Trade-Offs in Provisioning Artificial Intelligence as a Service : Critical Information Infrastructures, Winter Term 21/22

Background: The market for Artificial Intelligence as a Service is increasing. With rising adoption of this cloud paradigm, the service design becomes more and more important. Currently, setscrews and corresponding trade-offs in Artificial Intelligence service design are not well understood.
Objective: This work aims to deepen the understanding of Artificial Intelligence service design and the trade-offs that follow specific design decisions. The research questions are: What are the main setscrews for the design of AI as a Service products? What trade-offs do AI as a Service providers face, when choosing different designs of their AI as a Service products?
Methods: To answer the research questions we conducted ten semi-structured interviews with domain experts in Artificial Intelligence as a Service. Based on the interviews we identified setscrews and trade-offs in Artificial Intelligence as a Service design.
Results: There are five groups of setscrews for for Artificial Intelligence as a Service products: reliability and security, performance, relationship to competing products, ease of use, and functionalities. Because performance and ease of use are often affected by changes to setscrews of other groups, they are part of many of the trade-offs that we have identified.
... mehr
Conclusion: Artificial Intelligence as a Service has similar service design trade-offs to other cloud computing paradigms such as Software as a Service. However, new challenges arise such as demand peaks for computing power caused by deep learning services.