Page 16 - Fortune-November 01, 2018

installed on nearly all new phones. Amazon, colloquialisms, and the context of conversa-
in contrast, needs to get consumers to install tions by analyzing, for example, recordings of
and then open the Alexa app on their iPhones call-center agents talking with customers or
or Android devices. “The extra step to open interactions with a digital assistant.
the Alexa voice app puts Amazon at a distinct Voice-recognition systems rely as much on
disadvantage,” says Loup’s Munster, formerly physics as on computer science. Speech creates
a Wall Street analyst of computer companies. vibrations in the air, which voice engines pick
By contrast, all that’s required to activate Siri up as analog sound waves and then translate
and the Google Assistant is to say their names. NICK FOX into a digital format. Computers can then
That said, iOS and Android are open to GOOGLE analyze that digital data for meaning. Artificial
third-party developers of all stripes, and Ama- intelligence turbocharges the process by first
zon is one of them—meaning that nothing is “Every once figuring out whether the sound is directed
stopping developers on both platforms from in a while toward its systems by detecting a customer-
writing Alexa programs. Bezos bragged in an there is a chosen “wake word” such as “Alexa.” Then they
earnings release earlier this year that “tens tectonic use machine-learning models trained by what
shift in
of thousands of developers across more than technology, millions of other customers have said to them
150 countries” are building Alexa apps and and we before to make highly accurate guesses as to
incorporating them into non-Amazon devices. think voice what was said. “A voice-recognition system
Indeed, partnerships are a key battleground for is one of first recognizes the sound, and then it puts the
voice applications. Alexa is built into “sound- those.” words in context,” explains Johan Schalkwyk,
bars” from Sonos, headphones from Jabra, an engineering vice president for the Google
and cars from BMW, Ford, and Toyota. Google Assistant. “If I say, ‘What’s the weather in…,’
boasts integrations with audio equipment mak- the A.I. knows that the next word is a country
ers Sony and Bang & Olufsen, August smart or a city. We have a 5-million-word English vo-
locks, and Philips LED lighting systems, and cabulary in our database, and to recognize one
Apple has partnerships that allow its HomePod word out of 5 million without context is a super
to work with First Alert Security systems and hard problem. If the A.I. knows you’re asking
Honeywell smart thermostats. “The beauty of about a city, then it’s only a one-in-30,000 task,
these partnerships,” says Google’s Fox, “is that ROHIT PRASAD which is much easier to get right.”
they allow us to link voice into the whole smart- AMAZON Computing power allows the systems multi-
appliance ecosystem. I don’t have to open my ple opportunities to learn. In order to ask Alexa
phone and go to an app. I can just say to the “We wanted to turn on the microwave—a real example—the
device, ‘Show me who’s at my front door,’ and it to remove voice engine first needs to understand the
will pop right up. It’s simplifying by unifying.” friction command. That means learning to decipher
Artificial intelligence has long been a staple for our thick Southern accents (“MAH-cruhwave”),
customers,
of dystopian popular culture, notably from and the high-pitched kids’ voices, non-native speakers,
films such as The Terminator and The Matrix, most and so on, while at the same time filtering out
where wickedly clever machines rise up and natural background noise like song lyrics playing on
pose a threat to humankind. Thankfully, we’re means the radio. It then has to understand the many
FOX : COUR T ESY OF GOOGLE; PR AS A D: COUR T ESY OF AM A ZON voice-recognition programs were only as good ‘Choose other voice assistants match questions with
ways people might ask to use the microwave:
was voice.
not there yet, but advances in A.I. and the
It’s not
“Reheat my food,” “Turn on my microwave,”
availability of cheap computing have made im-
merely a
pressively futuristic applications a reality. Early
“Nuke the food for two minutes.” Alexa and
search en-
gine with
similar commands in the database, thereby
as the programmers who wrote them. Now
a bunch
these apps keep getting better because they
“learning” that “reheat my food” is how a par-
of results
ticular user is likely to ask in the future.
are connected through the Internet to data
that says,
centers. These complex mathematical models
The technology has taken off in part because
sift through huge amounts of data that com-
it has gotten so proficient at translating human
one.’ It tells
commands into action. Google’s Schalkwyk
panies have spent years compiling and learn
you the
to recognize different speech patterns. They
says his company’s voice engine now responds
answer.”
can recognize vocabulary, regional accents,
with 95% accuracy, up from only 80% in
117
FO R T U N E. CO M // N O V. 1 . 1 8