A multitasking supercomputer that attempts to mimic the human brain was switched on Nov. 2—and it could be used to help virtual assistants like Apple’s Siri and Amazon’s Alexa understand your accent.
The SpiNNaker computer can send billions of small pieces of information to different destinations simultaneously, much like the brain. It’s not quite like a regular computer, which sends one larger packet of data to a single destination at a time.
It’s the result of a £15 million (more than $19 million) project first conceived in England 20 years ago. It has spent a decade under construction. Its technology could be used to improve tools like the virtual assistants, its project leader told CNN.
If applied to commercial devices such as smartphones, that could be good news for people who find themselves shouting the same command at them over and over again.
The technology could be used to reduce the energy needed to run Siri or Alexa, “so instead of being extensively trained on huge data sets and then put into service as a fixed network, they could continue to learn ‘on the job’ and adapt to the user’s habits and requirements,” University of Manchester computer science Professor Steve Furber explained.
“One form of adaptation is adjusting to the speaker’s accent,” he added. “I know there are people who struggle with Siri because their accents are outside its capabilities.”
The computer, which was switched on Nov. 2, features a million-processor core, its creators said.
But it still can’t quite outsmart a human, they noted.
“Even with a million processors we barely approach 1% of the scale of the human brain,” said Furber. “We can model networks with about the complexity of a mouse brain.”
Nonetheless, the scientists behind SpiNNaker have already harnessed its power to create a robot—the SpOmnibot—and a second-generation model is already in development.
