人工智能应用的机遇与挑战.pdf

1、OPPORTUNITIES AND CHALLENGES FOR AI ADOPTIONProfessor Susan AtheyStanford UniversityMANY TYPES OF AI APPLICATIONS2Developers(B2B and/or B2C apps)Developer Services&ToolsAdd-InsConsumer UserGeneral PurposeSpecial PurposeCustomer Assistants,Platform-hosted AppsService ProviderClientInteractionGuidance

2、 and ContentInternal Business UserInternal Apps,Business Process AutomationB2B SAASBusiness Auto-mationData AnalysisExample application categoriesTarget user of AIAssistants SearchHealthTutorShoppingVacation planningData feedsSubscript.PaymentsBackendFrontendCoach-client dashbdCoaching guidanceForec

3、astHR processAcctingCustomer MgmtOptimize advertis-ing,productnUser/business adoption requires complementary investments,human capital,techEach type has distinct data&development requirements,general vs.specific data,user interface/guardrailsMODERN AI LEADS TO RESTRUCTURING2010s AI/ML:Shift towards

4、more general-purpose inputs Significant silos Barriers to entry/adoption3Storage&ComputingData MarketsSoftware&Developer ToolsFoundation model paradigm:General-purpose data and learning in foundation models Apps do not recreate basic understanding Low/no codeFoundation ModelFM Software&Developer Too

5、lsProprietary DataSoftware EngineeringAI,ML,AnalyticsData,Storage,ComputeEngineering,AI,MLSOME(OF MANY)POTENTIAL SCENARIOS FOR AI MARKETOpen/Interoperable Scenario Closed models better than open models,charges Open models available,“good enough”when combined with fine-tuning+add-ons Cloud providers

6、host many models+developer tools,support interoperability Low-code developer tools,plug-ins,analysis Content creators(?)may be data market Widespread adoption,competitive prices for applicationsImpact and Importance of Open Models4Closed/Concentrated Scenario Closed models much better than open,high