Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Rise of the Intelligent Machines in Healthcare
February 29, 2016
Kenneth A. Kleinberg, FHIMSS
Managing Director, Research & Insights
The Advisory Board Company
Conflict of Interest
Kenneth A. Kleinberg, MA
Has no real or apparent conflicts of interest to report.
Agenda
• Learning Objectives
• Overview of Intelligent Computing
• Use in Other Industries
• Uses in Health Care
• Challenges and Futures
• Summary/Wrap-Up
• Questions
Learning Objectives
• Identify what advances in intelligent computing are having the greatest effect on other industries such as transportation, retail, and financial services, and how these advances could be applied to health care.
• Compare the types of technological approaches used in intelligent computing, such as inferencing, constraint-based reasoning, neural networks, and machine learning, and the types of problems they can address in health care.
• Identify examples of the application of intelligent computing in health care and the Internet of Things (IoT) that are already deployed or are in development and the benefits they provide, such as robotic assistants, smart pumps, speech interfaces, scheduling systems, and remote diagnosis.
The Evolving Story of Intelligent Computing/AI
Intelligent computing/AI uses
algorithms, heuristics, pattern
matching, rules, machine/deep
learning, and cognitive computing to
solve problems typically performed
by humans, as well as complex
problems difficult for humans
Intelligent systems are often
inspired by biology (parallel
computation) and, through access to
large data sets, get smarter with
use
AI has been in development for
decades, but only recently
gotten good enough for people
to notice, mostly due to advances
in other industries besides
health care
The public perception of AI is often
influenced by hundreds of sci-fi
movies, fear of “bad robots,” and a
general skepticism that
“machines” will ever be able to
master human capabilities that we
hold so dear
The rise of intelligent machines
is approaching; and the world,
especially the health care
industry, is far from prepared for
what’s to come…
What How
Who
When
Why
http://www.himss.org/ValueSuite
STEPS Benefits of Intelligent Computing/AI
• Tasks get done faster and more consistently
• Enhances the abilities of human workers
• Interacting with AI can be fun!
• Clinicians have smart “assistants” they can query
• Stuff doesn’t’ fall “through the cracks”
• Larger and more complex data sets can be accessed
• Analytics can be made smarter
• Alerts and reminders can be more intelligent
• Supports more dynamic and adaptive patient engagement
• Catches problems and trends earlier
• Adapts education to the patient and context
• Reduces labor costs
• Operates continuously and with more capacity
• Becomes more effective over time
Some (Controversial) Definitions of Intelligent Computing/AI
Intelligent
Computing/AI (can learn and adapt)
Symbolic
(Logical)
Reasoning
Statistics
and
Analytics
Cognitive
Computing
(simulates human thought
processes)
Bio-inspired
Systems
• Neural networks (multilayer,
feedforward, recurrent,
convolutional)
• Genetic algorithms
• Progeny clustering
• Machine learning
• Deep learning
• Rule/Knowledge-based systems
• Induction and deduction
• Forward and backward chaining
• Fuzzy logic
• Regression
• Descriptive and inferential
• Bayesian networks
• Random forest
• Data mining
• Predictive analytics
• Computational learning
When is it Intelligent Computing?
statistician
programmer
researcher
analyst
clinician
modeler
The less
the
has to
determine
the
order of processing
order of training
data to apply
factors to focus on
steps to improve the model
the more the
system can be
described as
intelligent
IC/AI is Vastly More Powerful than Procedural Programming
Pattern Recognition
Classification
Which class does
something belong in?
Knowledge Discovery
and Data Mining
What relationships
exist?
Prediction
What will
happen?
Clustering
How many different
groups of similar
objects?
Optimization
How can it be
made better?
Scheduling/Planning
How can we
accommodate order and
constraints?
Decision Making
What should we do?
Speech/NLP/Translation
What do you say and what
did you mean?
Machine Vision/
Perception
What do you see?
Robotics
Can we effect action
in the physical world?
Typical Problem Types for Intelligent Computing
How Fast Is IC/AI Advancing: Are We There Yet?
Exponential growth:
Will AI take off thanks to
network effects and
disruptive innovations, or
will it only make modest
advances for the next
decades?
AI Winters: AI has already
gone through a few phases of
hype and troughs of
disillusionment (1974-80, and
1987-93)
Surpass human
intelligence: Some
predict we’ll see the
“singularity” of machine
intelligence in the next
few decades
Unpredictable Timing :
Some advances seem to
never arrive (speech
recognition), while others
come upon us unexpectedly
(GPS driving directions)
60s 70s 80s 90s 2000 2010 2020 2030 50s 2040
AI and intelligent computing
advances are starting to
accelerate
2050
IC/AI Being Used Successfully in Other Industries “Under the Covers”
Transportation
Autopilots, self-driving cars,
space vehicles, complex
scheduling
Example: American Airlines Sabre System
Retail and Manufacturing
Shopping assistants, product
launches, logistics, robotic factories
Example: Amazon Machine Learning Service
Financial Services
Auto-trading, check cashing, fraud
detection, market prediction
Example: Securities Observation, News Analysis,
and Regulation System (SONAR)
Emergency Response
Biohazard response,
environmental changes,
police/military presence
Example: DigitalGlobe’s Tomnod
Service and Support
Booking assistants and tech
support
Examples: USAA’s Military Veterans Advisor
Gaming and Simulation
Video games, entertainment,
simulations, education/training
Example: Computer Go
Security, Crime
Prevention, Military
Identification, case
analysis, logistics
Example: Avigilon
Commonalities
• Complex challenges with lots of data
• Speed and consistency are important
• Resistance from existing workers
• A gradual adoption over years (or longer)
• Eventually it’s no longer considered AI
Motion Requires Real-Time Control and Adaptation to Changing Conditions
Source: Health Care IT Advisor research and analysis.
IC/AI in Transportation
Autopilots
Large airline jets are increasingly automated, can land in zero-zero
conditions (control of power, direction, braking). Pilots cannot start
the approach without having these systems operational. Example: Honeywell Flight Control Systems
Self-Driving Cars
Trials in California have had a few accidents. There are challenges in
determining level of threat/impact on vehicle (a paper bag versus a
rock), driving in reduced visibility conditions, and human drivers. Example: Google and Uber
Spacecraft and Rovers
Time lags are so great they require some self-sufficiency to carry out
essential commands. Human handlers are not always sure how the
system will react to commands. Example: Rosetta and Philae spacecrafts
Complex Scheduling
Algorithms allow for automated scheduling that maximizes
efficiency and/or resource time. Used for trains, public
transportation, logistics, and military operations. Example: Stottler Henke’s Aurora system used by Boeing
TIT
AN
AE
RO
SP
AC
E
GO
OG
LE
CU
RIO
SIT
Y
BO
EIN
G
Dealing with a Ton of Information and Decisions
Self-Driving Cars: The Long Road (Pun Intended)
How Far We’ve Gotten
”
“That's artificial intelligence.
How do I get around obstacles
and detours and noticing all
those things. That's quite a
task and that shows us how
far we've gotten through
simulating intelligence.”
Steve Wozniak
Our Goal
”
“We don't claim that the cars are
going to be perfect…Our goal is
to beat human drivers..”
Sergey Brin, Google Co-founder
Source: www.kurzweilai.net/fully-self-driving-cars-expected-by-2030-says-forecast; spectrum.ieee.org/automaton/robotics/artificial-intelligence/how-google-self-driving-car-works; www.techrepublic.com/article/wozniak-talks-self-driving-cars-apple-watch-and-how-ai-will-benefit-humanity/; www.inc.com/associated-press/google-self-driving-cars-accidents.html; Health Care IT Advisor research and analysis.
No Industry Knows Consumers and Leverages That Knowledge Better
Source: Health Care IT Advisor research and analysis.
IC/AI in Retail and Manufacturing
• What did you buy before?
• What did those like you buy?
• What do you buy seasonally
or after certain events?
• How much will you spend?
• How can we place the right items in
the right stores, warehouses, and
distribution centers, at the right
time?
• How can we get items to the
consumer more quickly?
• How can we build products
constantly – 24/7?
• How can we leverage Intelligent
design and 3-D printing?
• How can we predict maintenance
requirements?
• Which markets will be most
successful?
• What do particular segments want?
Logistics
Shopping Assistants/Agents Product Launches
Automated Factories
Example: Netflix, Operator from Uber
Example: Spiegel’s use of
NeuralWare,Inc,
Example: Quintiq
Example: Ford’s Global Study and
Process Allocation System (GSPAS)
SA
WY
ER
Milliseconds Determine the Winner and Identifying Risk Reduces Liability
Source: http://www.slideshare.net/0xdata/paypal-fraud-detection-with-deep-learning-in-h2o-presentationh2oworld2014; http://www.npr.org/2015/10/20/445337189/would-you-let-a-robot-manage-your-retirement-savings; Health Care IT Advisor research and analysis.
IC/AI in Financial Services
Auto-Trading
Split-second decisions
based on immense data
that involve billions of
dollars. These problems
are too complex to leave
to humans.
Example: TickCOM’s iSTRAT
Check Cashing
Image recognition of
checks facilitates
deposits with high
accuracy. More difficult
cases are bounced back
to humans.
Example: Cyphermint
Market Prediction and
Portfolio Management
Robo-advisers can
identify best investment
options and determine
how portfolios should
change. Example: Blooom
Fraud Detection
Involves many different
forms of modeling and
inputs from many sources.
Applies to credit cards,
mortgages, insurance risk,
etc.
Example: PayPal’s Fraud Detection
Case in Brief: Fraud Prevention at PayPal
• Transaction Level: Employs machine learning and statistical models to flag fraudulent behavior up-front; uses more sophisticated
algorithms once transaction is completed
• Account Level: Monitors account-level activity to identify abusive behavior; abusive patterns include frequent payments and
suspicious profile changes
• Network Level: Monitors account-to-account interaction; frequent transfer of money from several accounts to one central account
Source: Health Care IT Advisor research and analysis.
1) http://www.forbes.com/sites/neilhowe/2015/05/14/artificial-intelligence-paves-the-way-for-ambient-intelligence/
2) http://www.usatoday.com/story/tech/2015/07/06/health-care-technologies/29160173/
Who Will Be Changing Our World?
Amazon
Apple
Baidu
Google and Deep Mind
IBM
Microsoft
OpenAI (Musk and others)
SRI International
Toyota
BlueBrain Project
DARPA
HumanBrain Project
Big Data Research and
Development Initiative
Corporate Research Centers Government Agencies
$17B Amount invested in artificial
intelligence since 20091
Carnegie Mellon
Georgia Tech
MIT
Stanford University
UC-Berkeley
Universities
$2B Amount invested in digital
start-ups in 20142
2,000+ Number of start-ups that
have key words “digital
health” or are new health
care technologeies2
High Aspirations
”
Larry Page, Google “Oh, we’re really making an AI”
“Over a quarter of all attention and resources" at
the Microsoft Research main lab now focused on
AI-related activities” Eric Horbitz, Microsoft
Intelligent Information
Gathering and Sensing (IoT)
1
What do we know about
the patient and his
changing environment to
aid in his health?
Six Related Categories of Application Development and Use
2 Intelligent Interaction
and Service
3
How can we communicate
with our systems in a more
natural manor?
What’s wrong with the
patient and what type of
evolving treatment plan
would be most effective?
Intelligent Diagnosis
and Care Plans
Intelligent
Medical Devices
4
How can we automate and
adjust medical devices to
be more real-time,
accurate, and responsive?
5 Robotics
6
What roles can robots take
on to assist with the
mundane, dangerous, or
complex jobs of humans?
Advanced
BI/Analytics
What can we learn from our
data, and how can we
predict futures states and act
on that knowledge?
Applications of IC/AI in Health Care
Audience Response Polling Question 1
Where will IC/AI in HC will have the greatest impact?
a. Internet of Things
b. Decision Support/Diagnosis
c. Robotics and Smart Devices
d. Analytics
Enabling Situational Awareness and Action with IoT
Frameworks + AI-based Tools + Progressive Providers
“Ambiant” agent and
machine intelligence-based
platform provides alerting
and workflow management
processes
Provides systems Integration
and services, partner
ecosystem development, and
the “Intelligent Health System
Framework”
Opened North America’s
“first fully digital” medical
facility in Toronto, October
2015
Evaluates innovation in
real-world settings
Hospital Example:
“Code Blue”
• How is it triggered (connected
medical devices?)
• Who is it sent to (who is on the
care team?)
• Who is nearest with the right
skills (and able to respond?)
• When will they arrive?
• Who needs to bring what
devices (crash cart) or medical
supplies (and where are these
items?)
• Who else needs to be notified
and what are the ripple
effects?
Source: CGI; ThoughtWire; Mackenzie Innovation Institute; Humber River Hospital
Answers Questions, and Explains “Reasoning”
MD Anderson Patient Concierge Using Cognitive Computing
Source: MD Anderson, Houston, TX; Health Care IT Advisor research and analysis.
Can You Understand Me Now?
Speech and NLP Advances for Health Care
Source: Nuance; Health Care IT Advisor research and analysis.
Increased
processing
power, cloud,
virtualization
Increased data and
cases from millions of
conversations and
encounters (with
specific regions,
accents, specialties,
users)
Algorithms that look at
words, snippets, and
sentences, and
increasingly, paragraphs,
documents, the EHR
(medical ontologies), and
additional context
Advances in probabilistic
(e.g., Max-Entropy Markov
Models) and logical (rules)
reasoning approaches and
the ability to use them in
combination (and in
multiple passes)
Advances and
investments in
consumer
digital
assistants (Siri,
Cortana, Google
Now, Alexa, M)
Human machine interaction – still a long way to go
Medical NLP and fact extraction– between “crawl”
and “walk” – already providing useful functionality
Medical speech recognition for specialties – at the
“run” phase - on par with human accuracy –- 97-99%
Menu-driven commands – really good – almost
flawless Better built-in
mobile device
microphones,
signal processing
Current State of Achievement
IBM Watson Health Launched in 2015 – Cognitive Computing
Company in Brief: IBM Watson Health (Part of IBM Watson Group)
Technical Approach
• Uses hundreds of computational techniques, including machine learning; conducts
NLP queries on structured and unstructured data; generates hypotheses, scores
evidence, and returns answers
• Uses IBM DeepQA software, Apache UIMA Architecture, clusters of Linux servers, and Hadoop
Key Factors for Success
• Focuses on breadth and depth scale, combination of approaches, and parallel processing
• Supports partner development with APIs, offers cloud capabilities
Feb 2011: Nuance,
Columbia University,
University of Maryland
Oct 2012: Cleveland
Clinic, Case Western
Reserve University
Feb: Memorial Sloan
Kettering, WellPoint,
Maine Center for
Cancer Medicine
Oct: MD Anderson’s
Moon Shot Program
Jun 2014: GenieMD
Mar: Modernizing
Medicine
Apr: IBM Watson Health
established; Apple,
Johnson & Johnson,
Medtronic; acquires
Explorys, Phytel
Jul: CVS
Aug: Acquires Merge
Healthcare
Sep: Boston Children's
Hospital, Columbia
University Medical Center,
ICON plc, Sage
Bionetworks, Teva
Pharmaceuticals
2011 – 2012 2013 – 2014 H1 2015 H2 2015
Source: IBM
Intelligent Medical Devices: Reducing Workloads
Case in Brief: Anesthesiology Automation—
Johnson & Johnson Sedasys
• FDA approval in 2013 for “narrow use” with expert available
(uses propofol)
• In use at four U.S. hospitals for colonoscopies and
endoscopies
• Business case: Anesthesiologist requires four years of
medical school and a median salary of $277K per year
• Now being tested for heart and brain surgery
Case in Brief: Artificial Pancreas and Smart
Infusion Pumps—Medtronic MiniMed Connect
• SMARTGUARD mimics some functions of a healthy pancreas;
predicts low glucose levels in advance and stops pump
• Insulin pump and continuous glucose monitoring can talk
directly to smartphone
• Partnered with Samsung
Source: Medtronic; Johnson & Johnson
Robotics: To Serve (and More)
Forecasted Impact from Robotics
$67B Spending on
robots in 2020 22% Reduction in U.S. labor
costs in by 2025
Hospital-Based
Robots
University of California
San Francisco at Mission
Bay uses 25 TUG Robots
by Aethon. They travel
481 miles per day in
1,300 trips, equating to a
time savings of 315
hours.
Similarly, Yujin Robots
can deliver drugs, linens,
and meals, and also cart
away medical waste,
soiled sheets, trash.
Robotic
Assistants
Developed in Japan,
the latest generation of
the Robobear medical
assistant can lift
patients into and out of
beds, help position
humans into sitting and
standing positions, and
lift patients from
wheelchairs.
Telepresence
Partner’s HealthCare
uses Vecna’s VGo
robots to provide
remote care to
children in their
homes. The robot
can do “rounds” on
the patient every
day, taking pictures
and gathering data
to track progress.
Aethon TUG Vecna VGo
Pets
Huggable is a
collaboration between
Boston Children’s
Hospital and MIT. The
social robot prototype
recently started a 90-
person study to
determine whether it has
therapeutic value for
children enduring long
hospital stays.
Another example is
Paro, the roboseal,
developed by the
Japanese firm AIST.
Home Assistants
GiraffPLUS, from the
European Union,
combines a network of
sensors that collects
physiological and
environmental data with
a telepresence robot for
social interaction. The
data is fed wirelessly to
doctors and utilizes
Skype to conduct remote
doctor consultations. It’s
geared toward older
patients who live alone.
Huggable GiraffPLUS RIBA Robobear
Source: http://www.cnbc.com/2015/07/06/robot-use-on-the-rise-through-2025.html
New Data and New Tools Bring Better, Stronger, and Faster Predictions
Source: http://www.ayasdi.com/; http://www.ayasdi.com/wp-content/uploads/2015/02/Healthcare_Mount_Sinai_Solution_Brief.pdf; Health Care IT Advisor interviews and analysis.
BI Analytics
Company in Brief: Ayasdi
• Headquartered in Menlo Park, CA
• Grown out of the mathematics
department of Stanford University and
initially funded by DARPA
• Uses Topological Data Analysis and
machine learning to capture the “shape
of the data”
• Applying multiple algorithms eliminates
approach biases and reveals previously
obscured patterns
–In Mt. Sinai’s diabetes study the data
exhibited clusters, loops, flares, and
line patterns revealing several distinct
subgroups that were previously
unidentified
• Collaborators: Mercy, UCSF, Merck,
Michael J. Fox Foundation, FDA
Ayasdi Topological Map for Mt. Sinai Diabetes Study
Ayasdi aims to make complex data
useful for healthcare providers and
payers through Machine Intelligence,
which represents the next generation of
healthcare data analytics.” Gurjeet Singh, CEO and Co-Founder. Ayasdi
Audience Response Polling Question 2
Will your job be eliminated by IC/AI in 10 years?
a. Highly unlikely
b. Possibly
c. Highly likely
d. All ready happened
Major Challenges to IC/AI in Health Care
Complexity: Medical issues don’t
appear in isolation and coordination
of care is difficult.
Business Challenges Legal and Ethical Challenges
Threat to human jobs: Strong fear
associated with technology displacing
human workers.
Cost: The high costs for
developing, testing, certifying, and
implementing can be a barrier.
Workflow: How do AI solutions fit
into existing workflows? How much
effort is required to use it? Does it
interfere or annoy unnecessarily?
Competing Priorities: EHRs,
portals, Meaningful Use, Payment
Report, ACOs.
Regulation: Health IT regulations
are hotly debated at the national
level. Finding the right balance of
public health protection and
fostering innovation is key.
Legal: Juries still award large sums
when health care is not applied
properly or expected outcomes are
not achieved.
Liability: How do we deal with
computer failings? It raises the issue
of data de-identification, privacy,
security, and espionage.
Human Touch: How will we interact
with AI? How strongly will we require
the human touch and human
compassion in health care?
Audience Response Polling Question 3
What is the largest barrier to IC/AI in healthcare?
a. Technical
b. Clinical complexity
c. Costs, skills
d. Regulation, legal, ethical
Can All These Brilliant Minds Be Wrong?
Source: Health Care IT Advisor research and analysis.
You Know You’re Onto Something When…
The development of full artificial intelligence could spell the end of
the human race.” Stephen Hawking, Theoretical Physicist; author of
A Brief History of Time
With artificial intelligence we are summoning the demon.”
Elon Musk, Founder of Tesla Motors,
SpaceX, and PayPal
I agree with Elon Musk…and don't understand why some people
are not concerned.” Bill Gates, Founder of Microsoft
and the Bill & Melinda Gates Foundation
AI Doomsday?
IC/AI Scenario Planning: Where Will We Be in 20 Years?
AI Fizzles
No Major
Breakthroughs
Every Company
Loves You
Promises, Promises
Battle of the Giant
Intelligences
Niche Advantages
“Do they have your best
interests in mind?
Which AI-run governments,
corporations, and systems will
dominate?
How many more times must we
open our pocketbooks ?
Intelligent curiosity or
secret weapon?
AI Super
Intelligence
Singularity and
Consciousness
AI Limited
Niche Companies and Research
Al Ubiquitous
All Major Corporations
Audience Response Polling Question 4
Do you welcome AI or do your fear it?
a. Fear it will be our downfall
b. Concerned – highly cautious
c. Excited by it – welcome it
d. Love it – humanity’s savior
Graphic
Steps to Intelligent Computing/AI Success
Combine the experience, knowledge, and human
touch of clinicians with the power of intelligent
computing to achieve more than either alone
Use Intelligent Computing to provide higher levels of
patient engagement and education, such as adaptive,
personalized response and gaming
Use intelligent computing to tackle the complexity and
expanse of new data sources to push the boundaries
of precision medicine and population health
Summary/Key Takeaways
Satisfaction
Treatment/Clinical
Electronic
Information/Data
Prevention and Patient
Education
Savings
Focus on the advantages of intelligent computing –
these systems should be viewed as assistants, not
threats
Use IC to reduce labor costs, increase consistency,
discover new clinical knowledge, and offer scalable
return on investment for value- and risk-based care
31
Final Thought 32
Our technology, our machines, is part of our humanity. We created them to extend ourselves, and that is what is unique about human beings.”
Ray Kurzweil
Questions
Kenneth A. Kleinberg, FHIMSS
Managing Director, Research & Insights
The Advisory Board Company
2445 M St NW, Washington, DC 20037
202-266-6318
Twitter: @kkleinberg1
33