to train machine learning software with example data before it can go to work. Numenta’s algorithms might make it possible to apply machine learning to many more problems, Wilcke said.
Machine learning is widely used by Google and other computing companies for various tasks, from categorizing images to processing spoken phrases. Many researchers have come to focus on a technique called deep learning, which trains multi-layered networks of artificial neurons to find patterns in data (see “10 Breakthrough Technologies 2013: Deep Learning”). The results have been striking, but deep learning does not mimic biology closely.
Numenta’s algorithms also operate in a network, but they are aimed at faithfully recreating the behavior of repeating circuits of roughly 100 neurons found in the outer layer of the brain called the neocortex.
“Our goal is not to be biologically inspired; I want to re-create exactly,” says Hawkins. He believes that the brain’s ability to make sense of the world is rooted in these repeating circuits, and that mimicking them in software will make machine learning software capable of much more. “This is how you would really build a machine intelligence,” he says.
In his Sandia talk, Wilkce said Numenta had struck a balance between taking cues from biology and making software that is practical. “It seems to hit a sweet spot,” said Wilcke. “It’s not oversimplified, and not so complicated that there is little chance to ever build a large scale model.”
The IBM group is working on using Numenta’s algorithms to analyze satellite imagery of crops, and to spot early warnings signs of mechanical failures in data from pumps or other machinery. Wilcke also described plans for a novel computer that is a kind of physical re-creation of Numenta’s algorithms.
The plan calls for stacking multiple silicon wafers on top of one another, with physical connections running between them to mimic the networks described by Numenta’s algorithms.
Some computer scientists and neuroscientists are critical of Hawkins’s ideas, saying they have not lived up to his claims. Gary Marcus, a professor of psychology at New York University and a co-founder of an AI startup called Geometric Intelligence, says Numenta’s models are arguably closer to how the brain operates than artificial neural networks. “But they, too, are oversimplified,” he says. “And so far I have not seen a knock-down argument that they yield better performance in any major challenge area.”
Marcus says Hawkins’s algorithms mimic only some of the known mechanisms at work in the brain, and that the majority of its function still remains a mystery. Demonstrations of Numenta’s technology have so far been limited, he adds. “I haven’t seen them try to handle natural language understanding or even produce state-of-the-art results in image recognition,” he says.
Although Hawkins points to the fact that IBM has picked up his ideas as proof of their merit, he seems in no particular hurry to see them make a mark on the world. He has retreated from an earlier plan to make money by marketing Numenta’s first product, software launched in late 2013, called Grok, that looks for anomalies in logs produced by software hosted in the cloud. Hawkins says that software will soon be released for free.
Instead, Numenta’s staff of roughly 20 is now focused on perfecting the algorithms built from Hawkins’s original theory. Getting the software to be capable of learning how to control motors and other physical equipment is a major focus. That could be useful for robotics—one day. “We’re very fortunate that—because of myself and other investors—we don’t have to build a business around this right now,” Hawkins says. “We think we’re building an intellectual property base for the next 30 years of computing.”
Keep up with the latest in machine learning at EmTech MIT.
Discover where tech, business, and culture converge.
September 11-14, 2018
MIT Media Lab