Invitation to Experiment with Abstractive Symbolic AI

What is Epistemology?

Epistemology is the theory of knowledge, especially with regard to its methods, validity, and scope.

Epistemology is the investigation of what distinguishes justified belief from opinion.

What is Abstractive Symbolic AI?

Abstractive Symbolic Artificial Intelligence (ASAI) uses the principles of epistemology to help users determine the subject/predicate relationship or supporting arguments to premises. In other words, it is a way to help people better process information and extract knowledge.

Mitigating the impact of Generative AI in research

Prior to putting pen to paper, an author will typically organize thoughts and ideas into connected stories. These connected stories (like supporting arguments) come together to form a narrative that the author commits to a document. The text of the document contains semantics that are used to help communicate ideas from the author to the readers. When a person reads a document, the brain analyzes the semantics and organizes the result into a structured hierarchy of stories and propositions that represent the document's key themes and concepts - similar to a "Table of Contents" in a book. This enables the reader to identify what is relevant.

The Impact to Knowledge Management

Researchers collect, analyze, and interpret information in their day-to-day. Each document consists of themes and concepts. The level of effort to process documents will vary based on the complexity and length. Processing information and extracting knowledge at scale can be time-consuming, difficult, or both. ELAINE is an innovative tool built on the principles of Abstractive Symbolic AI. The purpose of ELAINE is to help people better consume and manage knowledge. Using the principles of epistemology, ELAINE maps a document's themes and concepts into an easy-to-navigate hierarchy.

How it works (the science)

Everyone is using the word "narrative", but what is it and how is it created? The simplest definition of a "narrative" is a set of connected stories or events. When an author writes a narrative, it starts with a set of themes and/or concepts that are grouped into propositions and come together to tell a story. ELAINE uses semantic analysis to break down a narrative into a semantic hierarchy that models the author's mindset and intent.

A semantic hierarchy is an organization of terms (words) that represent themes and concepts that have been abstracted from a story. The abstracted terms are arranged in the order according to the semantic relevance of a theme and/or concept. The abstracted terms of a theme and/or concept form a "high level abstraction" (HLA) . The number of terms in each HLA represents the semantic depth (like nuance/context) of a theme and/or concept. The greater the depth, the richer the supporting argument of the author.

How does it work (in practice)

ELAINE takes a complex document and creates a semantic hierarchy - like a Table of Contents - that maps out key concepts and/or ideas that are ranked by the author's intent. For example, a high concentration of HLA tags signals a high density of concepts within the document. As a result, the mapping reveals the author's most valuable (concept dense) intent - the "good stuff" of a document.

Applied to research papers, it is entirely possible that documents with an extensive semantic hierarchy are indicative of thoughtful supporting arguments written by a human.

An Example: Compare and contrast two research papers - one prepared by a human and another with GPT generated content

ELAINE analyzed both documents. Below are the two knowledge maps from the analysis:

The example shows the difference of the human mindset organizing thoughts behind the composition of a document versus one that is generated with the help of GPT. Links to the original articles and the detail ELAINE analysis:

=> Research paper prepared by human - ChatGPT Recognize Its Own Writing in Scientific Abstracts?"

=> ELAINE analysis: Recognize Its Own Writing in Scientific Abstracts?

=> Research paper prepared with the help of ChatGPT - the Consequences of AI-Driven Academic Writing on Scholarly Practices

=> ELAINE analysis: Exploring the Consequences of AI-Driven Academic Writing on Scholarly Practices

It is important to note that this is a one-off example to show the difference between human composition and GPT assisted composition. From the knowledge maps in this example, it shows the inter-connection of terms between propositions in a human composed research paper is more intensive than the one composed by GPT. Similarly, the semantic hierarchies of the research paper prepared by human contain more semantic depth than the one prepared by ChapGPT. This one-off example is far from adequate. For this reason, we like to invite other researchers to help us to validate or to refute the idea behind this experiment while acknowledging LLM is capable of reproducing text that is highly similar to the trained materials. Our methodology can show the semantic organization of thoughts behind a research paper, however, it may never be an absolute certainty in determining if a research paper is composed with the help the GPT.

The Invitation

Our team is extending an open invitation to a limited number of professionals who work in research and are curious about how to de-risk the impact of Generative AI.

If you are interested in participating in the pilot of this scenario with ELAINE on research papers at large, please click Register to connect with our team.

NOTE: Participation in this pilot is free of charge and is not a grant of license. The pilot will run for a pre-determined period of time and may follow-up with a limited survey to collect feedback.