The (human) brain behind making our devices much more ... well, human
ANDREW Ng is hunched over his smartphone in a pantomime of keypecking, typo-ridden discomfort.
“This is how we do it today,” says the chief scientist for Baidu, China’s largest search engine.
“And this is how we should be doing it.” He sits back in his chair, speaking to no one in particular with his phone placed on the table.
The one-finger typing agony of millions of smartphone users should one day become a thing of the past, he says. All it would take is the creation of a reasonably accurate, pocket-sized electronic version of a human brain.
Ng is an expert in deep learning, a branch of artificial intelligence that focuses on teaching computers how to talk, listen, read and think like us. The area is fast becoming a priority for the world’s biggest technology companies, including Baidu, as it tackles the era of the mobile internet.
“The whole world is switching to mobile devices but no one has created a usable interface to input into the devices,” he says. With the development of artificial intelligence, “soon you’ll be able to order food and just say ‘Can I have some food delivered to my house before I get home?’ out loud. It won’t even feel like technology, it will just be in the background.”
In addition to better voice recognition, artificial intelligence is being talked about for any number of uses, from predicting advertising clicks to recognising faces.
Since joining Baidu last year, Ng has been steadily working to implement this vision. A Briton with Chinese roots, in 2011 he founded Google Brain, the US technology group’s deep learning project, and led it until he joined the Chinese company last year.
Poaching him was regarded as a coup in the technology world. He calls the advanced computers at Baidu’s Sunnyvale, California, lab “rocket engines” whose software can be taught to mimic the functioning of the human mind.
Their “fuel” is data, which he gets from Baidu’s trove of online video and audio output as he works to teach the electronic brain to listen and speak.
The company has an advantage in deep-learning algorithms for speech recognition in that most video and audio in China is accompanied by text — nearly all news clips, television shows and films are close-captioned and almost all are available to Baidu and Iqiyi, its video affiliate.
While a typical academic project uses 2,000 hours of audio data to train voice recognition, says Ng, the troves of data available to China’s version of Google means he can use 100,000 hours.
He declines to specify just how much the extra 98,000 hours improves the accuracy of his project, but insists it is vital.
“A lot of people underestimate the difference between 95% and 99% accuracy. It’s not an ‘incremental’ improvement of 4%; it’s the difference between using it occasionally versus using it all the time,” he says.
Thanks to the strides made in Chinese language voice recognition — a particular challenge because of the number of homonyms and the importance of context — Baidu will soon roll out Deepspeech, a voice recognition software that is similar to Apple’s Siri.
Other Chinese companies, including Alibaba and Tencent, are making advances in artificial intelligence but thanks largely to Ng’s reputation, Baidu is judged by industry experts to be ahead of its domestic peers, alongside US rivals Facebook, Google and IBM.
“Artificial intelligence is an oligopoly,” says Yang Jing, founder of AI Era, an association for the artificial intelligence industry in China. “It’s a game for the titans.”
Baidu already saves $2.7m a day at its data centres by using deeplearning algorithms to predict harddrive malfunctions, and it is using artificial intelligence to optimise the use of advertisements and photos to improve clickthrough rates. It would not say how much it is spending on artificial intelligence development.
But in spite of lofty long-term ambitions, translating deep learning into moneymaking projects is still largely on the horizon.
Ng is undaunted. “There’s no question that it is creating huge economic value; there’s no question that this will continue to create huge advances,” he says. “There is still a huge gap between the way machines learn and the way humans learn.” Financial Times Limited 2015(c)