Advancing AI-Based Research: Wizia's Contribution to NASA's AI Assistants Challenge

The Wizia team participated in NASA’s AI Assistants Challenge, a collaborative effort between NASA Tournament Lab and DrivenData, designed to encourage the development of artificial intelligence solutions that support NASA scientists’ demanding research tasks.

The “Research Rovers: AI Research Assistants for NASA” event invited participants worldwide to propose AI-driven methodologies that could perform a multitude of research tasks such as identifying seminal works, summarizing research across diverse publications, and pinpointing research gaps within a specific field.

In response to this call, the Wizia team presented the concept of Automated Information Processing System (AIPS), detailed in our midpoint submission. The document highlights the team’s intentions:

Bridging Semantic Technology and Large Language Models for NASA’s Research Challenges

The Automated Information Processing System (AIPS) is a pioneering framework designed to elevate Large Language Models (LLMs) like ChatGPT into efficient, self-guided research assistants for NASA. By integrating Semantic Technology and a unique “Language as a Tool” Ontology, AIPS addresses the limitations often associated with LLMs — such as lack of focus, inaccuracy, and absence of deep-context understanding.

Semantic Memory Management: Traditional LLMs are restricted by limited context windows. AIPS leverages Semantic Technology like RDF to create a dynamic, interconnected memory layer. This enhances the LLM’s capability to manage complex data relationships, facilitating deep dives into academic literature and synthesizing findings.

Task Management Multi-Agent System: AIPS incorporates a multi-agent system to distribute, monitor, and execute tasks.

Self-documented Messaging Protocol: Every interaction within AIPS is cataloged using a unique self-documenting protocol. This creates a data trail for each query and response, making it easier for researchers to audit the system’s decisions and for the model to learn and adapt over time.

Ontology-Driven Iterative Development: The “Language as a Tool” Ontology serves as the backbone, guiding the LLM in task execution and decision-making. This ontology is not static; it evolves iteratively, incorporating user feedback and system learnings.

Customizable and Scalable Architecture: AIPS is designed for modularity and scalability. Its architecture can be tailored to fit specific research requirements, making it adaptable to a variety of research domains and tasks.

This methodology was recognized by the NASA judges, who provided the following feedback:

The wizia proposal takes a unique approach to a specific sub-element of the AI research assistant problem space, the context memory used in queries of LLMs. This is a great example of focusing on one small part of the overall problem space and doing it well, so the proposal is doing well in relevance, effectiveness and novelty.

Progressing towards the final submission, the Wizia team shifted focus to a broader objective — development of a Multi-Agent Research Network. This initiative is designed to enhance the research process at large rather than targeting individual research tasks. It draws inspiration from the domain of Meta-Research, an evolving field that examines the methodologies, tools, and ecosystem within which research is conducted, with a focus on improving the efficiency and effectiveness of scientific investigation.

Meta-Research typically involves the critical study of research design, data analysis, publication practices, replication of studies, and research ethics. By applying an AI-based meta-research lens, the Wizia team strives to architect a network of AI agents that are uniquely specialized. Each agent is intended to undertake distinct facets of the research process — such as conducting thorough literature reviews, sifting through global datasets, formulating nuanced hypotheses, and evaluating research outcomes.

The team is continuing theoretical work on leveraging semantic technologies for effective use in Large Language Models’ context memory.

The Wizia Team in “AI Research Assistants for NASA”:

  • Vladyslav Kryzhnii, Independent Computer Science Researcher.
  • Andrii Hryhoriev, Research Assistant.