MPAI has developed an approach to standards for AI-based data coding that relies on the identification of Use Cases implemented as AI Workflows (AIW). These perform certain functions on received data of specified data types and produce output data of specified data types. AIWs are composed of AI Modules (AIM) arranged with a specified topology and exchange data of identified data types. AIMs can be Composite when they are composed of other AIMs and are called Basic when they are not, or when the internal structure has not been exposed.
Technical Specification: AI Framework V2.1 specifies an environment that enables initialisation, execution, dynamic configuration, and control of AIWs.
By applying this process, MPAI has developed 20 AI Workflows, 120 AIMs, and 180 data types.
The level of sophistication of AIMs that are part of the AIWs varies considerably in terms of number and variety of data types involved and of number and interconnected AIMs. The function of some may require an in-depth data analysis, while the function of others may be performed with simple data processing operations.
MPAI has developed and is currently refining the notion of Perceptible and Agentive AI (PAAI), a machine or set of machines performing functions that MPAI has standardised in its Technical Specifications.
MPAI has developed the Technical Report: AIW and AIM Implementation Guidelines (MPAI-WMG) V1.0 for the following purposes to review its base of AIWs and AIMs in the light of the Perceptible and Agentive AI:
