he says. “Voice is your place in society.”
His job also entails casting and coach-
ing the actor Google hired to play Assis-
tant’s voice. Before working at Google,
Giangola produced and directed the
original version of Siri. He plays me a clip
of one studio session in which he worked
with the voice actor behind Assistant to
get her into character. (Google declined
to reveal her identity.) In this particu-
lar scenario, Giangola played the role of
a manager asking Google how a recent
interview with a potential hire went, a
task Assistant can’t currently help with
but might someday.
The actor playing Assistant replies
in what seems like a characteristically
Google way: “Well, he was on time, and
he was wearing a beautiful tie.” Factual
and upbeat, if a little odd. Her voice is
nearly impossible to distinguish from that
of the inanimate butler she’s portraying,
except for a subtle break as she says the
wordtie, which reminds me she’s human.
Which also reminds me that even a voice
that sounds genuine isn’t all that useful if
its speaker doesn’t understand the intri-
cacies of spoken language. Tell Assistant
“I’m lonely” and it will recite an elabo-
rately crafted and empathic response that
Krettek and others helped create. But tell
it “I feel like no one likes me” and it re-
sponds that it does not understand.
THERE ARE MORE prosaic problems
for Google and its competitors than
decoding the fundamentals of speech.
Engaging users in the first place is a big
one. Google doesn’t typically make its
usage statistics public, but according to
a study last year by researcher Creative
Strategies, people said they used voice
features only rarely or sometimes: 70%
for Siri and 62% for Google.
Privacy is another growing concern.
Earlier this year, for example, ambient re-
cordings from an Amazon Echo were sub-
mitted as evidence in an Arkansas mur-
der trial, the first time data recorded by
an artificial-intelligence-powered gadget
was used in a U.S. courtroom. Devices like
Echo and Google’s Home are always lis-
tening but don’t send information to their
hosts unless specifically prompted—or
in Assistant’s case, when someone says,
‘O.K., Google.’ But given that firms such
as Amazon and Google profit from know-
ing as much about you as possible, the
idea of placing Internet-connected micro-
phones around the house is disquieting
to many. Google says it clearly lays out
what kind of data its gadgets collect on
its website and points to the light on top
of its Home speaker that glows to indicate
when it’s actively listening.
And then there’s the difficulty of the
underlying problem. Computers are still
far from being able to detect the cues
that make it possible to understand how
a person may be feeling when making
a request or asking a question. To do
so, Assistant would need to learn from
a huge amount of data that depicts the
user’s voice in various emotional states.
“Training data usually includes normal
speech in relatively quiet settings,” says
Jaime Carbonell, director of the Language
Technologies Institute at Carnegie Mellon
University’s School of Computer Science.
“You don’t typically get people in highly
stressful situations to provide data. It’s
very difficult to collect data under all of
these conditions.”
Getting more devices into more
homes would help. On Oct. 4, in a splashy
presentation, Google announced two new
Home speakers, including a smaller, less
expensive version and a higher-end
device. But Amazon is still far ahead.
The company’s Echo speaker will capture
70.6% of the voice-enabled speaker
market in 2017, while Google will account
for only 23.8%, according to eMarketer
projections.
Germick doesn’t seem daunted,
though. And he isn’t shy about Assis-
tant’s shortcomings. When I ask him
which artificial intelligence from science
fiction he hopes Assistant evolves into, he
doesn’t choose one of the super-advanced,
all-knowing varieties like Jarvis in the
Iron Man films or Samantha from the 2013
movieHer. He says he hopes to make As-
sistant like the perennially cheery char-
acter played by Ellie Kemper on Netflix’s
Unbreakable Kimmy Schmidt. Despite
having been held captive underground
for 15 years and being systematically bam-
boozled, Kemper plays Kimmy as almost
unnervingly able to find the bright side.
“We talk a lot about her relentless opti-
mism,” Germick says. “Just like she came
out of the vault, we don’t always under-
stand context, but we try to stay positive.”
That’s the thing about personality: quirks
can be part of the charm. MARKET SHAREGOOGLEAPPLEMICROSOFTAMAZON49.4%
42.5%
5.1%
1.8%
Which helpers do people have
on their smartphones?SAMSUNGOTHERGOOGLEAMAZONAPPLEMICROSOFT0.7%
0.4%
23.5%
37.1%
21%
15.8%
Which helpers do
people use the most?SOURCE: HIGHERVISIBILITY, 2017