What is an AI Marketplace? • Developing AI models often requires the sharing of data from the user side, and such data may be proprietary. Therefore, an AI marketplace may have a mechanism which ensures that developers use that data only for training purposes. • An AI marketplace needs a mechanism which can determine the quality of a final delivered AI model. Conventionally, accuracy has been the primary metric. However, alternative metrics that capture reliability, robustness, and fairness are also now considered important. • Like in any conventional software system, AI systems also require maintenance over time. In the standard software industry, the company which originally developed the software is usually responsible for providing support. However, in an AI marketplace, specific AI developers may not be available in the future. So, an AI marketplace needs standard guidelines which AI developers should follow while developing models for marketplace customers. Thus maintenance by other AI developers is much easier. • Unlike in an app store, an AI marketplace could allow customers to request new products on the fly. An AI marketplace can then quickly match AI customers with AI developers with relevant expertise. • Some AI models are proprietary. So, unlike in an app store, they can not be shared online with a wider audience, as this would further facilitate adversarial attacks and risk leaking intellectual property.
Marketplace stakeholdersdata owners (DO), cloud owners (CO), and model owners (MO).
Liftoff (Q4 2021)
Acceleration (Q1 2023)
Skyward (Q4 2023- Q1 2024)