/ai-dsl

Artificial Intelligence Domain Specific Language (AI-DSL) to enable autonomous interoperability between AI services.

Primary LanguageIdrisMIT LicenseMIT

AI-DSL

Artificial Intelligence Domain Specific Language (AI-DSL) for autonomous interoperability between AI services.

Overview

The general idea of the AI-DSL is to provide

  1. A simple and powerful language to express AI service assemblages, as well as to formalize mathematical properties, both crisp and statistical, to be met by these AI service assemblages. These properties may pertain to algorithmic behaviors, interactions with other servies as well as resource usage, computational, financial or otherwise.

  2. A tool set for not only verifying the validity of such assemblages but also automatically create such assemblages. Basically, if a user can formalize precisely enough the desired function, then the AI-DSL tool set should be able to automatically fetch and combine the right AI services to deliver that function.

Status

Up to now the work has been mostly exploratory resulting into prototypes and experimental code covering various aspects of the AI-DSL as opposed to a final product. This is justified by the fact that it is an ambitious project and requires a fair bit of research and development.

Progress has been taking place into phases. Details can be found in technical reports written at the end of each phase. An overview of is given below.

Phase 1

  • Overview:
    1. Formalize trivial properties about service resources, computational, finantial and performance using Idris.
    2. Formalize trivial properties of services doing simple arithmetic operations using Idris to experiment with dependent type checking based service assemblage validation.
    3. Explore existing ontologies such as SUMO to provide a rich vocabulary to the AI-DSL.
  • Technical Report of May 2021

Phase 2

  • Overview:
    1. Implement the machine algorithms gradient descent, linear and logistic regression in Idris.
    2. Formalize and prove a descending property for each algorithm in Idris.
    3. Explore program synthesis in various ways, including developing our own language framework as well as using existing tools provided by Idris. Program synthesis is important because it is the backbone of automated service assemblage.
  • Technical Report of Octover 2022

Folder Structure

  • doc: contain technical reports and other documentation
  • experimental: contain a number of experiments, ranging from representing and proving properties using dependent types, performing program synthesis to type checking protobuf specifications, and more.
  • ontology: experiment representing SingularityNET platform knowledge in SUO-KIF format.
  • snet-marketplace-space: scripts to build an atomspace of the SingularityNET Marketplace, as well as file dumps in MeTTa and JSON formats.

Further Reading