揭示 AI 的潜力和陷阱一次一个实验.pptx

1、Unraveling Potentials and Pitfalls of AI,One Experiment at a Time,Dayana Hernandez,Jeannie Louie,Lana Milter,Oct 28,2024,Our processAI POC use casesInsightsQ&A,Speakers,Dayana HernandezSalesforce Developer&Solution Architect,Jeannie LouieBusiness Analyst,Lana MilterBusiness Solution Architect,A team

2、 from UCSF SOM Tech(School of Medicine Technology Services)We are not Data Scientists with PhDs in Machine Language LearningWe are not creating new AI models or training new algorithms,Our background,Our expertise and approach,Apply and leverage AI tools that are available right now,mostly with Gene

3、rative AIStart with small but impactful AI features to existing apps and workflows,Our process,Our Process,Analysis,Discovery,AI Dev Cycle,Which tools should we explore,and how will we implement them?,What is the problem to solve?What are the requirements and use cases?,Is AI the right fit for our u

4、se cases?Which AI solution is best suited?,Discovery,Analysis,Traditional Programming,Good at,Considerations,Repetitive tasksWell-defined or rule-based logic,Sample use cases,Data sorting&filteringSimple calculationsForm validations,Pre-determined outcomesStructured data,Analysis,Traditional Program

5、ming,Generative AI,Good at,Considerations,Repetitive tasksWell-defined or rule-based logic,Sample use cases,Data sorting&filteringSimple calculationsForm validations,Pre-determined outcomesStructured data,New content(e.g.,text,images)Natural language processingAmbiguity exists-no single right answer

6、,“good enough”is fine,Virtual agent/Chatbot Conversational searchDigital scribe,Tolerance for errors Slower processing&most costlyUnstructured data,Model bias&fairnessData privacyTransparency,Analysis,Traditional Programming,Generative AI,Predictive AI/Machine Learning,Good at,Considerations,Repetit