AlphaGo on a smartphone? Chinese brothers plan to put AI in your pocket with start-up based on Cambrian deep-learning processor
Two scientists from Chinese Academy of Sciences expect Cambrian to beat smartphone vendor Xiaomi’s record valuation of US$45 billion, aim to change the world by commercialising new AI technology
A team of researchers at the Chinese Academy of Sciences expect to set a new valuation record in China with their start-up based on Cambrian, a deep-learning processor they invented that may take artificial intelligence to the next level.
The team, led by 32-year old Chen Yunji and his younger brother Chen Tianshi, who both teach at the academy’s Institute of Computing Technology in Beijing, said they are now wrapping up the first round of angel investment in their start-up, which is named after the “revolutionary” processor.
READ MORE: Chinese start-up on track to deliver artificial intelligence-on-a-chip
Their goal is to bring human-like AI to people’s mobile devices. Such computing power is at present only available on supercomputers.
“We are starting a company called Cambrian to commercialise our deep-learning processor,” Chen Yunji told guancha.cn, a Shanghai-based news website, over the weekend.
Cambrian refers to a geological period 500 million years ago when life on earth changed dramatically as the number of multicellular organisms exploded, paving the way for incredible diversity.
“The company will soon complete its first round of angel investment, and its valuation may become the highest in history for a Chinese company,” the older Chen was quoted as saying.
After five years or more, I think every cell phone can be as powerful as Google Brain
He declined to provide the exact amount of funding they have received.
Xiaomi, the Chinese smartphone maker and manufacturer of a range of consumer electronics and smart products that is shaking off its image as an “Apple clone”, raised US$1.1 billion in 2014 for a record valuation among start-ups in the country of US$45 billion.
But with momentum building to further develop AI, the Chens believe they can eclipse this figure.
WATCH: Davos 2016 - the state of artificial intelligence
“After five years or more, I think every cell phone can be as powerful as Google Brain,” Chen Yunji told MIT Technology Review last year. The journal named him one of the world’s top innovators under the age 35.
READ MORE: Google’s AlphaGo might have bested the world Go champ - but Chinese netizens say it’s not smart enough to win at mahjong
The AI movement got another shot in the arm last week when AlphaGo, a supercomputer built by Google engineers, beat Go grandmaster Lee Se-dol of South Korea in the first three games of their historic best-of-five tournament in Seoul.
This makes today’s final match in the Korean capital a dead rubber as AlphaGo has already claimed the US$1 million prize despite Lee claiming some bragging rights on Sunday with his first victory.
Winning the tournament is considered a huge milestone in computing because Go, a board game where players use black and white tiles to claim the other’s territory, is vastly more complicated than chess. The number of possible moves has been compared to the number of atoms in the universe. AlphaGo was built by DeepMind, a London-based Google AI lab.
READ MORE: End of the road for journalists? Tencent’s Robot reporter ‘Dreamwriter’ churns out perfect 1,000-word news story - in 60 seconds
But are the Chens developing something that could one day be as powerful as AlphaGo yet fit in your pocket courtesy of a small chip inserted in your smartphone?
“I think they have embarked on a journey with a bright future,” said Feng Jufu, a professor of AI at the Key Laboratory of Machine Perception of Peking University who has been following their progress.
“The new technology has the potential to become a huge commercial success,” he added.
Their chip and new set of instructions [DianNaoYu] is completely different from anything we have seen before. If the technology works, it will start a revolution
Meanwhile, tech giants like Google and Baidu, China’s search engine king, have been developing AI projects of their own on existing computer architecture. Some, like IBM, are even building new chips inspired by the structure of the brain.
But there is much debate within the AI community as to whether existing hardware and software can be upscaled to reach the levels of AI that are being sought, or whether scientists and computer engineers must go back to the drawing board and start over.
The Chen brothers are firmly in the second camp, and they have already created a buzz among the AI-dedicated research community.
READ MORE: Hong Kong artificial intelligence pioneer on immortality, the age of robots and manufacturing evil
A paper they submitted to the ACM/IEEE International Symposium on Computer Architecture in Seoul this summer was given the highest score by peer-reviews among nearly 300 works submitted to the world’s biggest academic conference on computer chips.
Their paper focussed on DianNaoYu, a new machine language they developed for deep learning on their AI processor. The name translates directly from Putonghua as “electric brain/computer language”.
Li Jianmin, an associate professor of AI at Tsinghua University in Beijing, said the Cambrian processor could spark a revolution in the computing industry.
“Their chip and new set of instructions [DianNaoYu] is completely different from anything we have seen before,” he said.
WATCH: Match 4 of the Google DeepMind Challenge - Lee Sedol vs AlphaGo
“If the technology works, it will start a revolution,” he added.
Many AI systems developed by governments or multinationals in recent decades have tried to mimic the operation of the human brain by making calculations via a network of artificial neurons on pre-existing x86 architecture.
READ MORE: ‘Soft robots’ and ‘mobile secretaries’ coming soon, says Hong Kong university working with China’s WeChat on artificial intelligence projects
On the whole, these have proven strong at performing commands but weak at “thinking” independently. For example, they need to receive and process thousands of instructions to conduct just one calculation on an artificial neuron.
In contrast, one instruction written in DianNaoYu on a Cambrian chip can command and operate a cluster of neurons, according to the Chens. This significantly accelerates the processing speed of an AI system and reduces the chip’s energy consumption, paving the way for mobile applications, they said.
Other issues to overcome include hardware requirements and privacy fears.
READ MORE: Workers are ready for the robot revolution, but are their managers?
Most of the bigger AI systems require massive server networks to operate, meaning that individual users would only realistically be able to use them via the cloud. As such, they would have to entrust others with their personal data.
But the technology being developed by the Chen brothers could liberate AI technology from the hands of big players, Li said.
Others are more sceptical whether this will ever see the light of day.
Professor Zhu Xiaoyan, a director at Tsinghua’s State Key Laboratory on Intelligent Technology and System, said the Chens still have many hurdles to clear if they hope to commercialise Cambrian.
One of those is keeping pace with the competition presented by quantum computers, said Zhu. Such computers could potentially prove billions of times faster than the most powerful computers today and more closely approximate the workings of the human brain, he said.
Qian Hui, an associate professor of AI at Zhejiang University in East China, pointed to Japan as an example of how such ambitions as those being bandied about by the Chens can go up in smoke.
Japan spent a decade trying to take computing to the next level with its so-called “fifth-generation” computer systems designed for AI, but the country ultimately fell behind because it underestimated the rapid pace at which the processing speeds of conventional computers would increase, Qian said.
“China could succeed this time, but I am afraid the probability is very low,” he said.
