从伦理到工程：设计值得信赖的人工智能系统.pdf

《从伦理到工程：设计值得信赖的人工智能系统.pdf》由会员分享，可在线阅读，更多相关《从伦理到工程：设计值得信赖的人工智能系统.pdf（21页珍藏版）》请在三个皮匠报告上搜索。

1、From Ethics To EngineeringRobbie JerromPrincipal Technologist AI:AI Business Unit1From Ethics To Engineering2From Ethics To EngineeringWe have powerful models,but we lack tools and architectures to make them trustworthy,auditable,and aligned with societal expectations.Its not enough to state princip

2、les;they must inform your architecture,platform,and decision-making.We need more than checklists;we need AI system design that embodies trust as first-class.Ethics In AI3AI Ethics its complicatedAI ethics addresses the moral principles and societal impacts of artificial intelligence systemsFairness

3、and BiasTreat all people equitably without discrimination.Transparency and ExplainabilityShow your working,and explain how decisions are made.Privacy and Data RightsProtect personal data and respect user consent.Safety and ControlPrevent harm and maintain human oversight.AccountabilityTake responsib

4、ility when things go wrong.Economic and Social ImpactConsider effects on jobs,inequality,and access.How can AI platform engineering help?Models4Technical governance through transparencyBlack-box AI no longer suffices under growing regulatory oversight and public scrutiny.Open-source models offer a m

5、ore auditable and transparent approach,which is essential for meaningful governance,compliance,and trust.5Lets start with the model,and then discuss the systems and platforms surrounding it.Generative AI Models6Open-Source AI ModelsUnlike open-source software,open-source AI is a little more complex.

6、License TypeExample ModelsCommercial UseKey RestrictionsCan Fine-tune?Can Distill?Training Data Available?Apache 2.0IBM GraniteMistralFalcon Unlimited None Use anywhere Any user scale Yes YesGranite:YesOthers:NoMeta Llama 3.xLlama 3,3.1,3.2,3.3 Yes(below 700M users)700M monthly user limit Cant train