In
a few months from now, or at least sometime next year, a
few IBM partners will release a series of software products that will be
unlike anything people have encountered so far. Instead of doing a task
for you through a software program, these products will prepare you
instead to do that task yourself.
You could ask the computer
about your health, a home purchase, or a travel plan. You don't need
visits from sales executives for product briefings. The computer will
gently guide you to make the right choice at the right time. Currently,
these products are being built around Watson, the famous IBM computer
that won the Jeopardy championship.
To be precise, it is built
around a more compact and powerful machine than the Jeopardy champion.
IBM threw open this machine to programmers on the cloud three weeks ago
to build products and services in a few industries to begin with. IBM
has applications from 200 potential partners, including a few from
India, focused initially on healthcare, financial services and travel.
"We
are trying to build an ecosystem of partners around Watson," says Jay
Subrahmonia, vice-president of development and delivery, WatsonSolutions
at IBM. IBM calls it cognitive computing, to distinguish it from the
more common term, artificial intelligence.
It is purportedly the
next wave of computing , infinitely more powerful and long-lasting than
any other computing wave we have seen. It changes the way we interact
with computers, the reason we use computers, and also the way we program
computers.
It is a big business opportunity as well. Just the
global healthcare market for such systems is projected to increase from
$201 million now to $239 billion by 2019, according to the market
research firm WinterGreen Research.
Next computing wave
Computers
that we use now are fast but dumb. Those in the cognitive computing era
will understand the context in which they function. They will also
learn constantly and improve their capabilities. Yet, they won't be
based on any single technology, as the current examples show. The core
of Watson is based on natural language processing (NLP), or the ability
to understand human languages.
But it combines this ability with
massive computing resources and a host of other computing technologies
like machine learning, information retrieval and automated reasoning.
Other companies—many based in the Silicon Valley—are building different
solutions based on different technologies.
Grok, a recent
startup from Palm Pilot founder Jeff Hawkins, mimics the human brain to
predict anomalies in IT systems. Palantir Technologies, based in Palo
Alto, uses cognitive analytics to predict suspicious or terrorist
activities. ColdLight, based in a small town in Pennsylvania, uses
machine learning to examine thousands of factors simultaneously, usually
to identify fraud.
All of them, and hundreds of similar
companies, are part of the coming cognitive computing wave. The common
theme: understanding data. In some ways, they are an extension of the
current wave of analytics and big data companies, but there are some
differences.
Traditional analytics requires a human being to ask
a question. In cognitive computing, we get the answers without knowing
what to ask. Deloitte estimates the US cognitive computing market will
expand in five years from the current $1 billion to $50 billion.
"Growth usually takes place through labour and capital," says Rajeev Ronanki, lead for Deloitte's cognitive computing practice.
"Here it is related to machine learning algorithms." This difference can make the cognitive computing market grow rapidly.
"Traditional
approaches are like giving the computing system a fish," says Ronanki,
"whereas cognitive systems are akin to teaching a computer how to fish."
This can cause a fundamental shift with how markets grow. For IBM,
taking Watson to the cloud was a nobrainer.
Its current price is
not known, but it is considered too expensive as a standalone system.
The hardware cost alone of the machine that won Jeopardy was $3 million,
but Watson contains plenty of algorithms and data as well.
Putting
it on the cloud would enable companies to pay as they use it. Watson is
also complex for even the brightest of programmers. An Application
Programming Interface (API) on the cloud would substantially simplify
the task of programming, as the programmer would not need to understand
what is inside the box.
When computers talk, see...
So far,
Watson has been used mainly to solve problems in healthcare. At the
Memorial Sloan-Kettering Cancer Center in New York, Watson goes through
millions of pages of cancer data and recommends the best treatment
options for patients.
The sports goods company North Face uses
Watson to provide customers recommendations for the ideal gear for a
trip. Over the next year, Watson will seep into more industries as
developers make applications on the cloud. IBM says that Watson will be
among its fastest-growing business ever.
In the near future,
analysts expect Watson to drive the cognitive computing industry as
well. Yet the arena is busy with startups , some of whom claim to have
developed breakthrough technologies.
Take Grok. It came out of a
project called Numenta started by Jeff Hawkins. Numenta is supposed to
have cracked how the brain works; its Cortical Lear ning Algorithm is
modelled on the neocortex, the part of the brain involved in higher
functions like reasoning, thought and language.
Grok was spun
off this year from Numenta, which is now an open source project to
advance the technology. Grok's first product—in Beta stage—is to detect
anomalies in IT systems and prevent problems before they become
manifest. Current products to detect anomalies look for patterns
exceeding a threshold. It is hard, if not impossible, to detect
anomalies below this threshold, which is usually the case at night, when
few people use the systems.
Grok tries to solve this problem
combining three methods: learning online by itself, creating models
automatically, and recognising patterns. "Our models learn
continuously," says Craig Vaughn, vice-president of marketing and
products at Grok.
"They keep changing as IT policies change."
This ability to learn is at the heart of any cognitive computing system,
and distinguishes it from traditional analytics. It is relevant
wherever there is a fast data stream: retail, healthcare, travel ,
telecom. But not if the data stream consists of imagescomputers cannot
understand images.
This is why robots are so unreliable. "If
robots could understand the world," says Dileep George, founder of
Vicarious, "the benefits could be enormous." For example , we could have
sent them inside the Fukushima reactor to fix it. Vicarious, near San
Francisco, is among the many companies trying to make computers see.
George,
an IIT graduate, has funding from top VCs like Peter Thiel. It had a
breakthrough recently; it cracked the Captcha, a set of overlapping and
contorted letters that humans can recognise easily but computers find
impossible to read. Vicarious created an algorithm that can separate the
overlapping letters with a high degree of accuracy.
It is a good step towards making computers understand the environment around them. But there is a long way to go.