chevron-icon
Active RFP

Causal learning guided PLN inference control

Top
chevron-icon
SingularityNET
RFP Owner

Causal learning guided PLN inference control

Study how PLN inference control can be guided by causal networks and/or how PLN rules can distinguish causal from correlational relationships

  • Type SingularityNET RFP
  • Total RFP Funding $150,000 USD
  • Proposals 3
  • Awarded Projects n/a

Overview

  • Est. Complexity

    💪 75/ 100

  • Est. Execution Time

    ⏱️ 6 Months

  • Proposal Winners

    🏆 Multiple

  • Max Funding / Proposal

    $75,000USD

RFP Details

Short summary

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.

Main purpose

  • To explore how causal learning and causal networks can help guide PLN inference control. 

  • To advance research in causal reasoning using PLN, which is foundational for AGI, and thus improve how AI systems distinguish between causality and correlation, leading to more accurate predictions and better decision-making processes.

Long description

Context and background: 

SingularityNET Foundation, in collaboration with other partners such as the OpenCog Foundation and TrueAGI, is working toward a scalable implementation of the Hyperon AGI framework running on decentralized infrastructure, and toward implementation of the PRIMUS cognitive architecture within this framework.

Hyperon and PRIMUS are complex systems involving multiple components, which need to demonstrate appropriate functionalities both individually and in combination. This RFP aims to address a portion of this overall need, via funding the initial iteration of one significant component of PRIMUS within Hyperon.

This project has two goals: 

  • To study how Probabilistic Logic Networks (PLN) inference control can be guided by causal learning, including causal networks; 
  • To study how PLN rules can be used to learn causal versus correlational relationships. 

Proposals can address either of these goals individually or explore both and the feedback between them. The ultimate aim is to improve PLN’s ability to manage uncertain and incomplete knowledge by incorporating causal reasoning, leading to more robust and accurate inference control.

  1. Guiding PLN Inference with Causal Networks:
    This thread seeks to determine how causal networks can be used to steer PLN's inference control in a specific domain or context. This may involve running large-scale inferences across domains (such as bioinformatics, genomics, or even simulated environments like Minecraft) and evaluating whether causal network-enhanced inference outperforms PLN operating independently.
  2. Distinguishing Causal from Correlational Relationships:
    This line of inquiry focuses on how PLN rules can be used to learn and differentiate causal relationships from mere correlations. This is critical in fields where distinguishing between cause and effect is essential, such as scientific research, AI planning, or healthcare. Proposals should examine how PLN can recognize causality versus correlation and validate this through examples where such distinctions improve decision-making accuracy.

Researchers may consider a variety of causal network learning systems for integration with PLN. Examples include Bayesian Causal Networks, Pearl’s Do-Calculus, Structural Causal Models, Potential Outcome Framework, and Probabilistic Event Calculus. These systems offer powerful tools for reasoning about causality and could enhance PLN's capabilities in distinguishing causal relationships from correlative ones. Additionally, Information Theoretic approaches can provide insights into the relationships between variables, helping PLN refine its inference control or causal learning. Deep Neural Network approaches also appear promising.

Collaboration:

This RFP may be followed by subsequent RFPs for applications that leverage Hyperon/PRIMUS to carry out various applications, and that aim to guide Hyperon/PRIMUS systems in cognitive development toward beneficial AGI

RFP Expected Outcomes:

  • For PLN Inference Control: Demonstrate how integrating causal networks enhances PLN’s ability to guide inference control within a chosen domain. Provide examples where this improves reasoning accuracy or decision-making.
  • For Causal vs. Correlational Learning: Develop methods that use PLN rules to distinguish between causal and correlational relationships. Offer concrete examples and validation metrics showing how this distinction leads to better outcomes in a given context.
  • Test Suite and Evaluation: Whether focusing on one of the above or both, provide a comprehensive set of test cases and evaluation metrics that measure performance improvements.
  • Domain-specific Application: Apply the chosen methodology (inference control, causal learning, or both) to a practical domain (e.g., bioinformatics, AI planning, or simulated environments like Minecraft), demonstrating clear improvements in causal reasoning or inference accuracy.

 

 

Functional Requirements

Must Have:

  • A well-defined methodology for either:
    • Using causal networks to guide PLN inference control, including a clear process for how causal information will be integrated into the reasoning framework; or
    • Enabling PLN rules to learn and distinguish between causal and correlative relationships, detailing how the rules will process and evaluate these distinctions.
  • Definition of a specific problem domain or use case (e.g., bioinformatics, AI planning, simulations) where the chosen methodology will be applied.
  • Empirical validation of the proposed approach through testing on a moderately-sized dataset or within a well-defined simulated environment. The dataset or environment must provide enough data to evaluate whether the system can accurately guide inferences or differentiate causality from correlation.
  • A test suite or benchmarking process to evaluate the effectiveness of the proposed causal reasoning or inference control, demonstrating clear improvements over standard PLN methods.
  • Proposals must outline how the data will be used to run multiple inferences, providing enough examples to ensure meaningful causal learning or inference control validation.

Should Have:

  • Demonstration of how the system's outputs (e.g., causal relationships, inference improvements) will be presented to validate results (visualization tools, video, report, or other appropriate methods).
  • Integration of external causal models or tools (e.g., Bayesian networks, causal discovery methods) to enhance the robustness of the approach.

Could Have:

  • Ability to apply different types of causal models, such as counterfactual reasoning or various forms of intervention-based causal analysis.
  • Exploration of cross-domain applicability, testing the system’s ability to adapt from one domain (e.g., genomics) to another (e.g., AI planning).
  • Visualization tools that map inference paths or causal structures, providing intuitive insights into how the system is making decisions.

Won’t Have Yet (probably):

  • Full-scale generalization or production-ready applications. The focus should remain on proof-of-concept research within defined domains.

Non-functional Requirements

  1. Architecture: Both options require integration with OpenCog Hyperon and PLN within Atomspace, regardless of whether you're guiding inference control or distinguishing causality from correlation.
  2. Programming Language: The need for Hyperon-compatible languages (e.g., Python, Rust, C++, MeTTa) applies equally to both options, as both involve working with PLN and Atomspace.
  3. Data Quality: Whether you are guiding inference control or distinguishing causality, quality datasets are essential for ensuring meaningful and valid research outcomes.
  4. Performance: Both options involve running inferences or testing PLN’s capabilities on datasets, so the system needs to be performant enough to handle this, though real-time performance isn't required for either.
  5. Modularity and Extensibility: For both options, modularity is important for future experiments. Option 1 might require different types of causal networks, and Option 2 might need flexibility in testing different types of causal vs. correlational distinctions.
  6. Documentation: Provide thorough, accessible documentation throughout the project. This should include:
    • Detailed methodologies: Clearly outline implementation processes, experimental design, and any encountered challenges.
    • Reproducibility: Ensure all methods and results are documented to allow for replication, future application, and further development by other researchers.

Main evaluation criteria

Alignment with requirements and objective

  • Does the proposal meet the requirements and advances the objectives of the RFP

Pre-existing R&D

  • Has the team previously done similar or related research or development work in other platforms / languages / contexts?

Team competence

  • Does the team have relevant skills?

Cost

  • Does the proposal offer good value for money?

Timeline

  • Does the proposal include a set of clearly defined milestones?

Other resources

References 


Hyperon and related AI-platforms are quickly evolving! This is a bit of a moving target, but the internal SingularityNET team will be available for help and expert advice, where needed. Also included:

  • SingularityNET technology links
  • Educational materials and resources for learning MeTTa
  • SingularityNET holds MeTTa study group calls every other week. Proposers are welcome to attend for support from our researchers and community.
  • Recurring Hyperon study group calls for community are currently being planned. These will cover MOSES, ECAN, PLN, and other key components of the OpenCog and PRIMUS Hyperon cognitive architectures.
  • Access to the SingularityNET World Mattermost server, with a dedicated channel for discussion and support among the RFP-winning teams and SingularityNET resources.

RFP Status

Completed & Awarded

The community and public are invited to view the full proposals and give feedback. During this time the RFP committee will doing their formal selection process to award winning proposals.

View Awarded Projects
3 proposals
rfp=proposal-img
EXPERT REVIEW 3.1

Teaching PLN Causal Intelligence

  • Type SingularityNET RFP
  • Funding Request n/a
  • RFP Guidelines Causal learning guided PLN inference control
author-img
Dominik Tilman
Dec. 2, 2024
rfp=proposal-img
EXPERT REVIEW 3.1

Bayesian Causal Networks for Probabilistic Logic

  • Type SingularityNET RFP
  • Funding Request n/a
  • RFP Guidelines Causal learning guided PLN inference control
author-img
evanluoicecream
Nov. 2, 2024
rfp=proposal-img
EXPERT REVIEW 1.3

Causal Learning-Driven

  • Type SingularityNET RFP
  • Funding Request n/a
  • RFP Guidelines Causal learning guided PLN inference control
author-img
Kirmair Lima
Dec. 4, 2024
0 projects

No Projects Avaliable

Check back later by refreshing the page.

Join the Discussion (2)

Sort by

2 Comments
  • 1
    commentator-avatar
    khellar
    Dec 2, 2024 | 1:53 PM

    Hi Dominik this work request is not meant to produce a service for our marketplace. Apologies for the confusion. Please disregard!

  • 0
    commentator-avatar
    Dominik Tilman
    Dec 2, 2024 | 1:35 PM

    @khellar @admin. The submission template for this RFP asks for a description of a service to be integrated into the SNET Marketplace. However, it is not clear from the RFP description what this should be (there is only the integration with Hyperon mentioned). Is this a mistake (because I don't see it in other RFPs either), or do I misunderstand it?

feedback_icon