Neuroscientist Poppy Crum says there is great opportunity in using AI to aid in healthcare diagnoses, but privacy challenges must be addressed first.

Published: May 9th, 2019

Privacy advocates worry about the pitfalls of smart speakers spying on us when we don’t intend it, but what if having an always-listening, always-connected microphone near us could be good for our health?

While adoption of voice-activated assistants like Amazon Alexa are growing faster than any previous technology in history, privacy is a major concern for the remaining holdouts from the technology. But what if having smart voice assistants like Alexa listening to us all the time is actually a good thing? What if it could even improve our health?

That’s the concept that Poppy Crum, chief scientist at Dolby Laboratories and a Ph.D. neuroscientist that’s an adjunct professor at Stanford University and often speaks of in keynote presentations around the world. IT World Canada spoke with Crum at the CIO Peer Forum in Toronto.

“It’s an unprecedented time when the amalgamation of our consumer technology can know more about our mental and physical wellness than most clinical visits can reveal,” she said. “Even something like an Alexa or smart speaker with our voice. There are many different conditions that can be detected in not what we say, but how we say it.”

Crum points to early biological indicators — or phenotypes — that indicate the onset of chronic conditions. One of the earliest ways to detect the degenerative cognitive illness of Alzheimer’s is to analyze speech — the cadence, tone, and fluidity at which an individual is able to verbalize. If a machine learning algorithm could record everything you say over a long period of time and analyze it with machine learning to detect degradation patterns, you could be warned much earlier about diseases than when a doctor would detect them. What Crum refers to the “temporal dynamics of your speech” are also affected by diabetes and heart disease, so there is potential to detect more than just cognitive conditions.

There are clearly privacy problems to be solved with voice-activated technology, Crum says. But the potential benefits could make it worthwhile.

“With these hearable devices, you have the issue where one person in the space has given permission for something whereas other people might be subjected to that, but their permission haven’t been shared,” she said. “That’s a problem we have to solve. Because the healthcare benefits are substantial. They are absolutely worth all of the difficulties we do have to solve. And that doesn’t mean we’re ready for this today.”

In Canada, 5.8 million people are using a smart speaker at least once a month, according to eMarketer. That’s a 51 per cent increase over 2018. The research firm projects continued growth, estimating 6.7 million Canadians will use a smart speaker in 2020. Globally, it’s estimated that the number of smart speakers worldwide will grow to 225 million installed by 2020. Amazon is the market leader and was the first to offer such a device, but other tech giants including Google and Samsung are also pushing voice-activated speakers as the next hot consumer device. (Google Home is the most popular smart speaker in Canada, as it launched before Amazon Echo here, and was the first to support both Canadian English and French languages.)

Yet ask those who don’t yet ask a smart speaker to turn on their lights, set timers, or play music why they don’t have one and privacy is a chief concern. Perhaps with good reason, as a Bloomberg report published April revealed that Amazon has a team of people listening to Alexa recordings in an effort to improve the service. Having strangers listen to what you say to Alexa in the privacy of your home was never part of the bargain when you signed up for Alexa (not even buried in a lengthy term of service agreement.)

Recently journalists have also gone through their record of personal Alexa recordings and found that false positive incidents of the trigger word are quite common. Alexa records many snippets of conversation not really intended for it when it thinks it hears the trigger word.

Ann Cavoukian at CIO Peer Forum
Privacy advocate Ann Cavoukian says technologies like Alexa provide no assurances of privacy.

Former Ontario Information and Privacy Commissioner and the head of the Privacy by Design Centre for Excellence at Ryerson University Ann Cavoukian has been a vocal opponent of smart speakers on the grounds they offer poor privacy standards.

“I just don’t think it’s ready for prime time,” she says of the technology being used to diagnose chronic conditions. “The benefits of early detection of Alzheimers’ will fade away because of all these unintended consequences.”

Cavoukian says her fear is that third parties can gain access to identifiable information through a service like Alexa. To operate Alexa, Amazon has to centralize a lot of user data in order to deliver useful results and instruct its machine learning algorithms. Cavoukian would prefer a decentralized model that would give individuals more control over when their data is used and for what purpose. But it’s hard to imagine how that could be applied in a smart speaker scenario.

“The home is the last bastion of privacy, so don’t get an Alexa,” she advises. “Or turn the thing off while you’re talking.”

Despite the concerns of privacy advocates or consumer holdouts, Amazon is clearly interested in the healthcare application potential for Alexa. On April 4, it announced it had achieved compliance with the U.S. Health Insurance Portability and Accountability Act of 1995 (HIPAA). The program allows select organizations to build Alexa Skills that can receive and transmit personal health information. The law requires that individuals provide written authorization for any third party seeking the use of their data.

And in a patent published in March 2017, Amazon scopes out methods for analyzing Alexa recordings to detect human emotions or mental health states.

In her keynote speech at CIO Peer Forum in Toronto, Crum asked the audience to consider the intent behind those deploying technologies. When a technology is designed with intent, there’s a big opportunity to ensure that each individual user is realizing the right outcome.

So what intent does Amazon need to have if it’s going to make healthcare services work through a service like Alexa?

“Data is necessary for improved experiences for personalized products,” she says. “It’s really critical that the data that’s been captured is split up and encrypted in ways that it can’t be linked back to my personal interactions.”

In other words, Amazon needs to find ways to improve Alexa without creeping us out in the process.