SUMMARY:
Summarize the case. Identify the main point (as in “What’s your
point?”), thesis, or conclusion of this case. (5 points)
Case StudyWhat’s
Up with IBM’s Watson?
In
February, 2011 an IBM computer named Watson made history by handily defeating
the two most decorated champions of the game show Jeopardy, Ken Jennings and
Brad Rutter. Watson was named after IBM’s founder, Thomas J. Watson, and its
achievement marked a milestone in the ability of computers to process and
interpret human language.
IBM
had been working on Watson for years. The project’s goal was to develop a more
effective set of techniques that computers can use to process natural language – language that human beings
instinctively use, not language specially formatted to be understood by
computers. Watson had to be able to register the intent of a question, search
through millions of lines of text and data, pick up nuances of meaning and
context, and rank potential responses for a user to select, all in less than
three seconds.
The
hardware for Watson used in Jeopardy consisted of 10 racks of IBM POWER 750
servers running Linux, with 15 terabytes of RAM and 2,880 processor cores
(equivalent to 6,000 top-end home computers), and operated at 80 teraflops.
Watson needed this amount of power to quickly scan its enormous database of
information, including information from the Internet. To prepare for Jeopardy,
the IBM researchers downloaded over 10 million documents, including
encyclopedias and Wikipedia, the Internet Movie Database (IMDB), and the entire
archive of The New York Times. All of the data sat in Watson’s
primary memory, as opposed to a much slower hard drive, so that Watson could
find the data it needed within three seconds.
Watson
is able to learn from its mistakes as well as its successes. To solve a typical
problem, Watson tries many of the thousands of algorithms that the team has
programmed it to use. The algorithms evaluate the language used in each clue, gather
information about the important people and places mentioned in the clue, and
generate hundreds of solutions. Human beings don’t need to take such a formal
approach to generate the solutions that fit a question best, but Watson
compensates for this with superior computing power and speed. If a certain
algorithm works to solve a problem, Watson remembers what type of question it
was and the algorithm it used to get the right answer. In this way, Watson
improves at answering questions over time. Watson also learns another way — the
team gave Watson thousands of old Jeopardy questions to process. Watson
analyzed both questions and answers to determine patterns or similarities
between clues, and using these patterns, it assigns varying degrees of confidence
to the answers it gives.
Although
Watson was only able to correctly answer a small fraction of the questions it
was initially given, machine learning allowed the system to continue to improve
until it reached Jeopardy champion level. IBM term cognitive computing to refer
to Watson’s ability to interpret speech and text, rapidly mine large volumes of
data, answer questions, draw conclusions, and learn from its mistakes.
The
Watson version used in Jeopardy took 20 IBM engineers three years to build at
an $18 million labor cost, and an estimated $1 million in equipment. IBM saw
the investment as a stepping stone to broader commercial uses of its AI
technology, including applications for health care, financial services, or any
industry where sifting through large amounts of data (including unstructured
data) to answer questions is important. Watson is expected to become more
useful and powerful by learning from new sets of experts in new fields of
knowledge. In January 2014 the company created a new division, the Watson
Business Group, which will have 2500 employees working largely in New York
City’s Silicon Alley. IBM has invested more than $1 billion in this Group, and
has allocated one-third of its overall research efforts to Watson.
In
September 2011, WellPoint Inc., the largest U.S. health care provider, with
34.2 million members, enlisted Watson for utilization management. The WellPoint
Interactive Care Reviewer application is designed to determine if physicians’
requested treatment meets the guidelines of the company and a patient’s
insurance policy. The Watson WellPoint application combines data from three
sources: a patient’s chart and electronic records maintained by a physician or
hospital, the insurance company’s history of medicines and treatments, and
Watson’s huge library of textbooks and medical journals. According to WellPoint
vice president Elizabeth Bingham, Watson initially took too long to “learn”
WellPoint’s policies. IBM was able to improve the system by revising the Watson
training routine for WellPoint, and the Interactive Care Reviewer is being
adopted by 1600 health care providers.
Cancer
treatment appears to be an especially promising application for Watson. Current
guidelines aren’t precise enough to determine treatments that are most appropriate
for a specific patient. For example, the recommended treatment may be
chemotherapy, but how do you pick among ten or more possible chemotherapy
options? How do you choose the dosage? What treatment frequency would work
best? Oncologists also can’t keep pace with the torrent of cancer research
findings and therapies, genomic techniques, and patient record data. It is just
too much for even a highly-trained scientist to manage.
In
2012 Memorial Sloan-Kettering Cancer Center began work on a Watson application
to recommend cancer treatments, using data from Sloan-Kettering’s clinical
database of over one million patients along with treatment guidelines and
published research to help Sloan-Kettering researchers to recommend
personalized treatment options for lung cancer patients. The Watson application
needs to pass a series of tests in order to be used on cancer patients, and
actually being able to use Watson is more complex than originally envisioned.
For instance, Sloan-Kettering oncologist Dr. Mark Kris displayed a screen from
Watson that listed three potential treatments, but Watson was less than 32
percent confident that any of them were correct. Ari Caroline, director of
Sloan Kettering’s quantitative analysis and strategic initiatives group has tutored
Watson and has said that system was still in pilot mode. But progress is
genuine, and Caroline believes Watson will soon be able to guide oncologists in
selecting treatment options and tackling new research. A final version of the
system has not yet been released.
Researchers
at the University of Texas MD Anderson Cancer Center worked with IBM for a year
to build a version of Watson called Oncology Expert Advisor (OEA) to recommend
cancer treatments by mining medical literature, with an initial focus on acute
leukemia. Watson learned from a variety of data about which cancer treatments
worked best and which should be avoided for specific patients. OEA “reads” the
medical records of patients to generate case summaries. It then weighs the
patient profile against its knowledge base to suggest treatment options
relevant to that particular patient, based on literature, guidelines and expert
recommendations. When asked by a doctor about a patient, Watson’s algorithms
search for possible treatments and rank them according to levels of confidence
up to 100%, with each option linked to supporting evidence.
The
project initially stumbled because IBM engineers and Anderson doctors couldn’t
understand each other. IBM developers worked elsewhere and only visited Anderson
every few weeks to talk to doctors. When IBM developers and doctors started
meeting several times a week, the application became much better and the
leukemia advisor is nearly ready for use. However, it might take two more years
before Watson could handle other types of cancer. And although Watson might
help oncology specialists at M.D. Anderson identify leukemia treatment options,
it can’t substitute for the expertise of an experienced doctor, according to
Lynda Chin, chairperson of the M.D. Anderson genomic department. The cancer
experts have seen patients a thousand times, and sometimes their decisions are
based on intuition that’s difficult to explain. The Anderson project was valued
at nearly $15 million, and IBM management is hoping it could grow to $100
million. The Anderson project plans to expand to other cancer types once the
prototype becomes more developed.
In
November 2013 IBM announced it would make Watson technology available via the
Internet as a cloud service that could be used by many different industries.
IBM will open parts of the system to outside developers to create businesses
and mobile applications based on cognitive computing. A Watson Developer Cloud
provides tools and methodologies for developers to work with a Watson system, a
content store supplying both free and fee-based data for new applications, and
about five hundred subject matter experts from IBM and third parties. Welltok
used these tools to create a mobile Watson app called CareWell Concierge for
Intelligent Health Itineraries for consumers. Users will be able to participate
with Watson in conversations about their health. Fluid Retail is developing a
personalized shopping assistant. MD Buyline is developing a Watson app to
advise hospital managers about procurement of medical equipment and supplies.
IBM
will deliver three new cloud-based products based on Watson’s cognitive
intelligence and capabilities. IBM Watson Discovery Advisor is aimed at the
pharmaceutical, publishing, and education industries, and will wade through
search results to deliver data faster and help researchers formulate
conclusions. IBM Watson Analytics is a cloud-based service that provides
insights, including visual representations, based on raw big data enterprises
send to Watson. IBM Watson Explorer is a cloud service that will provide a
unified view of a user’s information, facilitating the revelation and sharing
of data-driven insights
IBM
has also made Watson easier and less expensive to use. The latest version of
Watson is 24 times faster than the version used in the 2011 Jeopardy contest,
using only 10 percent of the hardware used in the Jeopardy version.
Nevertheless,
Watson thus far has not produced much revenue for IBM—only about $100 million
from commercialization efforts between 2011 and 2014. IBM CEO Virginia Rometty
is hoping Watson will be able to produce $10 billion in annual revenue within a
decade, and that Watson will bring in $1 billion in revenue per year by 2018.
In
order to effectively commercialize the technology, IBM will need to expand
Watson’s knowledge domains, and this is its greatest challenge. Turning Watson
into a useful business tool requires an enormous amount of work. Watson has to
learn the terminology and master the domains of expertise in many different
areas, including health care and scientific research, understand the context of
how that language is used, and how to correlate questions with the correct
answers. Watson doesn’t work yet with data from audio, video, and animations
and with languages other than English. It can’t come up yet with its own ideas.
IBM
will have to be careful not to oversell what Watson can do, so that Watson does
not end up like other artificial intelligence systems where expectations were
way overblown. Making machines that beat humans at chess or a TV game show is
much easier than solving problems in the real world. According to Curt Monash,
president of Monash Research, Watson hasn’t yet overcome the hurdle that
derailed AI in the 1980s, which was that AI was only able to capture small pieces
of a limited knowledge domain for a single-purpose use. Watson is having more
trouble solving real-life problems than Jeopardy questions. Watson’s basic
learning process requiring IBM engineers to master the technicalities of a
customer’s business and translate those requirements into usable software has
been very arduous. It remains to be seen whether the complexity of establishing
a body of knowledge and training an intelligent system is repeatable and
scalable for other types of work and whether it creates an opportunities for
differentiation and competitive advantage. Watson is very much a work in
progress.
Sources: Mohana Ravindranath, “How IBM Is Trying to Commercialize
Watson,” Washington Post, May 11, 2014; Spencer E. Ante, “IBM
Struggles to Turn Watson Computer Into Big Business, Wall Street Journal,
January 7, 2014; Lynda Chin, “IBM Watson: Providing a Second Opinion for
Oncologists,”www.ibm.com, accessed July 9,
2014; George Lawton, “IBM’s Watson Supercomputer Gives Developers Access to
Cognitive Cloud,” SearchCloudApplications.com, March 28, 2014; Jack Vaughan,
“For IBM Watson, No easy Answers on Commercial Cognitive Computing,”
Searchdatamanagement.com, January 10, 2014; Michael Goldberg, “Five Things to
Know about IBM Watson, Where It Is and Where It’s Going,” DataInformed, January
14, 2014; Larry Dignan, “IBM Forms Watson Business Group: Will
Commercialization Follow?,” ZDNet, January 9, 2014; Quentin Hardy, “IBM Bets
Watson Can Earn Its Keep,” New York Times, January
8, 2014 and “IBM to Announce More Powerful Watson via the Internet,” New York Times, November 13, 2013; Ian B Murphy,
“Predictive Analytics in Development: IBM Watson at Memorial Sloan-Kettering,
RPI Research Lab,” DataInformed, February 20, 2013; Anna Wilde Mathews, John
Markoff, “Computer Wins on ‘Jeopardy!’: Trivial, It’s Not,” The New York Times, February 16, 2011; Stanley Fish, “What
Did Watson the Computer Do?” The New York Times, February
21, 2011; and Stephen Baker, “The Programmer’s Dilemma: Building a Jeopardy!
Champion,” McKinsey Quarterly, February 2011.
