Converting Knowledge into Insights – Generative AI

Introduction

Generative AI, LLMs – Hearing a lot about It these days?

Let us try to uncover what it is.

Simply put – It’s an AI algorithm with human-like thinking abilities, like how humans can interpret text and numbers. This interpretation includes analysis, summarization of the data, and generative capabilities, where the model can next line of action. Generative models have gained significant intelligence, which can predict the next action by looking at the pattern of the complex multi-dimensional dataset.

Believe it or not, Generative AI will change the way manufacturing industries function today.

People act, processes react, and people again act. That’s what is called experience-based decision-making. Whereas, during the act-react-act process, there are several dimensions of data being generated of which some are relational, some are time-series, and some are of document type format. All these data are stored in different systems of which hardly any data is being accessed for deriving insights other than relational and time-series.

Though this might look lucrative, Gen AI doesn’t have any visibility of the domain or specifics of an organization. This means it will not have any understanding of specific domain keywords or any underlying construct of the data. In that case, how do we infuse it?

Every gen AI needs context to allow the extraction of meaningful insights. In manufacturing language context is what becomes the wisdom for the generative models. These contexts are presented to the language models using the knowledge graphs which store the relevant information about the organization’s processes and operations.

Let us go a little deeper into it.

The underlying foundation of LLMs is the constructs of language – semantics and syntactics. Semantics reveals the sentiments and the meaning of a text/number in the given context, and syntactic references the grammatical syntax as per the English standards. This is what we also term natural language processing (NLP).

Let's start building on top of the example use cases – compressors discussed in my previous article for Knowledge graphs. In this continued article, I will take an example of equipment maintenance, and troubleshooting workflows.

Current practice:

Imagine, there had been an abnormality observed in one of the compressors. The maintenance team would need an investigation which would demand for analysis of multiple data from multiple systems. This process is time-consuming and data-intensive, and beyond all-this is all experience-driven.

Enhanced workflow: (Gen AI-based):

With Generative AI, powered by knowledge graphs, the system will allow users to ask questions in human language. LLM (Gen AI) will process the user query to chunk/tokenize into smaller contexts. Contexts are the essential elements that LLM uses for comparison with the underlying data. Contexts are used to perform complex data analysis and similarity search from all the data sources. Domain context is handled by knowledge graphs which are used to interlink all types of data. This domain know-how is leveraged by LLM to extract the required insights.

Following shall be the key comparison between the 2 workflows: