They’re listening....The case for an always on Alexa
We should welcome Amazon’s plans for Alexa to listen to
more of what we say.
By Stephen L. Carter June 9, 2019, 6:00 AM PDT
They’re listening.
Amazon’s patent application for an always-on feature for
Alexa, its popular voice-activated personal assistant, has raised a lot of
concern. “If you’re already freaked out by the privacy implications of smart
speakers like Amazon’s Echo,” says Gizmodo, “we have some bad news.” A headline
in ScienceAlert is even more direct: “Newly Released Amazon Patent Shows Just
How Much Creepier Alexa Can Get.” You get the idea.
But the anxiety is much ado about nothing. An Alexa
that’s always listening will likely prove more useful than an Alexa that isn’t;
and, in any case, always-on devices are certainly our future.
As of January 2019, Amazon had sold more than 100 million
devices that include Alexa. That number is certainly dwarfed by the billions of
devices that include Siri and Google Assistant. But the difference is, those
features come bundled with your smartphone, which has many other uses. When you
purchase an Alexa device, you are choosing to invite Amazon into your home to
listen to you.
Even those who have no interest in obtaining an
Alexa-linked device know how the thing works, if only from Amazon’s flood of
television commercials. You say “Alexa, what’s today’s weather?” — and it tells
you. Whenever Alexa hears its “wake word,”
the software assumes that the user wants its attention. Thus Alexa
“wakes,” listens and processes the words that follow as a command.
The difficulty arises because, in the words of the patent
application, a wake word “may not always precede a spoken command, and may
sometimes be included in the middle of a spoken command.” The application gives
an example: “Play ‘Mother’s Little Helper’ by the Rolling Stones.” As currently
configured, the system will respond if the user prefaces the request with
“Alexa” (that is, “Alexa, play ‘Mother’s Little Helper’”) but not if the wake
word comes at the end (that is, “Play ‘Mother’s Little Helper,’ Alexa”).
Put otherwise, Alexa as currently configured does not
match the way people speak. The user must structure each command around a
relatively formal grammar. The invention described in the patent application
would deformalize the means for addressing the device, thus enabling Alexa to
fit more easily and naturally into everyday life.
Here’s Amazon’s solution: Alexa already stores what it
hears in a buffer. Under the new configuration, according to the application,
once Alexa detects a wake word, “the device will go backwards through the audio
in the buffer to determine the start of the utterance that includes the
wakeword.” After finding what it scores as the most likely start of the
command, Alexa will perform a similar calculation to find the end. The command
will then be processed exactly like one that was preceded by the wake word.
It all makes a great deal of sense. Why then the concern?
It seems to me that there are two potential issues.
One is a worry about what happens to the information from
the audio buffer. Alexa currently retains recordings for a period of time,
helping it model the user’s needs and wants. This feature, which can be
partially disabled, has already caused privacy problems. Courts have issued
subpoenas for Alexa recordings. And as Bloomberg News has reported, human
beings at Amazon already listen to much of what Alexa hears, in an effort to
improve the algorithm. But no always-on feature was necessary for those
recordings to survive.
The second concern might be that an Alexa which listens
more closely, responding to natural language commands, will soon become an
Alexa that fades into the background. The relative formality with which the
device must be addressed serves as a reminder that we are addressing just that
— a device. The more casually we can speak, the more casual we will likely be
about using it. We might, quite literally, forget that Alexa is there. Only the
consumer can decide whether that is a feature or a bug.
Amazon says that it has no current plan to change the way
Alexa listens, but bear in mind that the always-on feature can be implemented
whether or not it is ever patented. In other words, if the notion of a device
that is always awake worries you, the fact that a patent application has been
filed shouldn’t cause you to worry more. Any device that listens to you can
already be made always-on. (Including, by the way, your smartphone.)
We accept that our laptops and smart televisions are
recording the choices we make and sending them we know not where. The only
reason we imagine that our spoken words are safe is that speech is an older,
more instinctive technology. We still think of speech as special, a
distinctively human function, and when we are in spaces we consider private, we
consider our voice as something heard only by our most intimate and trusted
acquaintances.
But to the computers that now surround us, speech is just
another form of data. The various voice-commanded devices of today, whether in
our homes, smartphones or cars, work just like keyboards or touchscreens. The
only difference is that the human input is a voice. And the only way they can
get that input is to listen for it.
So let’s calm down. Yes, it can be fun to imagine a
future in which our homes are entirely connected and yet we’re able to keep private
everything we want to keep private. But that ship sailed long before Amazon
decided to seek a patent on a minor and welcome change to Alexa.
Stephen L. Carter is a Bloomberg Opinion columnist. He is
a professor of law at Yale University and was a clerk to U.S. Supreme Court
Justice Thurgood Marshall. His novels include “The Emperor of Ocean Park,” and
his latest nonfiction book is “Invisible: The Forgotten Story of the Black
Woman Lawyer Who Took Down America's Most Powerful Mobster.”
This column does not necessarily reflect the opinion of
the editorial board or Bloomberg LP and its owners.
Comments
Post a Comment