Dr Kelvin Ross discussed the rise of AI and data science and its impact on ethics and governance. He has over 25 years of experience in advanced technology and has participated in both successful and unsuccessful startups. He discussed how data is used in areas like driverless cars, precision medicine, and ICUs. Topics around AI included machine learning, deep learning, datasets, and the "super model." Challenges around scaling, bias, overfitting, and privacy were also covered. The talk concluded with a discussion of governance issues like oversight, explainability, appeal, and the need for regulation as AI impacts more areas of society.
The Rise of AI and Data Science: Ethics and Governance Challenges
1. ACS EMERGING & DEEP TECH WEBINAR:
THE RISE OF AI AND DATA SCIENCE
AND ITS IMPACT ON ETHICS AND
GOVERNANCE
Dr Kelvin Ross
kelvin.ross@kjr.com.au
2. Dr Kelvin Ross is an entrepreneur, technologist and researcher. He currently holds a number of roles,
including Founder & Chairman of KJR, a mid-tier IT consultancy, Adjunct Associate Professor in Intelligent
and Integrated Systems at Griffith University, and Director at IntelliHQ, a non-profit innovation Centre
focused on Artificial Intelligence in healthcare at Gold Coast University Hospital, and a medical research
data platform startup, Datarwe.
He has over 25 years of experience in advanced technology commencing with safety-critical systems in
the military, then moving on to transportation, banking, financial markets, government and healthcare
systems. He has participated in several successful and unsuccessful technology startups, as well
numerous successful and unsuccessful technology implementation programmes in medium and large
enterprises.
7. 0
kph
00
arterial
Stopped at traffic light
straight
day
PREDICT
Fatigue
+15 mins
+30 mins
+1 hr
+2 hr
Stress
Heart Arythmia
Stroke Risk
Diabetes Risk
8. Davos 2017
"You have to think about how you
develop trust for any new technology”
Ginni Rometty, IBM CEO
"There are some things that machines
can do that perhaps they shouldn't do”
Satya Nadella, Microsoft CEO
27. Training Bias
• Algorithm reflecting
underlying training
data bias
• Institutionalised behaviour
captured in training data
• Reinforced in feedback
loop
• Hidden variables
• Less explicit correlations,
e.g. postcode
http://wpj.dukejournals.org/content/33/4/111.full.pdf+html
33. 1. Oversight
2. Explainability
3. Appeal
http://www.innovationaus.com/2017/09/
Medcraft-on-ethics-and-algorithms
“Businesses and government
must be held accountable for their
use of computer algorithms and big
data and transparency should be
the first priority”
Greg Medcraft
ASIC Chairman
34. The app says you
require surgery in
the next 24 hours.
How did it
decide that?
It doesn’t say, but it
does have an AUC
of 0.94!
35. The app says
you require
surgery in the
next 24 hours as
your blood
pressure is low,
ECG rhythm is
irregular, and
protein test
returned positive.
47. Researchers design patch to make people ‘virtually invisible’ to AI detectors
https://www.computerworld.com.au/article/660283/researchers-design-patch-make-people-virtually-invisible-ai-detectors/
49. General
Data
Protection
Regulation
• Effective 25 May 2018
• Automated individual decision-making, including profiling (Article 22) is made
contestable. Citizens now have the right to question and fight decisions that
affect them that have been made on a purely algorithmic basis.
https://en.wikipedia.org/wiki/General_Data_Protection_Regulation
50. Australian Data Sharing
• Privacy Acts (Federal and States)
• Consumer Data Rights
• https://consumerdatastandards.org.au
• https://www.oaic.gov.au/consumer-data-right/about-the-
consumer-data-right/
• Office of the National Data Commissioner
• https://www.datacommissioner.gov.au/resources/discus
sion-paper
54. Australian Human Rights Commission
https://tech.humanrights.gov.au/sites/default/files/2019-
55. Medical Big Data & AI
• Challenging
• TGA SAMD
Classification
• Health Acts / Privacy
Acts
• Human Research
Ethics Committee –
NHMRC
• Site Specific
Agreements with
Hospitals
https://www.tga.gov.au/sites/default/files/presentation-proposed-reforms-
regulation-software-including-software-medical-device-consultation-results.pdf
https://www.tga.gov.au/sites/default/files/submission-consultation-regulation-
software-including-software-medical-device-samd-ranzcr.pdf
59. UK NHS - Code of conduct for data-driven
health and care technology
1. Understand users, their needs and the context
2. Define the outcome and how the technology will contribute to it
3. Use data that is in line with appropriate guidelines for the purpose for which it
is being used
4. Be fair, transparent and accountable about what data is being used
5. Make use of open standards
6. Be transparent about the limitations of the data used
7. Show what type of algorithm is being developed or deployed, the ethical
examination of how the data is used, how its performance will be validated
and how it will be integrated into health and care provision
8. Generate evidence of effectiveness for the intended use and value for money
9. Make security integral to the design
10. Define the commercial strategy
https://www.gov.uk/government/publications/code-of-conduct-for-data-driven-health-and-care-technology/
initial-code-of-conduct-for-data-driven-health-and-care-technology
60. Software 2.0 Stack Andrej Karpathy, Tesla
https://vimeo.com/274274744
Supporting developers to learn optimal rules (ML architectures and
weights) from example inputs and outputs
2 main areas supporting teams:
- Label
- Maintain surrounding “Data Infrastructure”
- Visualise datasets
- Create/edit labels
- Bubble up likely mislabeled examples
- Suggest data to label
- Flag labeler disagreements
- ...
Software 1.0
Software 2.0
Supporting developers to write rules (programs) to produce
outputs from inputs
E.g. IDEs, Test Automation
63. Modelling Evaluation Deployment Surveillance
TRAIN
TEST
Model
A/B
Test
Apply To
Production
Data
Monitor
Production
Interactions
Initial
Data
Further production data iteratively evolves model
Phase 1 Phase 2 Phase 3 Phase 4
Lab Single
Site
Multi
Site
Market
Surveillance
Clinical Trial Model
64. Closing Remarks
1. Extend human capability using
big data and AI
2. New classes of risk
3. Regulation is inevitable
4. Assurance, Test & Evaluation
will be key
69. Uncanny Valley
The concept of the uncanny valley suggests humanoid objects which appear almost, but
not exactly, like real human beings elicit uncanny, or strangely familiar, feelings of
eeriness and revulsion in observers.
https://en.wikipedia.org/wiki/Uncanny_valley
Editor's Notes
The opportunity
Where data can be used – emphasise collaboration with ANZICS data group, and Melbourne institutions
Andrew Ng, ex Baidu head of AI
https://www.youtube.com/watch?v=21EiKfQYZXc
33:00