Microsoft caught some flak today after showing off How Old Do I Look?, an app demonstrating the company’s new face-recognition application programming interface (API). But Microsoft appreciates the feedback. The new Face API underlying the app should become more accurate in the future.
“The age and gender-recognition features are labeled as experimental features,” Ryan Galgon, senior program manager for Microsoft technology and research, told VentureBeat in an interview. “These are ones that we definitely want to improve over time.”
Galgon said his team expected the app to generate a lot of interest — but not this much. It started trending on Twitter, with no shortage of people complaining about incorrect predictions.
But Galgon is right to talk about how the Face API could improve. Any usage of the app, even if it results in inaccurate renderings, is good usage, because Microsoft gets more data to improve its systems. That’s just the nature of deep learning, a type of artificial intelligence that Microsoft is using behind the scenes for the Face API. The more data you use for training purposes, the more accurate a system’s predictions can be.
Other tech companies working with deep learning primarily for use in their own applications or for analytical purposes. But Microsoft is now turning its deep-learning smarts into products for developers to use, beating fellow deep-learning powerhouse Google to it.
Microsoft last year introduced its Azure Machine Learning cloud service with similar goals in mind, and market-leading public cloud Amazon Web Services earlier this month introduced its own cloud service for machine learning.
Galgon couldn’t talk about whether Microsoft had compared its new technology — including APIs for image recognition, speech recognition, and language understanding alongside the one for face recognition — against alternative machine learning services in the world. But that wouldn’t be surprising. Other companies doing deep learning, including Baidu, have matched up their systems with competitors.
If nothing else, researchers at Microsoft care about improving their deep-learning models and then writing academic papers about their achievements. But it certainly couldn’t hurt to one day provide systems that exceed the capabilities of competitors’ implementations — and systems that make for eye-popping demos.
Galgon didn’t seem disturbed by the outcome of today’s demo. He thinks the new APIs could come in handy for tasks like automatic thumbnail creation, the detection of adult content, and even working with Internet-connected devices. And the best part is, you don’t need a Ph.D. to use it.
“The developer doesn’t need to have background in machine learning,” he said.