Overview
AIDDL is both a language and a framework for integrative AI. That means its main purpose is to facilitate the combination of different AI techniques into a larger (integrated) system.
As a language, AIDDL allows to write models for different AI problems in a common language.
As a framework, AIDDL allows to manipulate terms written in the language, check types, use a rich library of default implementations of AI algorithms.
Together language and framework allow to describe and execute integrated AI systems in the same language as their inputs.
Typical Workflow
Write one or more AIDDL files containing the required models.
Write a program that:
loads your AIDDL files and accesses required terms
loads existing AI functions from the AIDDL framework
add missing functions (e.g., application specific, glue, wrappers around existing AI libraries)
combine functions into an integrated AI system
Language
The language can be divided into three types of terms.
Basic terms: symbols, variables, numbers, strings
Composite terms: lists, sets, tuples, key-value pairs
Reference terms: references to entries of functions
Types
Types specify subsets of the language.
They are used to:
Describe models used by AI algorithms (e.g., a planning domain, machine learning data or models, a logical knowledge base).
Describe applications specific input/output models. This allows to write application specific models without choosing an AI approach.
Consider the following example: the language contains any set mixing symbols, numbers, variables, etc.
We can define a type to express all sets that contain positive integers.
The core library (briefly covered in the next section) contains functionality to load types and perform type checks on terms. In the previous example, the type sets containing only positive integers would test negative for the first set and positive for the second.
Core Library
The core library makes AIDDL accessibly in a specific programming language.
Load AIDDL files (also called modules) into containers
Default operations on terms (e.g., numerical, set or list operations)
Evaluate terms
Type check terms (given a type definition)
Run unit tests written in AIDDL format
Common Library
The common library is a collection of types and algorithm implementations of common AI algorithms. Its main purpose is to provide default components for rapidly crafting integrated AI systems.
External Libraries
External libraries provide wrappers around other libraries to allow them to use AIDDL models as input and output. These libraries are usually dedicated to specific AI problems and their implementations are expected to be more mature, stable, and closer to the state-of-the-art than the ones found in the common library.
External libraries are kept separate from the common library to avoid dependency overload.
Examples
A collection of projects that show how AIDDL can be used to create integrated AI systems.