Back [[3 Oct 2024 - HIMSS Asia Day 2]] Speaker Dr Gao Yulia, NUHS, Singapore Summary: - AI has its Pro and Cons, Good and Bad. - Need Ethics considerations - Human still figuring this out. ———————————————————————————————————————— The Dark Side of AI, Fear or Fearless. - Healthcare’s 4th Industrial Revolution - Moving from AI Implementation Pathway. - Level 1 to level 4 Dark Side? 1. Dissecting racial bias in an algorithm used to manage the health of population - trained on bias data set. 1. Self-fulfilling, cycle Addressing the Dark Side 1. Data Quality and Bias. - R.I.R.Out 1. Representative dataset 2. Confounders 3. Bias 4. Data purity Algorithm train for skin disease, it doesn’t work on dark colour skin. Overfitting and External Validation - Is HOPES model overfitting? Can it work if it’s outside of the data set it’s used to train on? Common pitfalls and recommendations for using ML to detect COVID 19, chest XRay. Information Accuracy - Hallucination, - Confabulation - Ability to handle rare cases - How to handle ambiguity Security and Patient Safety - Are they really who they are? — AI, VR, AR, deep fake - AI Influencers - Generally marketing. - Cyberattacks - Phishing. - AI Enabled future crime - feeding false information, hacking into system - Fraudulent Medical Research - someone use AI to generate a false paper. - Put through exert reviews, even experts doesn’t know (like wine experts) - Statistician analysis - - If didn’t know it’s AI generated, we would thought it’s done by human. - Can AI check AI? - Risks to patients? - Guardrails put in place to GPTs. - Prompt engineering can or may go around it - Psychological and Social impact - Does constant health monitoring - where is line obsession/abusive use of health tracking. - False positive = anxiety - Doctor - patient relationship - Dr AI, - Misinformation - Medical Education - AI in exams - Reduce Human knowledge? - Existential Crisis in healthcare works? - Replace human worker? - Consultation roles. Low level physical tasks - What if merge robot and AR and AI? Ethical Concerns - AI Ethics, - Huge impact on medical practice. Do no harm - Hippocrates. - How to train AI to follow that? Access to patient information 1. Who should have access to the data 2. Who owns the data 3. Implicit consent? 4. Can it be use against me? Project Nightingale. - Google, 50 million medical records uploaded to google clouds - Access to that information not only from hospital, but google employees. - Other cloud services we are using now. AI sees everything? - Highly complex digital twins. - Advance AI - Are we replica of human brain, Science of Morality - Moral enhancement, freedom and the God machine - Paper - AI monitors us all the time, intervene without human knowing.. in the name of morality - AI driven real time human analytics — covert interventions Skynet? Is it all doom and gloom? Jan 1, 1983 - launch of the internet Back then we have similar concerns for internet. Governance: - HIMMS solid maturity models to help us make sense of what we are doing - WHO guidance on large language models Remembering Humanity - Understand our role as human, humanity. - “Being human in a digital world” Deloitte. “Meaning and Utility” - Directed and Empowering world. - Responsibility and Regulation.. Learn from Cartoons - Training “how to be my friend” - when we ask the right questions, we get the right answers. Dark Side of the Moon - Far side of the moon. - Sense of wonder and mastery. - Less crater, smoother, thicker crust. The more we know, the more we appreciate. - putting transmitter on the far side of the moon. “Fear, Fearless or Fear Less”