Artificial Brain Edges Closer to Reality
We're getting closer to making 'wet' brains as well as making computers smarter.
Japanese researchers announced recently that they turned human embryonic stem cells into a three-dimensional structure similar to the cerebellum, the part of the brain responsible for motor movements and receiving information from our senses.
Even though the structure didn't last very long, it's another small step in building an "artificial brain," a challenge that has teams of scientists across the globe working to construct living tissue, silicon circuits and computer algorithms together into something that can perform the same tasks as our own living gray matter.
Some projects are massive. The European Union's decade-long $1.2 billion "Human Brain Project" includes 183 principal investigators from 24 countries whose stated goal is to simulate and map the human brain, develop brain-inspired computer technologies and explore human brain diseases. But last year, the HPB got mired in a dispute between neuroscientists and computer scientists about whether more of the work should be done on powerful supercomputers rather than medical labs. Now, the entire project is undergoing an external review.
Some brain projects are smaller and less ambitious. In La Jolla, Calif., a small startup company is building computer software that would allow you to "train" a robot to perform certain tasks. This artificial brain more closely resembles the cognitive capabilities of a small mammal than a modern human, according to Eugene Izhikevich, a cognitive neuroscientist and the founder and CEO of Brain Corporation.
"If we can build a brain for robots and steal as much as we can from biology," Izhikevich said. "You can train your robot by showing examples of desired behaviors." The Brain Corporation has already gotten a small dog-like robot to obey simple hand gestures (see the video here).
Izhikevich says he's providing the software to trainers so that others can make their robots do similar sorts of tasks. He foresees these simple robots taking over mundane chores like emptying the dishwasher, folding clothes, or picking strawberries, rather than replacing humans at the workplace."It's like training a dog," he said. "It requires certain skill, but you don't need a PhD in robotics. You can spend a day, a week or a month."
Izhikevich says the software "brains" used by these robots allow the device to only do what it's been trained to do, relying on training rather than complex decision-making. "It ends up as a fully autonomous system," he said. "It only does what you show it to do, it doesn't do more.Some roboticists are trying to develop more complex software so that their robots can talk, sort objects or play music. Others are trying to predict what humans want.
Still, these research robots still face obstacles, especially with things that our brains do well, such as recognizing human speech or facial cues. At IBM, a team is developing a computer chip called "True North" that mimics the way brains recognize patterns, relying on densely interconnected webs of transistors similar to the brain's neural networks. This project was announced last November in the journal Science and uses very little energy.
On the other end of the "brain" spectrum, there are teams working on building human brains made up of living tissue, like Robert Caplan's team at Tufts University. Caplan and his colleagues announced last summer that they had created a three-dimensional scaffolding shaped like a jelly roll of functioning rat brain cells that received electrical signals just like ours.
Caplan says his group continues to refine this "wet" brain with the hopes that someday it will be able to serve as a tester for new kinds of drugs for brain diseases such as Alzheimer's, Parkinson's or some forms of epilepsy.
"Ideally we would like to have a laboratory brain system that recapitulates the most devastating diseases," Caplan said this week. "We want to be able to take our existing toolkit of drugs and understand how they work instead of using trial and error."
Even this goal is still a ways off. In fact, both Caplan's team and the Japanese group are having trouble keeping their brain structures surviving for very long.
"We have not yet succeeded in maintenance of the culture for a long period enough to generate mature brain structure," said Keiko Muguruma, a researcher at the RIKEN Center for Developmental Biology and lead author of the recent paper in Cell Reports, in an e-mail to Discovery News. "The cultured cell aggregates (called spheroids) stop to develop at certain time point and start to degenerate. There may be some missing things for complete development."
As to when medical researchers, neuroscientists and computer specialists will develop a real human brain, experts say that is likely decades if not longer away. The biggest challenge lies not just in developing functioning tissue, or making lots of computations on a chip, but developing the connections and signaling that allow our brains to accomplish so many tasks so quickly.
"We are trying to capture what the brain is doing," Izhikevich said, "not what it looks like."
Multiple groups of scientists are working on ways to build an artificial brain.
If we can think it, we can control it.