Tech News

Apple and Google’s AI New Wizardry promises privacy at a cost

Since dawn of iPhone, many smart phone smartphones have come from other places: corporate computers Hodei. Mobile applications sent user data to the cloud for useful tasks such as transcribing speech or suggesting message responses. Now Apple and Google say mobile phones they are intelligent to make decisive and sensitive machine learning tasks like their own.

At Apple’s WWDC event This month, the company said its virtual assistant Siri it will transcribe speech in some languages ​​without touching the cloud on recent and future iPhones and iPads. In his time I / O developers event last month, Google has said its latest version Android the operating system has a function aimed at processing sensitive data securely and on the device, called Private Compute Core. Initial use should include a version of the company’s Smart Reply feature, which is included on the mobile keyboard that can suggest replies to incoming messages.

Apple and Google say it offers faster privacy and apps for learning the machine on your device. Failure to transmit personal data reduces the risk of exposure and saves the time spent waiting for data over the internet. At the same time, storing data on devices is in line with the interest of technology giants in keeping long-term consumers connected to their ecosystems. People who hear that they are processing their data in a more private way may be willing to agree to share more data.

The promotion that companies have recently made in machine learning is to limit the data that their clouds can “see” after years of working on technology.

In 2014, it began collecting some data on Google Chrome browser usage through a technique called differential privacy, adds noise to the data collected in ways that limit what these samples reveal about individuals. Apple has used the technique of data collected from phones to report emoji and typing predictions and for web browsing data.

Recently, both companies have adopted a technology called federated learning. It allows you to update your cloud-based machine learning system without receiving raw data; instead, individual devices process data locally and only share digested updates. As with differential privacy, companies have discussed the use of federated learning only in limited cases. Google has used the technique of writing predictions on mobile to keep up to date with language trends; Apple has published research on its use update voice recognition patterns.

Rachel Cummings, an assistant professor at Columbia and who has previously made inquiries about privacy with Apple, says the rapid change in machine learning on phones has been striking. “It’s very rare to see something that extends to scale so far from the first conception,” he says.

This advancement requires not only computer advances, but also the practical challenges of processing data on consumer-owned devices. Google has said that its federated learning system only touches users ’devices when they are connected, idle, and on a free internet connection. The technique was made possible in part by improvements in the capacity of mobile processors.

Harder mobile hardware has also helped Google 2019 announcement Knowledge of the voice of the virtual assistant on pixel devices would be completely in the device, free from the crutch of the cloud. Apple’s new voice recognition device for Siri, announced at WWDC this month, will use the company’s “neural engine” added to mobile processors to automatically turn on learning algorithms.

The technical feats are impressive. It is debatable to what extent users will change their relationship with technology giants.

Apple’s WWDC presenters said Siri’s new design was a “major privacy update” that inadvertently addressed the risk associated with streaming audio to the cloud, saying it was a major privacy concern for users. voice assistants. Some Siri commands (such as setting timers) can be fully detected locally for a quick response. However, in many cases, commands transcribed to Siri — presumably from unexpected recordings — will be sent to Apple servers to decode and respond to the software. Siri voice transcription will be cloud-based for smart HomePod speakers that are commonly installed in bedrooms and kitchens, as unexpected recordings can be worrisome.

Google promotes data processing on the device as a privacy winner and has stated that it will expand the practice. The company expects partners like Samsung that use the Android operating system to adopt the new Privacy Compute Core and use it for sensitive data-based functions.

Google has made local analysis of crawl data a feature of its proposal reinvent the targeting of online ads, called FLoC and claimed to be more private. Academics and some competing technology companies have said the design will help Google consolidate its dominance over online ads, making it more difficult to target other companies.


Source link

Related Articles

Back to top button