2. Artificial Intelligence (AI) in everyday use
• In the past, humans would recognise patterns in data, hand design
and write the steps (rules) in to a computer programme
• Now the AI looks for patterns, and from those helps create the rules
• Yes, you have used Ai this week:
• Google home, Amazon Alexa, Siri
• Google search, maps, traffic
• Credit card fraud detection, Superannuation stock market trading
• Recommendations on Netflix, shopping sites, Facebook advertisements
2
3. What is AI?
• Use of computer-based algorithms to perform tasks that would
require intelligence if performed by a human
• Includes predictive analytics, voice recognition,
language processing, image recognition, robotics
• And different learning techniques:
• supervised learning
• unsupervised learning
• reinforcement learning
3
5. Digital
Disruption
4th industrial
revolution*
• Machine learning / AI
• Big data analytics
• Internet of things
+ https://www.weforum.org/agenda/2016/01/the-fourth-industrial-revolution-what-it-means-and-how-to-respond/
• Desire for improved
health outcomes
• Increasing health
delivery costs
• Limited public funds
Healthcare Delivery
Transformation
• Role of health care providers
• Augmenting decision-making
processes
• Transforming hospital delivery
7. Industry AI Advantage
Source: ARTIFICIAL INTELLIGENCE: THE NEXT DIGITAL FRONTIER?, McKinsey Global Institute, June 2017,
http://www.mckinsey.com/~/media/mckinsey/industries/advanced%20electronics/our%20insights/how%20artificial%20intelligence%20can%20deliv
er%20real%20value%20to%20companies/mgi-artificial-intelligence-discussion-paper.ashx
9. • The factory of the future will
have only two employees,
a man and a dog.
The man will be there
to feed the dog.
The dog will be there to keep
the man from touching the
equipment.
12. Will Robots and AI take my job?
https://mashable.com/2017/05/31/website-tells-you-if-robots-will-take-your-job/#0XKmcnBtyOqN
https://free.vice.com/en_us/article/3kjy3k/job-bots-automation-future
https://www.isolatedtraveller.com/check-if-your-job-will-be-taken-over-by-robots/
https://www.thrillist.com/news/nation/will-robots-take-my-job-website-estimates-employment-future
13. Will Robots and AI take my job?
https://mashable.com/2017/05/31/website-tells-you-if-robots-will-take-your-job/#0XKmcnBtyOqN
https://free.vice.com/en_us/article/3kjy3k/job-bots-automation-future
https://www.isolatedtraveller.com/check-if-your-job-will-be-taken-over-by-robots/
https://www.thrillist.com/news/nation/will-robots-take-my-job-website-estimates-employment-future
14. Will Robots and AI take my job?
https://mashable.com/2017/05/31/website-tells-you-if-robots-will-take-your-job/#0XKmcnBtyOqN
https://free.vice.com/en_us/article/3kjy3k/job-bots-automation-future
https://www.isolatedtraveller.com/check-if-your-job-will-be-taken-over-by-robots/
https://www.thrillist.com/news/nation/will-robots-take-my-job-website-estimate-employment-future
15. Will Robots and AI take my job?
https://mashable.com/2017/05/31/website-tells-you-if-robots-will-take-your-job/#0XKmcnBtyOqN
https://free.vice.com/en_us/article/3kjy3k/job-bots-automation-future
https://www.isolatedtraveller.com/check-if-your-job-will-be-taken-over-by-robots/
https://www.thrillist.com/news/nation/will-robots-take-my-job-website-estimate-employment-future
20. Will we really replace entire Jobs?
• Most jobs are a series of related tasks
• In time individual tasks can be replaced, rather than jobs
• Many such tasks are monotonous and error prone
• So we will move from a series of repetitive simple tasks
to more variable, complex, cognitive tasks
21.
22. • An aging workforce is changing job expectations
23.
24. Cognitive Input per patient is and will be limited,
Care gap emerges without extra help.
2020 2025 2030 2035 2040 2045 2050
Total Care requirements
Care addition from clinicians
Care from pathways, DSS, AI
Growing care gap
24
25. Clinical decision support needs to grow to fill the gap.
2020 2025 2030 2035 2040 2045 2050
Total Care requirements
Care addition from clinicians
Care from pathways, DSS, AI
Need to grow clinical
decision support
25
26. Diabetic retinopathy (2017)
• 54 opthalmologists on panel, 3-7 per image
• 128,000 classified images fed in to tensor flow
• Results equivalent to human
• F-score 0.95 vs human 0.91
• Deploying in to India
• So provides resources where they don’t exist
• Shows levels of human fallibility
26
https://research.googleblog.com/2016/11/deep-learning-for-detection-of-diabetic.html
https://jamanetwork.com/journals/jama/fullarticle/2588763
CNN, Inception –v3 architecture
Optimization by distributed stochastic descent (Dean)
27. TREWScore
• Targeted Real-time Early Warning Score to predict septic shock
• MIMIC=II database, 13,014 patients (1836 septic)
• Supervised machine learning technique
• AUC 0.83, sensitivity 0.85 at specificity 0.67
• Mean detection 28 hrs before onset
of septic shock
• 2/3rds identified before onset of
sepsis-related organ dysfunction
27
28. C-Path - Stanford
• Machine learning to diagnose breast cancer
• Iterative process using known data set
• Better than pathologist eventually looking at cancer cells
• AND
• Found patterns in the stroma, which combined with the cell data
was a better predictor.
Sci Transl Med 9 November 2011: Vol. 3, Issue 108, p. 108-113
Systematic Analysis of Breast Cancer Morphology
Uncovers Stromal Features Associated with Survival
28
6642 features considered
L1 regularised logistic regression
31 features tested
8 fold cross-validation
Multivariate cox proportional hazard
29. Breast Cancer metastases (lymph nodes)
• 400 Gigapixel images, tensorflow CNNs
• AUC 0.925 (human = 0.966)
• Combined = 0.995
• Augmented Intelligence,
combining human and AI,
is the key to success
29
https://research.googleblog.com/2017/03/assisting-pathologists-in-detecting.html
https://drive.google.com/file/d/0B1T58bZ5vYa-QlR0QlJTa2dPWVk/view
https://arxiv.org/pdf/1606.05718.pdf
Used GoogLeNet, AlexNet,
VGG16, FaceNet
Won Camelyon16 both slide and lesion
30.
31.
32.
33. The serial job hopper – a new norm
• The job marketplace is changing, and will continue to change
• Within a single jobs, roles will constantly change
• Regularly changing jobs will not have associated stigma
• Continuous learning, broader view, regular new challenges
• AI itself is the fastest area of change
34. Find good oil
Drill and retrieve oil
Store oil
Refine oil
Turn to power
Create cars, electricity
Use cars, electricity
Sources of unbiased data
Readable structured data
In useable databases
Create data views
Turn data to knowledge
Turn knowledge to products
Engage users and customers
Domain experts
Engineers
Cloud architects
Database administrators
AI programmers
User interface design
Change management
35. INTELLI
a partnership between GCUH,
Universities and Industry to
transform healthcare through the
practical application of next
generation technologies.
1. Improve Health Outcomes
2. Create local opportunities for skills, jobs and venture development
3. Attract Entrepreneurs & Investment
4. Achieve Global Recognition for Innovation & Commercialisation