Smart Speakers Should Embrace Privacy by Design

Article Featured Image

Can you guess what’s coming soon to a smart speaker near you? Contentious debates around “regulation versus innovation.” That’s right! Regulating Big Tech is on the radar of governments globally, and opening salvos have been fired over smart speakers.

The California Consumer Privacy Act (CCPA) is an overarching regulation meant to enhance data privacy and data protection rights for California residents; it goes into effect in January 2020. Related to the CCPA, Assembly Bill 1395 (AB-1395) seeks to regulate privacy provisions for “connected devices with a voice recognition feature”—i.e., smart speakers such as Amazon Echo, Apple HomePod and Google Home. The clock ran out on the legislative process for AB-1395 this year, but it is expected to be taken up again by the California legislature later in 2020.

Privacy concerns are to be expected with always-on devices that are passively listening to your most intimate conversations. But media reports of employees and contractors of Amazon and Apple listening to such voice recordings, even if they did so only to improve the products, have escalated such concerns. AB-1395, dubbed the “anti-eavesdropping bill” by its proponents, prohibits collection and storage of voice recordings without explicit consent from users.

But AB-1395 also raises interesting questions about the best way to regulate artificial intelligence/machine learning technologies that power smart speaker applications. If we have speech recognition apps on steroids today, that’s because they have been trained using large speech datasets. Despite impressive performance gains in recent years, there is still much room for improvement when it comes to natural language understanding applications and speech recognition systems. A good way to bring about such improvement is to review recordings of interactions between users and speech apps and provide feedback to the system on hits and misses. This approach (called supervised learning) is used in a majority of the machine learning applications today. The crux? You’ll need humans to comb through voice recordings if we want to keep improving speech apps.

Meanwhile, brandishing the General Data Protection Regulation (GDPR), German authorities asked Google to halt processing voice recordings as they investigate further whether it’s in compliance with GDPR. Separately, Google also scored an own goal when the product specifications for its Nest security device did not disclose that there was a microphone, only for that fact to be later called out by the media. In short, snafus all around. Of course, the tech vendors say all the right things, but it does make you wonder whether privacy concerns are indeed getting the attention they deserve.

AB-1395 may have bipartisan support, but it is not a law (yet). The spotlight it has shone on data collection and the privacy issues of smart speakers, however, has already had some positive impact. Amazon, Apple, and Google now give you the ability to delete your voice interaction history and enable a bit more control over data retention. Of course, this is just a start, and more needs to be done in terms of switching to explicit opt-in/user permissions for data collection, with the ability to opt out at any time, communicating clearly what happens to the data collected, who has access to it and for how long, where it is stored, and whether such data is shared with any third parties.

Privacy cannot be an afterthought; rather, it should be by design. In other words, commit to privacy not just in letter, but also in spirit. That would help engender consumer trust and also puts Big Tech on a firmer ground with regulators.

Imagine the endless possibilities of a voice-first world. Consider that Amazon recently announced a whole slew of new Echo devices for every corner of your home, for your car, and even as wearable devices; other companies will follow suit. A magical genie is on standby everywhere, just waiting to be summoned and enabling a “your wish is my command.” For the first time ever, voice looks to be a viable interface for interacting with and making sense of the world around us. But if we bungle regulation or privacy, it may be enough to delay or even halt progress. 

Kashyap Kompella is the CEO of rpa2ai Research, a global AI industry analyst firm, and is the co-author of Practical Artificial Intelligence: An Enterprise Playbook.

SpeechTek Covers
for qualified subscribers
Subscribe Now Current Issue Past Issues
Related Articles

Security Threats to Speech Apps in the Age of Deep Fakes

As machine learning marches forward, we are becoming aware of the dangers of "deep fakes" generated by deep learning algorithms. Not surprisingly, machines can be both the problem and the solution.

What’s Your Enterprise Strategy for Voice Applications?

It matters who your target is