FGAI4H-C-002 Lausanne, 23-25 January 2019 Session
Description: FGAI4H-C-002 Lausanne, 23-25 January 2019 Session 1: Focus Group on AI for Health AI is an extremely dynamic, open field Unlike much of the technology industry, medical research cannot move fast and break things because it cant afford to
Related Topics
Download Presentation
"FGAI4H-C-002 Lausanne, 23-25 January 2019 Session" is the property of its rightful owner. Permission is granted to download and print the materials on this website for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Presentation Transcript
slide1. FGAI4H-C-002 Lausanne, 23-25 January 2019<br>
slide2. Session 1: Focus Group on AI for Health AI is an extremely dynamic, open field
Unlike much of the technology industry, medical research cannot “move fast and break things” because it can’t afford to fail
Currently no established ways to compare AI models (“Wild West”) and no common language or metric of success between AI and health
For worldwide adoption of AI in health, a standardized system of evaluating AI is required
The FG process would add substantial clarity to the field
Next step is to submit your data and projects!<br>
slide3. Session 2: Use Cases and Data Availability (1/2) Target local priorities and involve your workforce and population well
Shortage of doctors in China and India, especially in rural areas
AI validated in India is directly applicable to nearly half of the world
Work within established regulation and relevant frameworks
Use AI to empower hospitals and industry
Need to prepare the healthcare workforce through education and training to deliver the digital future<br>
slide4. Session 2: Use Cases and Data Availability (2/2) In the short term, descriptive AI is the most likely to be helpful
Data is the smallest and probably the most important unit of the AI ecosystem
The most important part of clinical practice is knowing “why”, more than “what” or “how”<br>
slide5. Session 3: Benchmarking and Security (1/2) Medical health data is valuable and thus a target to criminals
Increasing amount of medical data breaches (~5/week, each affecting 500+ people)
“Poisoning attacks” against algorithms are available on the darknet
Adversarial learning: When the adversary can tamper with the model itself
Data can be a trade secret if it creates a competitive advantage on the basis of being secret<br>
slide6. Session 3: Benchmarking and Security (2/2) The law voices societal concerns and is a medium for learning about these concerns
The law can also structure and influence the design process of IT-systems
E.g. GDPR - cybersecurity by design
Human rights and equality law are effective frameworks for negotiating and deciding upon questions of fairness
EU-wide strict product liability not applicable to software...which AI could be considered as<br>
slide7. Session 4: Regulations and Country Priorities for use of AI for Health Digital trust = Computational + operational trust
Countries are often not comfortable giving data to non-UN or non-high-trust institutions
Despite advances in ICTs, countries sometimes still rely on traditional methods for data sharing and reporting which can cause inefficiencies and inaccuracies
E.g. recording by hand
Health solutions can be based on behaviour change theory and societal contexts<br>
slide8. Session 5: Funding of AI for Health There is a spectrum of types of funding (“giving” “investing”) which is slowly blending
Investments tend to be more freely available when there are obviously higher returns
We have very poor data on philanthropy
Health & reproductive health is the main recipient of foundation grants
Crowdfunding -> emotional; philanthropic -> strategic; public -> replication
Through growing, philanthropy is overshadowed by public giving
Through their giving, foundations “buy” future impact in their areas of interest/focus
Looking at soundness of the project and solidity of the organization<br>
slide2. Session 1: Focus Group on AI for Health AI is an extremely dynamic, open field
Unlike much of the technology industry, medical research cannot “move fast and break things” because it can’t afford to fail
Currently no established ways to compare AI models (“Wild West”) and no common language or metric of success between AI and health
For worldwide adoption of AI in health, a standardized system of evaluating AI is required
The FG process would add substantial clarity to the field
Next step is to submit your data and projects!<br>
slide3. Session 2: Use Cases and Data Availability (1/2) Target local priorities and involve your workforce and population well
Shortage of doctors in China and India, especially in rural areas
AI validated in India is directly applicable to nearly half of the world
Work within established regulation and relevant frameworks
Use AI to empower hospitals and industry
Need to prepare the healthcare workforce through education and training to deliver the digital future<br>
slide4. Session 2: Use Cases and Data Availability (2/2) In the short term, descriptive AI is the most likely to be helpful
Data is the smallest and probably the most important unit of the AI ecosystem
The most important part of clinical practice is knowing “why”, more than “what” or “how”<br>
slide5. Session 3: Benchmarking and Security (1/2) Medical health data is valuable and thus a target to criminals
Increasing amount of medical data breaches (~5/week, each affecting 500+ people)
“Poisoning attacks” against algorithms are available on the darknet
Adversarial learning: When the adversary can tamper with the model itself
Data can be a trade secret if it creates a competitive advantage on the basis of being secret<br>
slide6. Session 3: Benchmarking and Security (2/2) The law voices societal concerns and is a medium for learning about these concerns
The law can also structure and influence the design process of IT-systems
E.g. GDPR - cybersecurity by design
Human rights and equality law are effective frameworks for negotiating and deciding upon questions of fairness
EU-wide strict product liability not applicable to software...which AI could be considered as<br>
slide7. Session 4: Regulations and Country Priorities for use of AI for Health Digital trust = Computational + operational trust
Countries are often not comfortable giving data to non-UN or non-high-trust institutions
Despite advances in ICTs, countries sometimes still rely on traditional methods for data sharing and reporting which can cause inefficiencies and inaccuracies
E.g. recording by hand
Health solutions can be based on behaviour change theory and societal contexts<br>
slide8. Session 5: Funding of AI for Health There is a spectrum of types of funding (“giving” “investing”) which is slowly blending
Investments tend to be more freely available when there are obviously higher returns
We have very poor data on philanthropy
Health & reproductive health is the main recipient of foundation grants
Crowdfunding -> emotional; philanthropic -> strategic; public -> replication
Through growing, philanthropy is overshadowed by public giving
Through their giving, foundations “buy” future impact in their areas of interest/focus
Looking at soundness of the project and solidity of the organization<br>