Seattle - Back in October, Deschutes Brewery’s Brian Faivre was fermenting a batch of
Obsidian Stout in a massive tank. Something was amiss; the beer wasn’t
fermenting at the usual temperature. Luckily, a software system triggered a
warning and he fixed the problem.
"We would have had to dump an entire batch,"
the brewmaster said. When beer is your bottom line, that's a calamity.
The software that spotted the temperature anomaly is from
Microsoft Corp. and it's a new type that uses a powerful form of
artificial intelligence called machine learning. What makes it potentially
revolutionary is that Deschutes rented the tool over the internet
from Microsoft's cloud-computing service.
Day to day, Deschutes uses the system to decide when to
stop one part of the brewing process and begin another, saving time while
producing better beer, the company says.
The Bend, Oregon-based brewer is among a growing
number of enterprises using new combinations of AI tools and cloud
services from Microsoft, Amazon.com and Alphabet’s Google. C-SPAN is using
Amazon image-recognition to automatically identify who is in the government TV
programs it broadcasts. Insurance company USAA is planning to use similar
technology from Google to assess damage from car accidents and floods without
sending in human insurance adjusters. The American Heart Association is using
Amazon voice recognition to power a chat bot registering people for a
charity walk in June.
AI software used to require thousands of processors and
lots of power, so only the largest technology companies and research
universities could afford to use it. An early Google system cost more than $1
million and used about 1 000 computers. Deschutes has no time for such
technical feats. It invests mostly in brewing tanks, not data centres.
Only
when Microsoft, Amazon and Google began offering AI software over the internet
in recent years did these ideas seem plausible. Amazon is the public cloud
leader right now, but each company has its strengths. Democratizing access to
powerful AI software is the latest battleground, and could decide which tech
giant emerges as the ultimate winner in a cloud infrastructure market
worth $25 billion this year, according to researcher IDC.
New generation
"There's a new generation of applications that
require a lot more intense data science and machine learning. There is a race
for who is going to provide the tools for that," said Diego Oppenheimer,
chief executive officer of Algorithmia Inc., a startup that runs a marketplace
for algorithms that do some of the same things as Microsoft, Amazon and
Google's technology.If the tools become widespread, they could transform work
as more automation lets companies get more done with the same human work
force.
C-SPAN, which runs three TV stations and five web
channels, previously used a combination of closed-caption transcripts and
manpower to determine when a new speaker started talking and who it was. It was
so time-consuming, the network only tagged about half of the events it broadcast.
C-SPAN began toying with Amazon's image-recognition cloud service the same day
it launched, said Alan Cloutier, technical manager for the network's archives.
Now the network is using it to match all speakers against
a database it maintains of 99 000 government officials. C-SPAN plans to enter
all the data into a system that will let users search its website for things
like Bernie Sander's healthcare speeches or all times Devin Nunes mentions
Russia.
As companies try to better analyse, optimize and predict
everything from sales cycles to product development, they are trying AI
techniques like deep learning, a type of machine learning that's produced
impressive results in recent years. IDC expects spending on such cognitive
systems and AI to grow 55 percent a year for the next five years. The
cloud-based portion of that should grow even faster, IDC analyst David
Schubmehl said.
"In the fullness of time deep learning will be one
of the most popular workloads on EC2," said Matt Wood, Amazon Web
Services' general manager for deep learning and AI, referring to its flagship
cloud service, Elastic Compute Cloud.
Image-recognition
Pinterest uses Amazon's image-recognition service to let
users take a picture of an item - say a friend's shoes - and see similar
footwear. Schools in India and Tacoma, Washington, are using Microsoft's Azure
Machine Learning to predict which students may drop out, and farmers in
India are using it to figure out when to plant peanut crops, based on monsoon
data. Johnson & Johnson is using Google's Jobs machine-learning algorithm
to comb through candidates' skills, preferences, seniority and location to
match job seekers to the right roles.
Google is late to the public cloud business and is using
its AI experience and massive computational resources to catch up. A
new "Advanced Solutions Lab" lets outside companies participate
in training sessions with machine-learning experts that Google runs for its
own staff. USAA was first to participate, tapping Google engineers to help
construct software for the financial-services company. Heather Cox, USAA's
chief technology officer, plans a multi-year deal with Google.
The three leaders in the public cloud today have also
made capabilities like speech and image recognition available to customers who
can design apps that hook into these AI features -- Microsoft offers 25
different ones.
"You can build software that is cognitive - that can
sense emotion and understand your intent, recognize speech or what’s in an
image - and we provide all of that in the cloud so customers can use it as part
of their software," said Microsoft vice president Joseph Sirosh.
Read also: LISTEN: What are the US' tech giants up to?
Amazon, in November, introduced similar tools. Rekognition
tells users what's in an image, Polly converts text to human-like speech and
Lex - based on the company's popular Alexa service -- uses speech and text
recognition for building conversational bots. It plans more this year.
Chris Nicholson, CEO of AI company Skymind, isn’t sure
how large the market really is for AI in the cloud. The massive data sets some
companies want to use are still mostly stored in house and it's
expensive and time-consuming to move them to the cloud. It’s easier to bring
the AI algorithms to the data than the other way round, he said.
Amazon's Wood disagrees, noting healthy demand for the
company's Snowball appliance for transferring large amounts of
information to its data centres. Interest was so high that in November
Amazon introduced an 18-wheeler truck called Snowmobile that can move 100
petabytes of data.
Microsoft's Sirosh said the cloud can be powerful for
companies that don't want to invest in the processing power to crunch the data
needed for AI-based apps.
Take Norwegian power company eSmart Systems, which
developed drones that photograph power lines. The company wrote its own
algorithm to scan the images for locations that need repair. But it rents the
massive computing power needed to run the software from Microsoft's Azure cloud
service, CEO Knut Johansen said.
As the market grows and competition intensifies, each
vendor will play to their strengths.
"Google has the most credibility based on tools they
have; Microsoft is the one that will actually be able to convince the
enterprises to do it; and Amazon has the advantage in that most corporate data
in the cloud is in AWS," said Algorithmia's Oppenheimer. "It's
anybody's game."