评估在飞机系统中使用人工智能（AI）的风险.pptx

1、EZA240,Harrell J.Van Norman AFLCMC/EZASDSN:986-6715Cell:937-474-8190harrell.van_normanus.af.mil,Assessing Risk of Using Artificial Intelligence(AI)In Aircraft Systems,AI RMF,1,1,Agenda,What is AI?How to Assess and Authorize Aircraft SystemsUnique Challenges:Assessing Non-Deterministic,Inscrutable Ca

2、pabilities in Safety-Critical Avionics is ChallengingAW Certification of Multithreaded Multicore CapabilitiesUnique Approach:Non-AI Components:Traditional RMFAI Capabilities:AI RMF&Cybersecurity FrameworkIdentify and Quantify AI RiskBenefits,Concerns,Future of AI,2,Artificial Intelligence Classifica

3、tion SAE G-34 Aerospace Taxonomy,A secondary classification is also proposed to take into account the different fields of application:computer vision,natural language processing,anomaly detection,and optimization,3,Labeled Data,Unlabeled Data,Partially Labeled Data,Trial-and-error process to obtain

4、a reward,Inscrutable Results,no one can predict when they might fail,4,Hidden meanings:In neural networks,data is passed from layer to layer,undergoing simple transformations at each step.Between the input and output layers are hidden layers,groups of nodes and connections that often bear no human-i

5、nterpretable patterns or obvious connections to either input or output.“Deep”networks are those with many hidden layers,Traceability?,Coverage?Explainability?As exciting as their performance gains have been,theres a troubling fact about modern neural networks:Nobody knows quite how they work.And tha

6、t means,Verifiability?,AI Unique Challenges,AI Unique Challenges,Money,5,Rapidly Adopted,Complex,Difficult to Assess,Vital Reliance,Dynamic,Better decisions Productivity gains Improved speed Personalized experiences Improved services Improved monitoring Better qualityReduce human error,Neural Networ