artificial intelligence and healthcare at the crossroads
TRANSCRIPT
Artificial Intelligence and
Healthcare at the CrossroadsGuy Barnard, MA, MBAChief Executive Officer, Co-Founder, Synchronous Health, Inc
September 2017
“The artificial intelligence will see you now”
Executive Summary
Models and approaches to Health are not solving the problem.
Status quo is not a viable strategy.
Technology has repeatedly failed to deliver the promised benefits, and so any
emerging trend can and should be met with skepticism.
Advances in technology, expansion of consumerism, infusion of capital, and
entries of new players, have led to much noise and confusion.
What’s different now is that stakes are higher and timelines so compressed, that
understanding and preparation are essential.
In fact, some will argue our sustained existence may depend on it, but regardless,
the essence of success may be simpler than one might think, but the pitfalls
unfortunately easier.
Consider these battle learnings from a field that is still emergent.
Why the explosion in AI?
Source: VentureScanner database of 826 companies in Artificial Intelligence Category
Analogy: The Cambrian Explosion,
500 million years agoDramatic burst of transformational changes
Ninety percent of the
animal phyla that exist
today appeared in this
short window of time
Deep LearningStatisticalbased AI
Rules based AI
GraphicsProcessing Units (GPU)
Von NuemannArchitecture
Moore’s Law
Natural Sight andLanguage
Bots
NLP Killer App
EdgeComputing
Cloud
Compute
VirtualizedExperience
Empathy
Self-Service
Multiple crossroads
leading to the
AI Cambrian
Explosion
After 240 minutes of
training.
This is where the magic
happens. It realizes that
digging a tunnel through
the wall is the most
effective technique.
After 120 minutes of
training
It plays like an expert!
After 10 minutes of
training
The bot tries to hit the ball
back but is too clumsy to
score much.
Example: Deep LearningGoogle’s DeepMind
This bot plays Atari breakout.
It is given nothing more than
the pixels on the screen and
the goal to optimize the score.
Video plays in slideshow mode
Source: Nature Vol 518, 26 Febuary 2015; http://tinyurl.com/atariai accessed August 30, 2017
MILEPOST
Example: Moore’s Law . . . .
1965, “Electronics” Magazine4 data points 120 years of data points
2016 Kurzeil / Jervetson Update
GPU
Specialized electromagnetic
Hollerith
TabulCator
von Neumanarchitecture
IBM
Blue Gene
NVIDIA
GTX
$199
SPEEDOMETERSource: Electronics (1965); Steve Jervetson Moore's Law over 120 Years (December 2015), DFJ Venture Capital
. . . . Extended by Gaming
Applications in Healthcare ?Expert Augmentation, Clinical Efficacy and Efficiency,
Consumer Engagement, Enabling processes and more
Source: Association of American Medical Colleges , The Complexities of Physician Supply and Demand 2016 Final Report, IHS analysis;Accenture Artificial Intelligence: Healthcare’s New Nervous System
700,000
750,000
800,000
850,000
900,000
950,000
2017 2019 2021 2023 2025
Supply Demand Shortfall
US Clinician Demand vs. Supply Forecast, 2017 –
2025
15-20% unmet demandaddressableby AI
Myths & Detours
Always need experts to operate at top of
license
Creativity and human-to-human work flows
cannot be digitized
AI in healthcare is about robotic surgery and
amazing diagnoses/decisions
AI at scale is too expensive and will
generate huge cloud invoices
Privacy is the same concern it was last
year
AI brings less bias and more objectivity
The biggest safety risk is a bad decision
that leads to a poor outcome
Myth FactEvidence overwhelming that relying on data
and algorithms* usually leads to better
decisions than even expert humans
Moving from the Art of Medicine to the
Science of Lifestyle Health (I)
Your members/employees/patients are
increasingly turning to virtualized
experiences (ATM, self-checkout, etc.)
Creative abilities are expanding rapidly
Always need experts to operate at top of
license
Creativity and human-to-human work flows
cannot be digitized
AI in healthcare is about robotic surgery
and amazing diagnosis/decision support
Back-office processes may be some of the
biggest early wins; (re)think recruiting,
revenue cycle management, underwriting,
fraud, etc.
Consumer engagement, prevention and
post-acute are ripe for disruption
* Where data and algorithms are available
Myth Fact
AI at scale is too expensive and will
generate huge cloud invoices
The amount of compute power available to
startup in a garage is unprecedented; open-
source, as-a-service, or on-device
Moving from the Art of Medicine to the
Science of Lifestyle Health (II)
Privacy is the same concern it was last
year
Breadth of data combined with the compute
power to re-associate make security an
exponentially bigger challenge
AI brings less bias and more objectivity Biases are often codified in training
datasets; the AI models start as just that --
models of existing work flows
The biggest safety risk is a bad decision
that leads to a poor outcome or even
death
No facts yet on this one, but plenty of
concerns about potentially way more;
planning and research advised
Questions to Consider
Do you have an AI strategy? Who is responsible for machine learning in your
organization? For back-office business processes too?
Are you systematically tracking the performance of the decisions by people and
algorithms? What data should you start collecting?
Which key decisions or operations, if any, would you consider turning over
entirely to AI? Which will be the hardest to turn over? Why?
Which work flows require an empathetic understanding of the human condition?
Should any of these be shifted in time or space?
What are the safety considerations? Is the current governance process
sufficient? Are you monetizing data or adopting “private by design”?
What are the skills/job types you will need in 2018? in five years?
What is your edge computing strategy? Is augmented reality in your facility or
on your roadmap? Where will vision/speech have the most impact? What are
the opportunities from new sensors / inputs ?
STRATEGY
DATA
PROCESSES
PEOPLE
SAFETY
TECHNOLOGY
Source: Synchronous Health AI strategy
GPS MAP
Field Learnings in Population Health (I)Context
Screen
Match
Engage
Measure!
FROM
20,000 business rules
$1B platform investment
Annual IT investment /
operating cost $125M
600 IT staff
Wellness portal
+ 20 apps
11 call centers
Conditions and risks
TO
Machine learning + human-
enabled AI
Bots not docs not apps
Human compassion
+ AI technology
Lifestyles, habits, behaviors
Serverless
Empathy screening of
specialists, bot interviews
Digital anthropologists,
Chief Human and Digital
Talent Officer
Aim: Reduced Costs. Improved Performance. Improved Health. Better Experience. Source: Barnard experience
Intent: Track feelings
Mood: sadness
GPS: 36.1246085,-86.8487669, 48°
Time: 2016-09-
16T19:20:30.45+01:00
People nearby: Kai, Ailani
Media: It’s My Life, Bon Jovi, 65db
Ambient light: Outside Movement:
Walking
Weather:Raining
Steps: 354 steps walked today
Location context: Home
Day of week context: Normally at work
Clip: 92352352352352.mp415
Evidence-based design,on proven methodologies
Designed for behaviors of consumers,not for a healthcare system
Accessibleand convenient
Advanced technology thatgets out of the way:
human experts that operate at top of license, with empathy,
powered by AI
Prove-as-you-go metrics
Life context specificity: Leverage tiny data, not big data,
to deliver highly targeted interventionsin the moment someone needs it,
at scale
Simple on-ramps, withaligned benefit/incentive design
Lifestyles
Behaviors
Habits
1 2 3
4
5
6
7
Field Learnings in Population Health (II)Seven Tenets that Seem to Matter
Karla, I am feeling sad.
Intent: Track feelings
Mood: sadness
GPS: 36.1246085,-86.8487669, 48°
Time: 2016-09-
16T19:20:30.45+01:00
People nearby: Kai,Ailani
Media: It’s My Life, Bon Jovi, 65db
Ambient light: Outside Movement:
Walking
Weather:Raining
Steps: 354 steps walked today
Location context: Home
Day of week context: Normally at work
Clip: 92352352352352.mp4
Okay, I’ve tracked your feelings of sadness
for your specialist.
Example (I): Human-Enabled AI in Support of
Depression
Humanslabel data
Machines learn fromlabeled data
competitive advantage from data +
networked humans
AI
Example (II): Human Compassion Bridges Time,
Space, and “Uncanny Valley”
Questions to Consider (Recap)
Do you have an AI strategy? Who is responsible for machine learning in your
organization? For back-office business processes too?
Are you systematically tracking the performance of the decisions by people and
algorithms? What data should you start collecting?
Which key decisions or operations, if any, would you consider turning over
entirely to AI? Which will be the hardest to turn over? Why?
Which work flows require an empathetic understanding of the human condition?
Should any of these be shifted in time or space?
What are the safety considerations? Is the current governance process
sufficient? Are you monetizing data or adopting “private by design”?
What are the skills/job types you will need in 2018? in five years?
What is your edge computing strategy? Is augmented reality in your facility or
on your roadmap? Where will vision/speech have the most impact? What are
the opportunities from new sensors / inputs ?
STRATEGY
DATA
PROCESSES
PEOPLE
SAFETY
TECHNOLOGY
Source: Synchronous Health AI strategy
GPS MAP
Where Opportunity
and Preparation Meet
AI at the Crossroads
“Once this
Pandora’s box
is opened, it will
be hard to
close”
Source: Open letter to the united nations convention on certain conventional weapons, August 2017
“Let's say you create a self-improving AI to pick strawberries and it gets
better and better at picking strawberries and picks more and more and it
is self-improving, so all it really wants to do is pick strawberries. So
then it would have all the world be strawberry fields. Strawberry fields
forever.”
“Our biggest existential threat”
“The development of full artificial
intelligence could spell the end of
the human race”
“We need to rethink the way
we have built society on top
of the web”
“I don't understand why some
people are not concerned.”
“The future is scary and very bad for
people”
“It’s more important right now to build
consensus in the industry and academia
around what are the things that would have a
chilling effect.”
So keeping AI beneficial is
probably important.
But can we afford not to be
using it now?
Where Opportunity
and Preparation Meet
AI at the Crossroads