- Google thinks smartphones will soon be powerful enough to do certain types of AI tasks that previously had to be done on powerful computers in data centers.
- It's releasing tools to help app makers take advantage of these capabilities.
- The end result: more powerful Android apps that can do more useful things.
Alphabet's Google has been enhancing its own mobile apps with artificial intelligence (AI).
Now Google is trying to make it easier for other companies to use AI in Android apps. And that could help Google's effort to make Android more appealing than Apple's iOS mobile operating system.
Google believes Android device makers will come out with new models featuring special chips called digital signal processors that will support methods for deep learning, a type of AI that Google and other technology companies have been using inside their apps in the past few years.
Deep learning involves two steps: First, researchers train software on lots of data, like photos. Then, the researchers show new data to the neural networks and tell them to make inferences about the data. For example, neural networks that have been fed pictures of dogs and other animals that are properly labeled can make an educated guess about whether a new picture contains a dog.
Historically, this work was done on powerful computers in a data center somewhere, with the results fed back down to the mobile device. But Google foresees a day when phones and tablets will be able to do it themselves.
As a result, apps will be able to get smarter and do more helpful and interesting things more quickly, without calling back to data centers. Apps could refine their image-recognition or speech-recognition systems on the fly. More broadly, Android devices will have more computing power onboard.
Google will make it easier for developers to make use of that added computing power by introducing a new application programming interface for neural networks, Dave Burke, vice president of engineering for Android at Google, announced on Wednesday at the Google I/O developer conference in Mountain View, California. The tool will be part of TensorFlow Lite, a special version of Google's TensorFlow open-source software for deep learning.
TensorFlow Lite is "a library for apps designed to be fast and small, yet still enabling state-of-the-art techniques," Burke said. Last year Google enhanced TensorFlow with Android and iOS support, but now Google is going further.
Today's news comes a few weeks after Facebook introduced Caffe2, an open-source deep learning framework that can work inside Android and iOS apps. Apple has not yet added tools for incorporating AI systems into iOS apps.
Watch: Google conference kicks off