Recent news of Apple’s intent to leverage differential privacy has potential implications for intelligent assistance. Unlike other consumer technology giants, Apple has been steadfast in its refusal to compromise personal privacy, even it if means compromising on personalization and other features that rely on knowledge of the user.

But as the major tech giants race to get ahead in the critical area of machine learning, in particular deep learning, Apple can hardly wait on the sidelines. Deep learning promises significant improvements in the functions of intelligent assistance. One example of such assistance is the auto-suggestion of words when we’re typing and texting.

In order for auto-suggestion to work, companies need to collect lots of real language data to understand patterns of word usage. By harvesting what millions of people are texting, deep learning algorithms can learn to predict that the word “night” is extremely likely to follow the words “it rained last.” So the next time you start typing “it rained last” the system will suggest the word “last” before you even get ready to type it.

Differential privacy allows the collection of massive amounts of data, while protecting the privacy of the individual. It works by adding noise to raw data, so that the company collecting the information is never able to trace it back to a particular individual.

As Apple works toward implementing differential privacy for iOS 10, Google seems to be experimenting with increased use of personalization for its Google Now assistant. While Google has also delved into differential privacy and has open sourced its work on the RAPPOR project, it’s not clear how or if Google is leveraging differential privacy to protect the privacy of users.

Intelligent assistance is increasingly relying on insights gleaned from deep learning algorithms. Those algorithms can only work and continue to improve if they have access to vast real-world data sets. As our use of social media, messaging platforms, voice assistants, and other data collecting devices and applications grows, that data is certainly available. Will companies protect our privacy as they collect our data in order to serve us better? Differential privacy seems to offer a solution. It remains to be seen how practical or widespread that solution will become.

Categories: Conversational Intelligence, Intelligent Assistants, Intelligent Authentication, Articles

Tags: privacy, data, Apple, machine learning, intelligent Assistants