
Soumil Rathi
Project OwnerProject Owner
MeTTa was designed as a language for AGI, yet there is a noticeable lack of agent-based demos showcasing its full potential. This proposal seeks to address that gap through an innovative project that highlights MeTTa’s unique capabilities. This project proposes a MeTTa-based simulation of a virtual dinner party involving neoteric agents with dynamic goal-setting, procedural learning, and introspective reasoning. Each agent will have their own goals to hit during the dinner party. With their goals and knowledge stored in their Atomspace, each agent will be able to recursively subdivide their current goal, thus taking the best action at the moment to eventually reach their goal.
Create educational and/or useful demos using SingularityNET's own MeTTa programming language. This RFP aims at bringing more community adoption of MeTTa and engagement within our ecosystem, and to demonstrate and expand the utility of MeTTa. Researchers must maintain demos for a minimum of one year.
This milestone will handle the setup for the environment and all of its agents. This includes the larger environment with the dinner party environment and each of the different agents within SophiaVerse. Moreover the milestone also includes the structure setup for an agent as MeTTa programs. Finally, it also includes setting up the agents - the states, goals, memory and actions that any agent can take.
The overall deliverable for this milestone is a setup for the multi-agent environment. For each agent, the states, actions, goals will be defined within MeTTa, and the memory component will be initialized. The format of the deliverable will be the codebase.
$3,000 USD
A successful implementation would allow an agent to be setup within the dinner party environment within SophiaVerse. This agent would be able to have its goals and potential actions setup.
This milestone refers to the integration of the MeTTa program with the SophiaVerse environment. As a part of this milestone the agent will be able to execute any selected actions which will run within the SophiaVerse environment as well.
The deliverable for this milestone would be the codebase for the demo with the integration within MeTTa to execute actions in the SophiaVerse. This effectively means that actions will be mapped from Atomspace states to grounded functions.
$2,000 USD
A successful implementation of this milestone will have an agent setup within MeTTa that can take actions within SophiaVerse when explicitly programmed. For example, the agent could be told to walk forward, speak, pick up things, eat, and would be able to execute such actions within the multi agent environment.
The most important thing for an agent is to create sub-goals and use the state information to map to the best action at the time. This milestone will address that creating rules for each of those mappings such that the agent can autonomously run. The matching would be based on probabilistic logic with MeTTa’s pattern-matching.
Once again, the primary deliverable will be the codebase. Within this codebase, all MeTTa rules for subgoal creation and action mapping will have been created and implemented. Thus, within this codebase, agents would be able to operate autonomously within the constraints of its defined goals.
$10,000 USD
A successful implementation of this milestone would have an agent able to decide which action to take based on its goals. This agent, when given a goal (eg. pick up a particular) object, would be able to break down the goal into subgoals continuously (eg. face object, reach object, pick up object).
This milestone refers to each agent being able to learn and update from its experiences in the environment. This includes gaining knowledge from the environment as it runs such as information about other agents and the environment. It also includes getting better at certain actions and goals by updating internal probabilities of success for undertaken actions.
Once again, the deliverable for this milestone will be the codebase at the current stage. At this level, the MeTTa rules and functions for identifying / gaining information from the environment will have been setup. This includes having functions to update probabilities of certain actions and goals based on learned experience.
$9,000 USD
A successful implementation of this milestone will see the agent learning from any mistakes it makes, or any actions that don't work in the environment (limited by the defined rules). As an example, if other agents are unfriendly to it, it would be able to perceive this and use that information to update it's internal goals and actions, such as removing a goal to become their field.
This milestone covers the final work for the proposal - writing all required documentation and deploying the project. The documentation includes the full structure and functionality of the environment as well as each individual reason. It also includes a technical report about the demo. Experiments will be conducted with the interactive demo with different starting goals and will be included in the technical report. Finally I will create a video explaining the demo structure and functionality to make it easy to follow. The code will be deployed in a Hyperon instance hosted by SingularityNET.
This milestone includes the final deliverables for the proposal. It will contain the full codebase under an appropriate open source license. This codebase will be deployed on a public Hyperon instance. Finally, all relevant documentation will be written and provided, including video and technical report
$1,000 USD
A successful implementation of this milestone will mean that the Hyperon instance running the codebase is running without any errors, the codebase itself is documented and maintained, and that the documentation is extensive and well-written enough to be understood by other developers, and that the code is extensible in the future.
Reviews & Ratings
While reviewers rated this submission highly, ultimately the panel of experts selected another proposal for strategic reasons.
Please create account or login to write a review and rate.
© 2024 Deep Funding
Join the Discussion (0)
Please create account or login to post comments.