With AskADoctor, the computer voice translation couples with another deep learning model that ropes in the health data owned and scrapped by Baidu across the Chinese Web. Wei said the product can assess 520 different diseases, representing upwards of 90 percent of the most common medical problems nationwide.
The team is also betting big on voice, a field where it may advance more in China than other rivals. Since February, Baidu’s deep learning stateside team, around 40 researchers, has worked on building artificial neural networks to process Mandarin. The technique allows machines to render the language — a complicated one, as it’s tonal and character-rich — with far more computing power.
Asked what sets Baidu’s AI division apart, Ng referenced size. Baidu is investing heavily in A.I. hardware — it has 16 graphics processing units training speech models — something that Ng may not have had at Google, which tends to favor a more dispensable approach to hardware. “I’m pretty confident we’re building supercomputers that let us scale with these deep learning algorithms bigger, faster than anyone else,” he said.