<p>about your learning experience to help us improve.</p>

You've completed

cleanFiles/bookmap_Putting_AI_Ethics_into_Practice_at_SAP_en-US_learning-sap-com_lsc-prod.zip (MMCP-DROPZONE) - 2024-11-07T08:50:40.000Z

Managing Standard AI Use Cases - Ethics in Everyday AI

Lesson Content

Putting AI Ethics into Practice at SAP

Learn about managing standard AI use cases and ethics in everyday AI in this informative video. Explore the governance of AI ethics in SAP and the implementation of AI-driven visual inspection in SAP Digital Manufacturing.<br><br># This description was auto-generated #

SAP_MOOC_aie2_Lesson_7_USECASE_Video.mp4

Welcome to the lesson on Managing Standard AI Use Cases, Ethics in Everyday AI. My name is Sherene Tan and I'm the Product Owner for the Visual Inspection Team in SAP Lab Singapore. Here is a quick recap on a governance of AI ethics in SAP. The diagram here summarizes the AI ethics assessment process. Any work that involves AI in SAP will need to go through this process before it can be developed, deployed, sold or used. In this lesson, I will walk you through the following use case, AI-driven visual inspection in SAP Digital Manufacturing. This has been identified as a standard AI use case. This solution aims to transform human visual inspection experience and overcome human cognitive limits. Production operators can now provide inputs to train an AI model. The trained model can then be integrated into business processes to visually inspect parts in production. The use of AI now assists workers in identifying defects earlier, faster, and more accurately, allowing businesses to improve productivity and quality significantly throughout the value chain. As visual inspection involves AI, it must go through the AI ethics review process and the first step is to complete an impact assessment that classifies the use case as a redline case or not. The set of redline criteria has three main categories assumed, personal freedom, society and environment. Since the visual inspection use case did not violate any of these red lines, it was not classified as a red line case. Do note that if your use case violates any of these red lines, it goes against the AI ethics policy and further development is prohibited. The next step is to then determine whether the use case is considered high-risk or not. There are five high-risk conditions that will result in a high-risk classification. If there is at least one yes to any of these conditions, a follow-up is necessary with the AI ethics office and a potential approval by the steering committee. As visual inspection is designed for quality inspections in the manufacturing process, no sensitive personal data is collected nor used. The AI models are mainly trained to identify any defects on the inspected parts and the process also ensures that a human can always override the AI decisions if necessary. Hence, this use case was classified as low risk. Do note that self-assessment is still required even if your use case has been identified as a standard AI use case. Now let's take a closer look into the visual inspection solution. I would like to highlight two key features. The first key feature is allowing human oversight in the inspection process. A human-in-control approach has been implemented where all recommendations from the AI model must be verified by the human. It is required for production operators to explicitly confirm via a button click that that the non-conformances found by the AI model are correct. Any decisions by the AI can be easily overridden by the human. If the AI misclassifies the inspected part as conformant or non-conformant, the production operator can indicate otherwise and easily remove any incorrect non-conformances found by the AI. Probability scores are also clearly shown for each identified non-conformance code to ensure transparency and to support human decision making. The second key feature of the solution is explainability. Non-conformances that are identified by the AI are clearly labelled on the inspection image, which allows production operators to easily identify the recommendations by the AI, making the verification process much easier for them. So to conclude, these are the three key takeaways from this lesson. The first, AI-driven visual inspection use case was classified as a standard use case in SAP, No red lines were violated and no high risk criteria were met. Second, all AI use cases, including standard use cases, must adhere to SAP's global AI ethics policy. Third, AI-driven visual inspection has a human-in-control governance mechanism implemented to ensure appropriate human oversight in the inspection process. With that, we've come to the end of our lesson. Thank you and goodbye.