Will Apple Catch Up to Its Competitors in Artificial Intelligence?

Believe it or not, in the 80s Time Magazine called Apple a chaotic mess without a strategic vision and certainly no future. Since then, Apple has made one of the biggest business comebacks in the last 50 years, achieving a valuation over $900 billion in 2017. This valuation is due in large part to the iPhone. The iPhone does employ AI for Siri and voice recognition ” but compared to companies like Alphabet and IBM, Apple hasn’t necessarily been a big name in the AI field.

That very well may change. In late November 2017, the richest company in the world gave Quartz a glimpse of how its deep neural network, VoxelNet, identifies data points from a LiDAR sensor. This is part of Apple’s effort to develop AI for a self-driving car. Originally, it looked like there would eventually be an Apple iCar out driving around on its own cognizance. But the company has since scrapped efforts to produce a car in favor of concentrating on software for what Tim Cook calls the mother of all AI projects. To have the best self-driving car, you have to have the best AI.  

Will Apple be able to compete in the battle for the best AI, the mother of all tech battles? If the iPhone X’s features are any indication, the answer is yes. Face ID uses a TrueDepth camera to analyze 30,000 invisible dots, creating a precise depth map of your face. Then, the neural network stores your facial data and compares subsequent datasets to the original, applying deep learning in order to adjust to changes in your appearance over time. This is the first application of face ID in a smartphone.

While Apple has been largely silent on its work with AI, companies like Google, Facebook, and Microsoft have published plenty of papers in academic journals. This would give anyone the impression that these companies are more intent on making real strides with AI, while Apple appeared to be content with making incremental progress on voice recognition and Siri.

But then, Apple published its first research paper late in 2016, placing it squarely within the ranks of every company working on solving the problem of how to make a machine generate new data from old input. Possibly, the paper and Apple’s latest revelation about its self-driving car software place the company ahead of the pack. The paper, titled Learning from Simulated and Unsupervised Images through Adversarial Training, won one of two awards for best paper at the Computer Vision and Pattern Recognition conference. The other winner was Facebook.

Apple’s research is admirable. Unlike Facebook and Google, Apple doesn’t harvest raw user data to train its AI. A great deal of data is necessary to get a machine to learn ”to get a neural network to create a new algorithm, which initiates a new process, based on analytics. When it comes to something like image recognition, Facebook and Google both have huge troves of user data they feed into their networks. When Facebook tags your picture without a human telling the network who you are, you might not be aware that old photos of you helped train the software how to do this.

Not so with Apple. The company’s new machine learning blog details its process, which involves local differential privacy: Local differential privacy has the advantage that the data is randomized before being sent from the device, so the server never sees or receives raw data.

Our system is designed to be opt-in and transparent. No data is recorded or transmitted before the user explicitly chooses to report usage information.

The research team has trained VoxelNet by feeding it real images and then instructing it to alter synthetic images to make them look more like the real thing. After that, the network compares the altered synthetic image to the real image and makes a judgment as to which one is real. It then applies this judgment towards identification of new images.

The Verge provides a little more detail on how VoxelNet works. The software takes LiDAR data points, which are 3D pixels called voxels, categorizes them and compresses them in the neural network in order to create a single, active map of a car’s surroundings. According to Roland Meertens, a Dutch engineer who builds computer vision systems for autonomous vehicles, this isn’t so much a breakthrough for self-driving cars as it is an impressive application of LiDAR data for other AI purposes. The Verge reports that researchers Yin Zhou and Oncel Tuzel benchmarked VoxelNet’s performance against a number of rival programs, and it handily outperformed them.  

This indicates Apple is further along with AI than anyone thought. But in order to beat frontrunners like Waymo and Tesla, Apple will have to make some spectacular moves. The iPhone X has been labeled the best smartphone you can buy, a breakthrough unrivaled by anything else in the industry. It’s not hard to imagine Apple rising to the occasion with its AI too.