Generative Artificial Intelligence (gen.AI) in Healthcare

In this presentation we introduce AI and generative AI. Discuss application in healthcare. Talk about challenges, limitations and opportunities.

1. Generative Artificial Intelligence ( genAI ) in Healthcare Sonika Tyagi, PhD Associate Professor (Digital Health and Bioinformatics) Data Science & AI division | School of Computing Technologies RMIT University Australia

2. What is AI?

3. What is AI? 3

4. — What is AI? The study and construction of agents that do the right thing. 4

5. Beneficial machines What is the right thing? The value alignment problem. “Machines are intelligent to the extent that their actions can be expected to achieve their objectives” ​ -Stuart Russell 5

6. — Inception of AI 8

7. — Inception of AI 9

8. The idea of a ”perceptron”: Neuron goes “artificial” 10

9. The idea of a ”perceptron”: Neuron goes “artificial” 11

10. The idea of a ”perceptron”: Neuron goes “artificial” 12 Ref: Machine Learning and Other Artificial Intelligence Applications, An Issue of Neuroimaging Clinics of North America, E-Book   McCulloch & Pitt's MCP neuron to Rossenblatt's Perceptron (1958)

11. Artificial Neural Networks (ANN): From a single perceptron to multi-layer perceptron approach a.k.a deep learning

12. Artificial Neural Networks (ANN): From a single perceptron to multi-layer perceptron approach a.k.a deep learning Predict clinical outcome: Length of Stay Mortality ICU admission etc. Image analysis Detect tumours Analyse x-ray etc Linking data Genetics Lifestyle Healthcare other

13. — Training a machine learning model The training data is used to train the model. 15 The test data is used to evaluate the model. "Black box" ? X_train Y_ train X_test Y_ test x1 x2 x3 X Y Model Output Label Y'  Data

14. Generative AI GenAI creates entirely new content based on its understanding of data The bird flies in the ….. The cat sat on a …... I love Apple OpenAI's chatGPT , Google’s Bard, Microsoft Bing, Chatsonic , Github Copilot, and ChatSonic ….

15. Health data: Electronic Medical Records (EMR) EMR Immunisations Vitals Pathology Medications Admin Prescriptions Clinical Notes

16. GeneAI applications in healthcare Medical Foundation model Health data: Electronic Medical Records (EMR) A EMR Specialized tasks Healthcare data >clinical documentation (e.g. discharge summaries; clinic, operation, & procedure notes) >summarizing research papers;  >working as a chatbot .. .. .. >Risk stratification >finding similar patients Human in the loop

17. ...there are two sides to a coin

18. Hurdles and challenges in using healthcare data for research: 20 Ref: Book chapter "Technical Issues in Implementing AI in Healthcare" -  Sonika Tyagi;  ISBN 9781032200880

19. AI and Bias Historical Bias Representation Bias Measurement Bias Modelling Bias Interpretation Bias

20. AI and Bias Historical Bias Representation Bias Measurement Bias Modelling Bias Interpretation Bias

21. GenAI specific issues

22. Evaluating geneAI models Wornow et al. (2023) elaborate upon six criteria: predictive performance, data labeling, model deployment, emergent clinical applications, multimodality, and novel human-AI interfaces.  Leadership Regulation  LLMs differ from already regulated AI-based technologies Hallucination vs design shortcomings? Incentives

23. Standadise AI? Best practice Adoption 20-30% Media hype Negative press <5% Adoption

24. Conclusion Generative AI has the potential to transform healthcare and improve patient outcomes Integration of gen.AI necessitates meticulous change management and risk strategies Strategic adoption: implementation science, incremental deployment, hype vs limitation Focus on inclusivity and sustainability and governance that is human centric.