Researchers have just created a very self-sufficient bicycle.
Thanks to an artificial intelligence chip modeled after the human brain, a new self-driving bike can heed vocal instructions, avoid obstacles in its path, and, perhaps a bit unnervingly, track and follow a person jogging up ahead of it.
The contraption won’t be wheeling itself into your garage anytime soon, though. For the scientists behind the invention, reported Wednesday in the journal Nature, the bicycle’s hat tricks are really more proof of concept than evidence of a new method of transportation.
What’s most exciting, the authors argue, is the AI chip itself, called Tianjic—the product of a team of computer scientists at Tsinghua University in Beijing, China, and the bike’s functional “brain.” Other groups are hard at work on similar chips, and if the research progresses, these tiny processing units will find their way into most machines of the future. The goal, the researchers say, is to help artificial intelligence learn more efficiently and adaptively. In other words, more like we humans do.
Machine learning technology has advanced quickly in recent years, but most devices share a common pitfall: the amount of time, energy, and human input required to get the skills of these systems up to snuff. When artificial intelligence learns, it often does so through brute force, cycling through countless rounds of trial and error until it converges on the best set of tactics.
People, on the other hand, are much better at thinking on their feet, and require much less brainpower to do so. To bridge this processing gap, many independent groups of computer scientists are trying to build computer chips with an internal architecture that mimics that of the human brain.
So-called neuromorphic chips are hybrids. Half of their makeup is standard AI fare, relying on standard computer algorithms. The rest, however, is biologically inspired, incorporating hundreds of thousands of faux neurons that attempt to approximate how human brain cells communicate: through electrical impulses, sent only when an input signal reaches a critical threshold. The two parts of the chip then relay information to each other in a way that’s meant to combine their strengths in the learning process and consume less energy in doing so.
“This is about trying to bridge and unify computer science and neuroscience,” Gordon Wilson, chief executive of Rain Neuromorphics, a start-up company that is developing its own human-brain-inspired chip, told Cade Metz at The New York Times.
Of course, such a system remains a gross oversimplification of an actual human brain—especially considering how little we understand even the version in our own heads. But these chips are at least a small step closer to computationally mimicking the brain’s processing power.
For now, the aforementioned bicycle’s neuromorphic chip remains imperfect. Its abilities, like obstacle detection and voice recognition, are all executed with pre-trained software, which means the bicycle still can’t learn from its experiences in real time. But the chip was able to establish the necessary connection between its computer science and neuroscience halves, and requires less energy than a more typical processor would, John Timmer at Ars Technica reports.
These qualities could make similar chips suitable for other applications, as long as it proves itself capable of other complex tasks, like manipulating objects, Alessandro Oltramari of Bosch, who was not involved in the study, told Donna Lu at New Scientist.
As for the self-driving bike? It’s unclear what the future holds for this wheeled wonder. Don’t feel bad for it, though. Fake brain or no, the bike’s still not sentient; it’s just along for the ride.