I should start this story out by saying I am a fan of the Google Home and Amazon Echo ecosystems. Both add convenience to my life, my family loves them, and I genuinely appreciate much of the information that both companies suck up in an attempt to deliver a better assistant.
Indeed, I’ve been waiting a long time to have my own robot, one that can suggest when I leave for the airport to avoid traffic delays, or let me set a screen time timer for my kid and have it turn off the television or broadband on her devices when that time has been exceeded (especially since a lot of that screen time is consumed after I’ve gone to bed).
To be clear, neither of those things are currently available, but we’re getting close to the day when they are. That said, until we figure out a way to assure the public that all these microphones and cameras inside their homes are designed for their benefit and will only activate when asked, such devices run the risk of becoming a short-lived novelty or, worse, the basis of a surveillance system that no one can escape.
Recent stories about both Google and Amazon outsourcing the review of people’s utterances to those devices have caused many to question the wisdom of having smart speakers made by either company in their home. Indeed, both Amazon and Google send a small segment of recordings to contract workers who then listen to make sure the natural language processing understood the commands and subsequently did what people asked. Which raises three distinct issues worth detailing.
The first is that contractors are listening to a small segment of recordings from people’s smart speakers, people who might ultimately be identifiable. The second is that in a fairly significant number of cases the recordings were made even though someone hadn’t ask for Alexa or Google and may not have been aware that they were being recorded. And finally, the contract workers listening in on these recordings are hearing what they think are instances of domestic violence and they don’t know what to do about it.
The response to this news has been split between those dismissing it out of hand, saying it’s to be expected because how else one would check quality, and those gleefully pointing out the lack of privacy. But a more productive way to talk about it would be to figure out if and if so, how we can take these disclosures and use them to create a more privacy-aware way to handle QA, address people’s concerns, and figure out how to address disturbing overheard conversations.
So let’s talk privacy. The single best way to solve privacy issues related to smart speakers is to change the underlying business model of the web. Companies suck up your data to better target you with advertising. So the best way to preserve your privacy is to pay for your digital assistant as a service so it doesn’t collect your data to later use it for advertising. That, of course, is not happening.
This leaves us with good policies for data control and access. Amazon and Google both let you see what the machine is recording, and let you delete those recordings (although not necessarily the data collected from them, especially if it’s necessary for tasks you’ve asked the machine to do). So my first recommendation is to regularly go to the apps to see what Alexa or Google has heard and delete anything you find problematic. Here’s how to do it for Alexa, and here’s how to do it for Google’s Home.
The second option is one that neither Google nor Amazon have implemented but should. They should run people’s voices through some kind of voice modulation program so contractors can’t recognize a person’s voice. That would eliminate the privacy problem associated with random people hearing your Alexa or Google requests.
It won’t help with the false positives issue, though, where Google and Amazon activate and record despite people in the home not saying the wake word. Based on a recent Google leak from Belgian broadcaster VRT, there were 153 false positives in “over 1,000” recordings shared with the broadcaster. And even if one or two of them accidentally squeezed their phone or hit a button to activate Google, that number is far too high.
On the consumer side, you can help mitigate the problem of false positives by ensuring your smart speaker makes a noise when it activates. That will at least let you know you are being recorded. In the meantime, the companies themselves should conduct audits on the percentage of false positives and share the results of those audits with the public. Ideally, we’d get a third party involved to sample the recordings (with the voice modulation) and perform the audits themselves. Doing so would also create a market incentive to build a better algorithm around the wake word.
Another suggestion is to implement a long and complicated wake phrase, which would make it much more difficult to accidentally trigger the device. Offering consumers a choice between a longer, more-difficult-to-trigger phrase and a more convenient one would be a good way to put more control in the hands of users. Combined with audits, consumers could make an informed decision on the trade-off between convenience and their privacy.
Finally, to address the concerns related to contractors hearing what are potentially abusive situations, companies should come up with formal policies that they share with their contractors. Though personally, I would love to see these contractors made into employees, because an employee is far more likely to treat the potentially sensitive data in these recordings appropriately than an hourly worker who has little loyalty to the organization. Instead, the current policy seems to be to ignore the situation altogether.
At a minimum, contractors should be able to refer disturbing recordings to a higher-up at Google or Amazon who has the training and authority to make what may be a difficult call. From there, I do think both companies have a duty to forward situations that might involve domestic violence to the police. The problem is that the data is likely old by the time it gets to the tech firms. It’s can also be tough to discern what might be violence and what might be something more innocuous. For example, I once chased my then-3-year-old daughter down a sidewalk as she shouted “Don’t beat me” at the top of her lungs because we were in a race she clearly wanted to win. Based on the startled stares of passersby, this could have gone very badly for me.
If Google and Amazon take the path of discerning reporting, they should disclose that they will share such recordings to local authorities so users can decide whether or not they want this device in their homes.
I don’t think I’ll be giving up my smart speakers anytime soon, but I did turn off Google’s ability to record my family (here’s how) and I do check the recordings every two months to see what Google and Amazon have picked up. Both of my devices light up and make a sound when they have been triggered, and I do turn them off when I’m discussing something sensitive. As a consumer, I’ve done all I can do. But I would like to see Google and Amazon implement a few changes to help consumers make more informed decisions — and be a little bit more comfortable as a result.
Like this story? Get it and more every Friday when you sign up for my free IoT newsletter. Sign up here.