AI-DSL Phase 2 — Final report

Matt Iklé
SingularityNET
Published in
4 min readJan 13, 2023

The AI-DSL Project for an AI interconnectivity layer on the SingularityNET decentralized AI Marketplace.

Also available in Russian, Chinese, French, Portuguese, Spanish

We are excited to announce the completion of Phase 2 of the AI-DSL development project. SingularityNET’s AI-DSL (AI Domain Specific Language) project will allow simple AIs to be linked together like legos and assembled into complex workflows by anyone, even non-technical users. This project is creating a revolutionary, intelligent interconnectivity layer to augment the utility of every service on the SingularityNET Platform/Marketplace.

With enthusiastic support from the Cardano community, Cardano Catalyst Fund7 funding was awarded to Phase 2 of the AI-DSL development project, to support ongoing research & development.

The AI-DSL project is a cornerstone of the expansion and evolution of the SingularityNET Platform/Marketplace into a revolutionary tool for Decentralized AI and will:

  1. Contribute to the broad usability and interoperability of services on the Platform
  2. Allow networkable decentralized AI workflows to be easily created
  3. Provide structural protocols and linkages to allow SingularityNET to evolve as a self-organizing network
  4. Create a framework for the potential emergence of AGI from this network of AI

The short-term benefits of this project are service discovery, interoperability, and linkage; the long-term potential is a self-organizing and self-assembling set of complex AI tools. These opportunities make this project a remarkable foundation for building the future of decentralized AI.

AI-DSL Overview

Currently, most AI systems require input and output to have a highly specific file format, data format, data organization, etc. This specificity comes at the expense of flexibility; data from one AI agent is not readable to other agents. The AI-DSL project will make AIs interoperable for the first time by providing a standardized communication protocol where AI services will be able to:

  • Locate compatible AI services on the Platform (or anywhere AI-DSL is used),
  • Verify exactly what the service requires,
  • How it processes, and what kind of data it outputs.

The initiating service will then be able to call those compatible services as needed to complete a workflow. This capability will exponentially increase the AI processes that can be achieved by combinations of AI services on the Platform/Marketplace. Thus making all of the SingularityNET services collectively more powerful and more generally capable. This innovative protocol will also make it simple for users to access all of the AI services they require via a single entry point. This saves time by eliminating the need to go through multiple providers or search for specific services across multiple platforms.

AI-DSL Phase 1 was a proof of concept using simple ‘toy problems’ (basic math arithmetic operations) to demonstrate the path forward for linking logical sequences, and pre-verifying inputs and outputs. In addition, ontology work was begun to define the requirements of the system. This foundational progress created the basis for the fundamental research completed during Phase 2.

Phase 2 Achievements

Phase 2 picked up from this proof of concept, taking the research further. The AI-DSL team first built a set of basic AI logic tools; these are mathematical foundational tools that most types of AI systems call and use:

  • Error minimization/gradient descent,
  • Linear regression,
  • and Logistic regression.

These services were then configured to call each other via the AI-DSL project protocol, focusing on connecting the services using types. Our team focused heavily on program synthesis and verification in this Phase - checking that the service to be called will actually output the result needed. The practical results from this work were very exciting, and the learnings are informing future directions of development in intriguing ways.

The fundamental research in this phase focused on six broad focus areas:

  1. Registry Tasks
  2. Ontology Tasks
  3. Exploratory Smart Contract Tasks
  4. Test Cases
  5. Production Tasks
  6. Co-Evolutionary Tasks

See the closing report and video for further details.

Next, AI-DSL Phase 3

During this upcoming phase, the team will implement the AI-DSL on live platform services, set up a demo workflow, and make it available as an SDK for power users. These developments will lay the practical basis for the AI-DSL project's capabilities for program synthesis and verification, and registry search functionality.

The specific aims of Phase 3 will be to:

  • Optimize the protocol by connecting and testing the protocol on real-world AI applications running on the SingularityNET Marketplace.
  • Focus on the underlying technologies required for AI-DSL project implementations,
  • Identify the project’s critical development areas, as well as the most promising future directions.

The AI-DSL project development team is on track to roll out an initial version of the product in 2023.

One of the keys to future AI-DSL project iterations identified by the team is the potential use of MeTTa, the language of OpenCog Hyperon. The AI-DSL project currently utilizes the Idris programming language, but MeTTa is expected to eventually replace Idris to provide greater flexibility, and efficiency and allow larger compositions of AI services to take place. The team will begin considering this conversion during this phase of development.

SingularityNET’s Future with AI-DSL Project Integration

Developing tailored AI algorithms that can solve real-world problems has been tedious, expensive, and time-consuming. The AI-DSL project is paving the way for users to access all the AI services they need in a single place. These self-assembling workflows will replace the current labor-intensive process for creating specialized, one-off AI processes. This protocol will create a universal mode of AI intercommunication and collaboration, making the benefits of complex AI processes accessible to all.

Stay Up to Date With the Latest News

Join the conversation, and follow SingularityNET for the latest news and updates:

--

--