Teaching PLN Causal Intelligence

chevron-icon
RFP Proposals
Top
chevron-icon
project-presentation-img
Expert Rating 3.1
Dominik Tilman
Project Owner

Teaching PLN Causal Intelligence

Expert Rating

3.1

Overview

With this proposal, we want to contribute to improving PLN's capabilities through our research of causal reasoning. We want to address both objectives of the RFP by combining our research with the development of a theoretical framework and its validation in two complementary areas: bioinformatics and reputation systems. With this proposal we aim to deliver: 1. Theoretical frameworks for integrating causal reasoning into PLN's inference control 2. Validated methods for distinguishing causation from correlation using PLN rules 3. Proof-of-concept implementations demonstrating our findings in AtomSpace 4. Guidelines for future PLN improvements based on our research results

RFP Guidelines

Causal learning guided PLN inference control

Complete & Awarded
  • Type SingularityNET RFP
  • Total RFP Funding $150,000 USD
  • Proposals 3
  • Awarded Projects n/a
author-img
SingularityNET
Oct. 4, 2024

Implementing causal, versus strictly correlative, inference can lead to a deeper understanding of many complex problems. Probabilistic Logic Networks (PLN), with their robust mathematical foundations for handling uncertain and incomplete knowledge—such as induction, abduction, analogy, and reasoning about time—offer a strong framework for this task. This project has two main goals:

  1. To guide PLN inference control using causal networks.
  2. To explore how PLN rules can distinguish causal from correlational relationships.

Proposals can address either of these goals separately or explore both together and the feedback between them. Submissions should include testable examples to validate the approach and demonstrate how causal reasoning improves PLN’s overall performance.

Proposal Description

Company Name (if applicable)

TrustLevel

Project details

Overview

With our proposal we aim to improve PLN's capabilities in causal reasoning by addressing both RFP objectives: how causal networks can guide PLN inference control, and how PLN rules can distinguish between causal and correlational relationships.

Our approach is to explore these questions through theoretical development followed by practical validation using two domains: bioinformatics pathways (primary focus) and reputation systems (secondary focus). 

We have experience in both domains and think that they offer complementary characteristics for validating our findings - bioinformatics providing well-documented causal chains for verification, and reputation systems being more dynamic in their nature.

Methodology

We plan with four interconnected phases, each building upon the findings and outcomes of the previous phase. With this approach we will: 

  1. Prioritize research and understanding
  2. Validate findings with practical implementation

Phase 1: Research & Analysis

1.1 Causal Network Study

We will research and analyse:

  • existing causal network frameworks and their inference guidance mechanisms, specifically:
    • Bayesian Causal Networks for probabilistic causal modeling
  • how different causal representations affect inference efficiency and inference control refinement
  • different methods for causal strength assessment and uncertainty handling

1.2 Causation vs Correlation Analysis

We will research and analyse:

  1. formal methods for distinguishing causal relationships
  2. temporal and interventional approaches to causality detection

1.3 PLN Integration Study

We will research and analyse:

  • PLN's current inference control mechanisms
  • potential integration points for causal reasoning
  • truth value handling for causal relationships

Outputs for Phase 2:

  • Formal models for causal representation
  • Identified methods for causation/correlation distinction
  • Integration points with PLN's existing mechanisms

Phase 2: Framework Development

2.1 Theoretical Framework

Building on Phase 1 research, we will design and develop:

  • formal models based on identified causal frameworks
  • algorithms incorporating studied temporal and interventional approaches
  • validation methodologies aligned with PLN's mechanisms

2.2 Algorithmic Approaches

We will translate theoretical findings into concrete methods:

  • Design pattern recognition based on studied causal frameworks
  • Prioritization algorithms using analyzed inference mechanisms

2.3 Validation Design

We will prepare the validation framework for Phase 3, including:

  • Test cases based on theoretical framework
  • Metrics aligned with research objectives

Outputs for Phase 3:

  • Theoretical framework with a set of new algorithms ready for validation
  • Validation criteria and metrics

Phase 3: Implementation & Validation

3.1 Research Validation Components

We will implement two key proof-of-concept components using AtomSpace and PLN's native capabilities:

  1. Causal Pattern Detection

    • Temporal and intervention patterns implemented as typed metagraphs
    • Evidence combination using PLN's truth value system
    • Pattern validation through AtomSpace queries and PLN inference
  1. Pattern Matching Framework

    • Direct integration with AtomSpace's pattern matcher
    • Type constraints using AtomSpace's type system
    • Causal validation through PLN rules

3.2 Use Case Testing and Demonstration

  1. PLN Inference Control Enhancement Demonstration:
  • Bioinformatics Use Case:
    • Test with known biological pathways
    • Show how causal networks guide pathway discovery compared to standard PLN
    • Validate causal chain discovery
    • Demonstrate improved reasoning
  • Reputation Systems:
    • Test with simulated user behavior patterns
    • Validate dynamic causal discovery
    • Demonstrate improved decision-making in trust assessment and improved inference paths in manipulation detection
    • Test robustness against noise and manipulation
  1. Causal vs Correlational Distinction:
  • Concrete Examples:
    • In bioinformatics: Distinguish for example between disease mechanisms and symptoms
    • In reputation: Separate direct trust signals from coincidental patterns
  • Validation Metrics:
    • Accuracy against known causal relationships
    • False positive reduction in correlation identification
    • Decision-making improvement

Outputs for Phase 4:

  • Validation results across both domains
  • Implementation insights and challenges

Phase 4: Documentation & Integration Guidelines

4.1 Research Documentation

  • Comprehensive documentation of findings
  • Analysis of results and limitations
  • Recommendations for future research

4.2 Integration Guidelines

  • Detailed guidelines for PLN enhancement
  • Implementation recommendations

Risks and Mitigation

Technical Risks

  1. Complexity of Causal Patterns

    • Risk: Patterns too complex for efficient implementation

    • Mitigation: Incremental development, starting with simpler patterns

  2. Integration Challenges

    • Risk: Difficulty integrating with existing PLN mechanisms

    • Mitigation: Regular consultation with SNET team. And using a modular approach

Research Risks

  1. Validation Complexity

    • Risk: Difficulty establishing ground truth in complex domains

    • Mitigation: Start with well-documented pathways and controlled scenarios

  2. Generalization Challenges

    • Risk: Findings too domain-specific

    • Mitigation: Start with general principles and always have cross-domain validation in mind.

Validation Use Cases

We want to validate our research findings across two complementary domains. The combination of bioinformatics and reputation systems provides ideal validation domains because they:

  • Deal with uncertain and incomplete knowledge at different scales
  • Enable testing with both structured scientific data and chaotic real-world data
  • Offer complementary validation approaches – clear test cases in reputation systems versus complex uncertainties in bioinformatics
  • Allow validation of both simple and complex causal relationships

Primary Domain: Bioinformatics

Bioinformatics provides an excellent validation environment through:

  • Rich complex relationships between gene expression, protein interactions, and disease pathways
  • Well-documented causal pathways for validating our findings
  • Clear distinction needed between mechanistic causes and symptomatic correlations
  • Critical domain where accurate causal reasoning has significant impact

Secondary Domain: Reputation Systems

Reputation systems offer complementary validation through:

  • Clear causal chains (actions → reputation changes → system responses)
  • Controlled test scenario possibilities
  • Real-world adversarial cases through manipulation attempts
  • Both direct causation (user action → reputation) and indirect effects (reputation → behavior)

Integration with Hyperon

We think that this proposal and our research can directly contribute to SingularityNET's AGI development by helping to enhance PLN capabilities within the Hyperon framework. With this proposal our intention is to:

  • leverage and extend Hyperon's metagraph-based knowledge representation for causal reasoning
  • improve PLN's inference capabilities through improved causal understanding
  • support PRIMUS's learning and cognitive control mechanisms

Links and references


https://trustlevel.io

Proposal Video

Not Avaliable Yet

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

  • Total Milestones

    9

  • Total Budget

    $50,000 USD

  • Last Updated

    8 Dec 2024

Milestone 1 - Project kick-off and research preparation

Description

Project Kick-Off meeting of the team and coordination with the SNET team on the project design and make any final milestone changes as necessary.

Deliverables

- Signed contract - Detailed research plan - Initial literature review outline

Budget

$1,500 USD

Success Criterion

Signed Contract

Milestone 2 - Theoretical Research

Description

Core research phase investigating causal networks and PLN inference.

Deliverables

1. Research Reports: - Causal network theory analysis - PLN inference control study - Causation vs correlation differentiation methods 2. Initial theoretical frameworks 3. Proposed algorithmic approaches 4. Analysis of integration points with PRIMUS architecture 5. Review of potential MeTTa implementation approaches

Budget

$9,500 USD

Success Criterion

- Comprehensive theoretical foundation established - Clear approach for causal network integration identified - Methods for causation/correlation distinction defined

Milestone 3 - Framework Development

Description

Development of formal frameworks and algorithms.

Deliverables

1. Formal Frameworks: - Causal inference guidance models - Pattern recognition algorithms - Validation methodologies 2. Algorithm specifications 3. Validation criteria 4. Framework compatibility assessment with Hyperon's knowledge representation 5. Integration considerations for PLN within Hyperon

Budget

$9,500 USD

Success Criterion

- Complete formal framework documentation - Algorithm specifications ready for implementation - Validation approach defined

Milestone 4 - Proof-of-Concept Design

Description

Design of validation implementations.

Deliverables

1. Technical Design: - PLN rule prototypes - Causal pattern specifications - Test case designs 2. Implementation plan 3. Validation metrics 4. Design patterns suitable for MeTTa implementation 5. Consideration of Hyperon's metagraph capabilities

Budget

$9,500 USD

Success Criterion

- Design documents approved - Test cases defined - Implementation approach validated

Milestone 5 - Bioinformatics Validation Implementation

Description

Implementation and testing of biomedical validation cases

Deliverables

1. Proof-of-Concept Implementation: - Basic causal pattern recognition - Pathway analysis examples - Test suite 2. Validation results 3. Analysis documentation

Budget

$6,500 USD

Success Criterion

- Working proof-of-concept for biological pathways - Validation results documented - Key findings analyzed

Milestone 6 - Reputation System Validation

Description

Implementation and testing of reputation system validation cases

Deliverables

1. Proof-of-Concept Implementation: - User behavior analysis - Trust computation examples - Test scenarios 2. Validation results 3. Analysis documentation

Budget

$6,500 USD

Success Criterion

- Working proof-of-concept for reputation systems - Validation results documented - Comparative analysis completed

Milestone 7 - Research Analysis & Findings

Description

Comprehensive analysis of research results.

Deliverables

1. Research Analysis: - Cross-domain findings - Method effectiveness analysis - Limitations and challenges 2. Future research recommendations 3. Integration guidelines 4. Specific recommendations for Hyperon/PRIMUS integration 5. Future development paths within the cognitive architecture

Budget

$3,000 USD

Success Criterion

- Complete analysis of findings - Clear recommendations documented - Integration paths identified

Milestone 8 - Integration Guidelines

Description

Development of PLN integration guidelines.

Deliverables

1. Integration Documentation: - Technical implementation guidelines - Best practices - Example patterns

Budget

$2,500 USD

Success Criterion

- Clear integration guidelines provided - Implementation examples documented

Milestone 9 - Final Report and Documentation

Description

Final research documentation and presentation.

Deliverables

1. Final Documentation: - Complete research findings - Validation results - Implementation guidelines 2. Research presentation 3. Code examples and prototypes

Budget

$1,500 USD

Success Criterion

- Complete research documentation delivered - All findings and guidelines approved - Presentation completed successfully

Join the Discussion (0)

Expert Ratings

Reviews & Ratings

Group Expert Rating (Final)

Overall

3.1

  • Feasibility 4.0
  • Desirabilty 3.2
  • Usefulness 3.0
  • Expert Review 1

    Overall

    1.0

    • Compliance with RFP requirements 1.0
    • Solution details and team expertise 1.0
    • Value for money 1.0
    Proposal structure is there, but crucial details missing

    Proposal with a generic proposal structure included, but no details are given of how to tackle the challenges involved in the RFP, only high-level descriptions are given. Without relevant ideas this project will unlikely lead to useful outcomes.

  • Expert Review 2

    Overall

    2.0

    • Compliance with RFP requirements 3.0
    • Solution details and team expertise 1.0
    • Value for money 1.0
    Extremely general, not actual team record, no trackable experience.

    Not recommend! While I understand that Dominik and his Trustlevel.io is already receiving grants from SNET, I cannot recommend him for the given RFP. After reading his proposal I can see that Dominik at least understands what PLN is (meaning it’s not just some simple probabilistic network as being viewed by others) and what RFP is asking for on a very high level. The proposal remains very general and can be seen as a blueprint for any other project. Dominik breaks down the work in 4 phases: Analysis, Theoretical Development, Implementation & Testing and Documentation; and while I agree on the phasing, any project can be divided into mentioned steps. I am rejecting the proposal because of the following reasons: 1. Not technical and lack of details: the proposal is not technical and very marketing, it does not provide any technical details, nor does it justify the chosen methods, for example it does not mention why Bayesian networks were chosen and not something else. 2. Team expertise is questionable: Dominik himself has education in marketing and business according to his LinkedIn profile and while he wrote the proposal, it is obvious he will not be the person who will be able to complete technical work. The actual developer’s name/profile is not disclosed, nor the proposal has any references to other similar projects or publications. 3. Lack of background work: In the proposal, Phase 1 states the following: “We will research and analyse: existing causal network frameworks and their inference guidance mechanisms, specifically: Bayesian Causal Networks for probabilistic causal modeling” Does the team has experience with casual models? If yes why they need to research it, instead they already should PROPOSE the model and point to their prior works demonstrating successful results. In order to recommend Dominik’s proposal I would like to see an understanding of how PLN control works, what are its pain points, where and why are the good coupling spots, why he thinks that Bayesian is the right network, how to deal with high dataflow/hardware/resources etc. None of details are provided unfortunately…

  • Expert Review 3

    Overall

    5.0

    • Compliance with RFP requirements 5.0
    • Solution details and team expertise 5.0
    • Value for money 5.0
    Good fit

    Reputable team with strong proposal. Appreciate the clarity provided and in favor of seeing this research.

  • Expert Review 4

    Overall

    3.0

    • Compliance with RFP requirements 5.0
    • Solution details and team expertise 3.0
    • Value for money 3.0
    Good albeit somewhat generic proposal.

    The proposal is sensible. I can't find much of a "personal touch" into it, which somewhat worries me. I find it difficult to evaluate how many fruits it may bear. Of course any research will bear some fruits, on the researcher's education if anything else.

  • Expert Review 5

    Overall

    4.0

    • Compliance with RFP requirements 5.0
    • Solution details and team expertise 5.0
    • Value for money 4.0
    A solid proposal that addresses the RFP points and seems to make sense, but gets vague on some key scientific/technical points

    The approach and programme suggested make sense and are well planned out... It's however not too clear what are the "causal patterns" to be used as templates to distinguish causation from correlation. I suppose some information-theoretic patterns can be plugged in here. But I gave it 4 not 5 stars because this lack of clarity makes it a little unclear how well this will actually work..

  • Expert Review 6

    Overall

    4.0

    • Compliance with RFP requirements 5.0
    • Solution details and team expertise 4.0
    • Value for money 4.0

    Detailed causal learning proposal focusing on two primary use cases, bioinformatics and reputation.

feedback_icon