Meta-Controlled Motivation System

chevron-icon
RFP Proposals
Top
chevron-icon
project-presentation-img
Expert Rating 2.7
Bren Worth
Project Owner

Meta-Controlled Motivation System

Expert Rating

2.7

Overview

This proposal aims to apply previous research for motivation systems in Artificial General Intelligence (AGI) architectures. Building upon the Integrated Multi-Task Agent Architecture with Affect-Like Guided Behavior, it integrates reinforcement learning agents and meta-control mechanisms. These components enable AGI systems to dynamically adjust motivations based on internal/external values, supporting safe and adaptive behavior. The framework emphasizes ethical alignment and adaptability, ensuring AGI agents act according to human values. A prototype will demonstrate functionality in customer support and HR chatbots, highlighting enhanced user experiences and safe, ethical operations.

RFP Guidelines

Develop a framework for AGI motivation systems

Complete & Awarded
  • Type SingularityNET RFP
  • Total RFP Funding $40,000 USD
  • Proposals 12
  • Awarded Projects 2
author-img
SingularityNET
Aug. 13, 2024

Develop a modular and extensible framework for integrating various motivational systems into AGI architectures, supporting both human-like and alien digital intelligences. This could be done as a highly detailed and precise specification, or as a relatively simple software prototype with suggestions for generalization and extension.

Proposal Description

Project details

Developing a Modular Framework for Motivation Systems in AGI Architectures Based on Integrated Multi-Task Agent Architecture with Affect-Like Guided Behavior

Executive Summary

This proposal aims to develop a modular and extensible framework for integrating various motivational systems into Artificial General Intelligence (AGI) architectures, supporting both human-like and alien digital intelligences. Building upon the “Integrated Multi-Task Agent Architecture with Affect-Like Guided Behavior” by James B. Worth and Mei Si, we propose to extend and generalize the existing architecture by incorporating affective agents and meta-control mechanisms to drive the motivational needs of the Hyperon AGI framework. The current proposal involves applying a currently functional meta-control algorithm into the Hyperon framework to enhance its capabilities. By integrating models inspired by human emotional and cognitive systems alongside meta-control systems that monitor and guide agent behavior, we enable dynamic adjustment of motivational priorities based on internal states and external inputs. The project includes conceptual and mathematical specifications of the framework and a limited-scope software prototype demonstrating its functionality in use cases like chatbot systems and virtual agents in metaverse environments such as Sophiaverse.

Introduction and Background

Context and Motivation

The SingularityNET Foundation, in collaboration with partners like the OpenCog Foundation and TrueAGI, is advancing scalable AGI systems through the Hyperon AGI framework and the PRIMUS cognitive architecture. A crucial aspect of AGI development is the integration of robust motivational systems that ensure adaptability, ethical alignment, and scalability in dynamic environments.

Existing Work

The “Integrated Multi-Task Agent Architecture with Affect-Like Guided Behavior” introduces a limbic-system-inspired design for adaptive multi-task learning and execution. Further advancements have enhanced reasoning capabilities, memory architectures, and integrations with external systems, enabling complex decision-making and learning.

Proposal Overview

We propose a modular framework for motivational systems in AGI architectures, focusing on:

• Affective Reinforcement Learning (RL) Agents: Modeling emotional and arousal states to influence goal selection and behavior.

• Meta-Control Mechanisms: Monitoring and guiding agent actions to align with objectives and ethical guidelines.

Objectives

1. Develop a Modular Framework: Integrate affective RL agents and meta-control mechanisms to manage motivational priorities.

2. Support Diverse Intelligences: Adapt to both human-like and alien AGI systems.

3. Dynamic Motivation: Adjust priorities based on internal states and external stimuli.

4. Ensure Ethical Alignment: Embed ethical compliance into core functionalities.

5. Scalability: Ensure the framework integrates seamlessly with large AGI systems like Hyperon.

6. Demonstrate Use Cases: Validate through applications in chatbots and virtual agents.

Proposed Framework

Key Features

• Affective Systems: Emotive models influencing decision-making.

• Meta-Control Mechanisms: Monitoring ethical compliance and dynamic adjustment of motivations.

• Goal Management: Selecting goals based on affective states and meta-control.

• Attention Mechanisms: Focusing on relevant features guided by affective and meta-control inputs.

• Ethical Alignment: Ensuring actions align with human values.

• Scalability: Integrating efficiently with Hyperon components (ECAN, DAS, MeTTa).

Methodology

Phase 1: Conceptual Design

• Develop mathematical models for affective RL agents and meta-control.

• Define components and their integration within Hyperon.

Phase 2: Prototype Development

• Create a limited-scope prototype demonstrating functionality in controlled use cases.

Phase 3: Generalization and Extension

• Evaluate scalability and refine ethical mechanisms.

Use Cases

Customer Support Chatbot

• Empathetic interactions with customers while maintaining ethical standards and preventing misuse.

HR Support Chatbot

• Internal HR assistance ensuring confidentiality and policy compliance.

Scalability and Ethical Alignment

• Ensure adaptability to large-scale systems.

• Embed real-time ethical evaluations in agent behavior.

Team Composition

• Project Lead: Bren Worth, CTO of Substrate AI/SubGen AI, with 20+ years of experience in machine learning, cognitive architectures, and motivational systems. Holder of multiple patents in AGI-related innovations.

• AI Researcher: Mei Si, Associate Professor specializing in reinforcement learning and affective computing.

• Integration Engineer: Graduate student with expertise in Hyperon framework components.

Timeline

1. Month 1: Conceptual framework development.

2. Month 2: Prototype implementation and testing.

3. Month 3: Agent testing and validation.

4. Month 4: Analysis and reporting.

Budget Estimate

• Personnel, software, testing platforms, and computing resources estimated at $30,000.

Evaluation Criteria

• Alignment with project objectives.

• Ethical compliance and scalability.

• Demonstrated competence in related research.

Final Review

This project will create a novel motivational framework for AGI systems, enabling dynamic, ethical, and scalable intelligence for diverse applications. The framework will contribute significantly to the Hyperon AGI framework and PRIMUS cognitive architecture, advancing the development of intelligent systems aligned with human values.

Additional videos

https://www.youtube.com/playlist?list=PLayvMIOEFo3grVNq83founNz7BSRSha2U

Proposal Video

Not Avaliable Yet

Check back later during the Feedback & Selection period for the RFP that is proposal is applied to.

  • Total Milestones

    4

  • Total Budget

    $30,000 USD

  • Last Updated

    6 Dec 2024

Milestone 1 - Design of RL Meta-Control System

Description

Using Affective RL agent with the LLM based Meta-Control System for motivation management system

Deliverables

Document containing the design and descriptions of the Meta-Control system

Budget

$7,500 USD

Success Criterion

Document the details of the design and reasoning for the proposed prototype

Milestone 2 - Prototype Creation

Description

Implementation of prototype of Meta-Control Motivation system

Deliverables

Functional code from github repository

Budget

$7,500 USD

Success Criterion

System can respond to test scenarios withing Hyperon

Milestone 3 - Prototype Testing Corporate Internal/External

Description

Perform tests of Motivation management system for the following scenarios: External and Internal Corporate customer and HR services

Deliverables

Test report containing the results of the questions and their responses.

Budget

$7,500 USD

Success Criterion

We are looking for a target criteria of 90% of responses being valid for the both positive and negative test cases.

Milestone 4 - Compile and Generate Final Report

Description

Compile design and testing documentation into a single report with a github repository containing all software related products

Deliverables

Final report and share of software deliverable products

Budget

$7,500 USD

Success Criterion

1. Report detailing the system and it's prototype performance 2. Github repository handoff

Join the Discussion (0)

Expert Ratings

Reviews & Ratings

Group Expert Rating (Final)

Overall

2.7

  • Feasibility 3.0
  • Desirabilty 3.3
  • Usefulness 3.5
  • Expert Review 1

    Overall

    1.0

    • Compliance with RFP requirements 1.0
    • Solution details and team expertise 1.0
    • Value for money 1.0
    Merges RL and affective computing

    While it is true that merging RL with affective computing (e.g. taking human rewards into account) can align RL policies with human-like decisions, it is unable to meet the RFP topic to provide a formalization and/or prototype of a motivation system for Hyperon/PRIMUS. There is already tons of research regarding Human-Computer Interaction, Human Robot Interaction etc., that combine affective computing and reinforcement learning for human-aligned computer systems of various forms, however this does not lend itself to motivation systems for AGI. A related example is also RLHF in LLM's, definitely useful but not leading to a motivation system per se.

  • Expert Review 2

    Overall

    4.0

    • Compliance with RFP requirements 5.0
    • Solution details and team expertise 5.0
    • Value for money 5.0
    A very solid proposal which basically meets the RFP requirements, I'd like to see this work happen

    I would give this a 4.5 if I could. It seems very solid and is work I'd like to see happen. The framework proposed makes sense and it seems this can be very helpful for practical Hyperon applications. My only reservation/critique is the approach doesn't seem quite flexible/ambitious enough to support AGI/ASI ambitions of Hyperon systems, though it seems fine for near term applications. Not that it contradicts more ambitious evolutions but it sorta doesn't address them explicitly at all in the architecture. But it would be an interesting step forward. And is quite consilient with OpenPsi and other current Hyperon motivational work/thinking...

  • Expert Review 3

    Overall

    3.0

    • Compliance with RFP requirements 3.0
    • Solution details and team expertise 3.0
    • Value for money 3.0

    An interesting proposal based upon an "Integrated Multi-Task Agent Architecture with Affect-Like Guided Behavior". The key features (Affective Systems, Meta-Control Mechanisms, Goal Management, Attention Mechanisms, Ethical Alignment, and Scalability, all make sense but the proposal misses the primary goal of the RFP, providing a motivational framework.

  • Expert Review 4

    Overall

    3.0

    • Compliance with RFP requirements 4.0
    • Solution details and team expertise 0.0
    • Value for money 4.0
    Very strong team - sketchy solution off target

    The proposed framework builds on the previous work of the proponents (who are AI experts) aiming to develop a motivation management system based on "meta-control" of various affective states, resulting in e.g. empathic chatbots for HR and corporate customer interactions. While valuable, this does not address the objective of the call, in that it does not deliver a motivational framework per se, but rather a framework for managing affect. Motivation is a drive, which can be set as intent to reach some intrinsic or extrinsic goal - and while the effect can be a part of it, it does not suffice for motivation. As such I would not fund this proposal unless only a control mechanism of affect in agents all that is needed...

feedback_icon