Individuals and institutions need better tools for analyzing, testing, and planning around AI-ML or Artificial Intelligence and Machine Learning.
At this github repo is a paper on a proposed framework to better define behaviors, abilities, limits, and goals: https://github.com/lineality/object_relationship_spaces_ai_ml
This paper is a work in progress.
From the abstract:
Definable Units of “Intelligence” for Evaluating AI Performance
Object Relationship Spaces for AI-ML: A Framework for Clearly Defined, STEM-Compatible, Project-Level, Functional Units of “Intelligence” For AI Design, Analysis, Performance, Architecture, and Operating Systems
There is a need for the use of well defined performance frameworks to describe the goals and skills/abilities of systems including AI.
The overall agenda here is to move toward clearer communication and better definitions, including the pragmatic utilization of universal intersecting/interlocking areas.
This proposed object-relationship-space framework can be used for guiding project-specific system design, goal-setting, discussion, testing, analysis, reporting, regulation, documentation, etc.
For more detail on what is meant by ‘design’: to manage and enable smaller or larger scale AI projects coordinating required abilities across internal and external components, including “symbolic” logistics and “sub-symbolic” training (including for AI-self-management), and whole operating-systems for AI.
AI must be able to handle “objects” in the following interlocking contexts:
2. (internal/external) project-object-database (in a project-framework)
3. project-participants (in a project-framework & participation-space)
such that, ‘objects’ are defined as existing outside of the AI for managing the project, and that so long as the AI effectively deals with these project-objects, it does not matter how the AI handles the objects ‘internally’.
A repeating theme, context, and agenda in this paper is to pragmatically leverage the interconnected functionality of clear definitions, STEM, projects, participation, positive values, and productivity.
Part one concerns a brief overview of the framework.
Part two concerns using the framework,
e.g. so you can construct your own well defined goals and tests for abilities of AI systems.
Part three concerns a discussion of the discussion of AI,
e.g. so you can critique statements in what you read about AI.
Part four concerns goals and agendas, background concepts and principles, and future design factors.