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 RFP has two primary goals, both centered on the exploration of concept creation techniques within the OpenCog Hyperon system using the MeTTa programming language. 

1. The first goal is to experiment with methods for generating new concepts based on existing concepts and data, evaluating the resulting algorithms for effectiveness and creativity. While Formal Concept Analysis (FCA) was implemented in OpenCog Classic to support these tasks, this project encourages extending FCA into more advanced frameworks such as fuzzy FCA and paraconsistent FCA. These variations allow reasoning under uncertainty and inconsistency, making them particularly valuable for more flexible, creative systems. There may also be other frameworks worth exploring beyond FCA that can be integrated into this experimentation.
2. The second goal is to implement one or more of these concept creation algorithms in Hyperon. MeTTa, the core language of the Hyperon architecture, provides an ideal environment for these experiments due to its flexibility and symbolic representation capabilities. Successful implementations will contribute to the broader AGI development effort, focusing on dynamic and adaptive concept formation.

In this context, creativity is a critical property to consider when evaluating new concept formation algorithms. Hyperon’s cognitive processes, which operate concurrently and cooperatively, offer multiple pathways for concept creation. These include evolutionary learning, concept blending, and formal concept analysis (potentially involving fuzzy or paraconsistent methods), alongside various other heuristics. The blending of these techniques allows Hyperon to generate novel concepts by abstracting from learned experiences and evidence.

The formation of new concepts in Hyperon also ties into core memory structures. For example, long-term memory and working memory play a role in clustering and concept formation, while map formation and pattern mining help abstract high-level structures from raw data. Techniques such as concept blending, Occam’s-Razor-driven concept predicatization, and heuristic-based map formation support this dynamic process of concept generation.

A key feature of Hyperon’s architecture is that the world model it constructs consists of the abstractions it has learned from specific cases via probabilistic logic networks (PLN) and concept creation algorithms. The system continuously refines its understanding through these mechanisms. Reinforcement learning (RL) methods can also contribute to this learning process, particularly in identifying explicit causal relationships within the data. These relationships may often be concrete but are also subject to abstract reasoning, further enriching the concept formation capabilities.

Moreover, one promising area of research suggested for this RFP is using LLMs or multi-agent systems of LLMs to evaluate the creativity and novelty of the concepts generated. LLMs, with their vast knowledge of existing concepts, could serve as a metric for assessing whether newly generated concepts are innovative and useful, while avoiding purely random or irrelevant outputs. By framing concept creativity within these systems, new pathways for automated and meaningful concept evaluation can be explored.

Ultimately, proposals should aim to integrate these blending and formal analysis techniques into Hyperon’s broader framework, driving the development of new abstractions and enhancing Hyperon’s ability to create novel, meaningful concepts. This will support ongoing AGI research by contributing to a more flexible and adaptive system capable of sophisticated cognitive functions.

https://wikidocs.net/197350

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:

• Implementation of Concept Creation Algorithms: Successful integration of one or more formal concept analysis (FCA), fuzzy FCA, paraconsistent FCA, or concept blending algorithms within the Hyperon framework using MeTTa.
• Creativity Evaluation Framework: Development of methods to evaluate the creativity and novelty of new concepts, potentially leveraging LLMs or multi-agent systems for assessment.
• Demonstrated Practical Use: Concrete demonstrations of concept creation algorithms generating valuable, innovative outcomes, such as novel abstractions or knowledge structures integrated into Hyperon’s Atomspace.
• Enhanced AGI Cognitive Synergy: Contributions toward cognitive synergy in Hyperon by improving concept formation processes, with practical applications in long-term and working memory systems.