Are you in a dilemma when choosing between a personal assistant and your privacy? Are you afraid that someone is tracking you when you want personalized services with just a command? Well, [slight pause] worry less if you are using an iPhone (more specifically, iOS13).
It has always been difficult for users to choose between services and privacy, especially with intelligent services that require Machine Learning and AI techniques to understand the user. And with all the data breach news from the tech giants, it is getting just worse. However, with the new iOS 13, Apple has taken several steps towards privacy because you know, as they say, “Privacy matters.” With the new iOS 13, they have enhanced the privacy and security aspects of your data on several fronts but, the most important is Siri, Apple’s voice AI assistant.
Siri now recognizes your voice and wakes up only when you say it. It recognizes your voice without actually storing your audio data. Strange? Isn’t it? We felt it too but, as Apple’s head of privacy, Julien Freudiger explained, Siri primarily uses a technique called, Federated Learning. A privacy-preserving machine-learning method introduced by Google in 2017. Shocking right? But, that’s the fact (you can make a joke here with Google preserving privacy). This method allows Apple to use the audio data available locally to train different speaker recognition models across all of its user’s devices. Only then it sends the updated models to a central server and combines with a master model. This way, your raw voice data is not leaving your iPhones and iPads but, Siri continuously gets better acknowledging you.
But, that’s not it, Apple further adds another layer of protection using something they call “differential privacy”. This method injects small amounts of noise into any raw data before applying machine learning on them to make it difficult to reverse-engineer the original audio files from the trained model.
With the rise in privacy concern, Google has released a new set of tools to allow developers to implement their own federated learning models. With the technique still being relatively new, we hope that it finds its ways through several other domains and get adopted by other companies.
So what do you think of this new machine learning technique? Are you excited to again get back to your voice assistant without losing your privacy? Do let us know in the comments section.
Thank you and Stay Knotted!