Junzi Moral Framework for AGI Alignment

chevron-icon
RFP Proposals
Top
chevron-icon
project-presentation-img
Expert Rating 2.7
Augustin Chan
Project Owner

Junzi Moral Framework for AGI Alignment

Expert Rating

2.7

Overview

Junzi Moral Framework for AGI Alignment Using the five constants (五常) as base vectors for moral alignment: 仁 (Benevolence) - compassion vector 義 (Righteousness) - ethical decision vector 禮 (Propriety) - social harmony vector 智 (Wisdom) - knowledge integration vector 信 (Trustworthiness) - reliability vector

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

Company Name (if applicable)

junzi.institute

Project details

Key Innovation Points:
1. Using the five constants (五常) as base vectors for moral alignment:
   * 仁 (Benevolence) - compassion vector
   * 義 (Righteousness) - ethical decision vector
   * 禮 (Propriety) - social harmony vector
   * 智 (Wisdom) - knowledge integration vector
   * 信 (Trustworthiness) - reliability vector

2. Technical Implementation (Phase 1):
* Python/TypeScript stack using proven components
* BGE embeddings for moral vector encoding
* pgvector for similarity search
* Basic vector operations for moral reasoning
* Simple chatbot demo interface

3. Unique Advantages:
* Framework for encoding classical wisdom in vector space
* Measurable moral reasoning through vector operations
* Clear path to future integration with AGI systems

4. Use Case (Phase 1):
* Chatbot demonstrating basic Junzi principles through:
  - Vector-based moral evaluation
  - Simple explanations with classical references
  - Basic decision consistency

5. Future Extensions (Phase 2):
* ECAN integration for attention management
* MeTTa implementation
* Additional use cases (virtual agents, robots)
* Advanced self-cultivation mechanisms

Milestones (4 months):

Milestone 1 ($10k):
- Vector encoding of five constants
- Basic pgvector setup
- 20 example scenarios encoded
- Simple similarity matching

Milestone 2 ($10k):
- Basic moral reasoning engine
- Decision-making logic
- Example explanations with quotes
- Core API functionality

Milestone 3 ($10k):
- Simple chat interface
- Basic visualization of moral space
- Documentation
- Phase 2 proposal

I want to be transparent about my implementation strategy and explain why I believe it will benefit the Hyperon ecosystem.

For Phase 1 (4 months), I propose using established technologies (Python, BGE embeddings, pgvector) to prove the core concept of encoding classical Chinese virtues as vector representations for moral reasoning. This approach will:

1. Allow rapid validation of the framework's fundamental principles
2. Produce clear empirical results about vector-based moral reasoning
3. Create a solid foundation for future Hyperon/MeTTa integration

While I understand the value of implementing directly in MeTTa, I believe creating this proof-of-concept first will:
- Generate valuable insights for future MeTTa implementation
- Provide concrete examples for mapping moral vectors to Hyperon's architecture
- Enable faster iteration and testing of core concepts

Phase 2 would focus on integrating with Hyperon/MeTTa and ECAN, building on validated results from Phase 1. This stepped approach ensures we're building on solid ground while moving toward full Hyperon integration.

I'm excited about contributing to SingularityNET's vision and would appreciate your thoughts on this implementation strategy.

Open Source Licensing

AGPL - Affero GPL

Can use the license that singularity.net prefers

Links and references

I will be using what I learned from building 8bitoracle.ai for this project.

Proposal Video

Not Avaliable Yet

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

  • Total Milestones

    3

  • Total Budget

    $30,000 USD

  • Last Updated

    3 Dec 2024

Milestone 1 - Virtue Vector Foundation

Description

Vector encoding of five constants Basic pgvector setup 20 example scenarios encoded Simple similarity matching

Deliverables

Python library with basic vector operations Small but well-documented training dataset Simple PostgreSQL/pgvector implementation README with examples

Budget

$10,000 USD

Success Criterion

Generate embeddings for 20 clear example scenarios Demonstrate basic virtue classification works Show similarity matching between related scenarios Document core vector operations with examples Basic tests covering main functionality

Milestone 2 - Moral Reasoning Engine

Description

Build reasoning system using vector operations Implement moral decision-making logic Create explanation generation system

Deliverables

Python or Javascript backend for moral reasoning Vector-based decision-making system Basic test suite

Budget

$10,000 USD

Success Criterion

Successfully demonstrate basic moral reasoning chain: Input → Vector Analysis → Decision → Explanation Show consistent results for similar input scenarios Include at least 3 classical references/quotes in explanations Provide working API examples for each core function Document limitation cases and error handling

Milestone 3 - Integration & Demo

Description

Build proof-of-concept chat interface showing moral reasoning Create basic visualization of virtue vectors Document framework for future integration

Deliverables

Working chatbot demo that can: Process user input through virtue vectors Generate responses based on moral reasoning Provide explanations for decisions Simple visualization tool showing how decisions map to virtues Documentation and future roadmap

Budget

$10,000 USD

Success Criterion

Successfully demonstrate 5 distinct moral reasoning scenarios Provide clear explanations linking decisions to virtues Create visualization showing relationship between virtues Document 10 example scenarios with step-by-step analysis Outline proposal for ECAN integration

Join the Discussion (0)

Expert Ratings

Reviews & Ratings

Group Expert Rating (Final)

Overall

2.7

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

    Overall

    1.0

    • Compliance with RFP requirements 1.0
    • Solution details and team expertise 1.0
    • Value for money 1.0
    Moral framework / AGI alignment

    Vector-based approach to encode vectors for moral alignment. While it could be used for this purpose, this approach does not explain how motivation systems could work in the first place, and hence it will not be relevant to the AGI systems we will build. Retrospectively fitting this framework on to a functioning motivation sytstem will unlikely lead to moral or ethical alignment, and most importantly it does not comply to the RFP requirements to formalize and/or build a prototype of a motivation system.

  • Expert Review 2

    Overall

    3.0

    • Compliance with RFP requirements 4.0
    • Solution details and team expertise 3.0
    • Value for money 4.0
    It's a clearly specified and interesting proposal which happens not to align with my own intuition/orientation... it's imposing a specific value system at the code level

    This is an interesting sort of experiment, encoding a highly specific quasi-Confucian value system at the code level ... it happens not to gel with my own approach would would be more to "wire in" abstract moral principles/capabilities and motivational principles/capabilities and then let more specific content be learned/adapted based on experience and interaction. But there is so much we don't know, I can't say for sure this approach is not good/interesting too... there can be use for embedding concise hand-coded ontologies at a deep level like this ofc..

  • Expert Review 3

    Overall

    3.0

    • Compliance with RFP requirements 3.0
    • Solution details and team expertise 4.0
    • Value for money 4.0

    The proposal does not target the motivation problem, focusing instead solely on moral alignment based on the five principles of Benevolence, Righteousness, Propriety, Wisdom, and Trustworthiness. Only limited/vague/generic discussion of Hyperon integration.

  • Expert Review 4

    Overall

    5.0

    • Compliance with RFP requirements 4.0
    • Solution details and team expertise 0.0
    • Value for money 4.0
    Very high value - if successful

    Although it only partially covers the Call - in that it offers a moral framework for motivation - this proposal can solve the crux of "the AI Alignment" problem if it succeeds in delivering a way to build wisdom in the fabric of AGI. Given where our world is today, with moral confusion stemming from values devoid of any morality, if successful this proposal can offer a simple, yet compelling solution to the design of an AGI that could guide humanity toward "the light". The simplicity of the solution is backed by a sound and reasonable approach, gradually verifying the success against use cases (20 chatbots in the initial phase). Given that motivation per se has no morality (can be geared towards doing bad ...) - this morally-driven motivational framework - although static and rigid - when combined e.g. with Anna Mikeda's proposal, can deliver a universal solution (IMHO).

feedback_icon