How will ChatGPT affect BI/Analytics? And Semantic Layers? And how will ChatGPT affect illumex?

I’ve been getting A LOT of those questions in the past couple of months.

This is my take on the possible effects ChatGPT/Bard and similar apps will have on the Data and Analytics space.

TLDR: the future is here, and it is more evenly distributed.

ChatGPT is the conversational and feedback-loop interface on top of the GPT 3.5 language model (LLM). Most likely, your organization is already using one of those LLMs based on Transformers (the GPT of the previous versions or BERT, for example) either embedded in your homegrown solutions or via 3rd parties.

LLMs are a subset of the more extensive Generative AI domain, which addresses multi-modal auto-generated experiences in audio, video, and text. It was hard to miss the auto-generated pictures and deep-fake video clips that filled our social threads last year. In the business world, we see more and more usage of synthetically generated data for many industries, for example – for drug manufacturing.

My short answer on the “ChatGPT effect” is that the crazy popularity and adoption by the broad public are great for the whole Generative AI industry. The openness  that businesses demonstrate in  trying to  adopt this new tech is much higher than any previous tech trend, maybe because users can actually play with it in their spare time as a hobby. Familiarity makes it less intimidating. In addition, this new tech comes from new vendors and is mainly served by the “top 4”, giving more chances for smaller vendors to penetrate enterprises directly or via partnerships.

What about Analytics? Imagine you ask questions via a chatbot, they are translated into a query and are navigated to the correct data.Your favorite BI tool displays the result’s visualization. Why do we still employ analysts and data engineers?

Let’s assume that the current very limited capabilities of the LLMs to do mathematical analysis, especially over time, will improve rapidly.

Are you familiar with the saying: “Torture your data long enough, and it will confess to anything”?

In this context, it means that at least for the mid-term, you will still need someone to curate the answers and help with asking the right questions, and interpreting the results to turn them into actionable decisions.

But the biggest problem is that LLM has its data model. Your data layer has one data model, and the application you use to represent the answers has its own data model. This creates a high chance of getting lost in translation. There is a common and wrong assumption that data objects’ names are semantically meaningful within organizations, and thus LLMs will automate the generation of knowledge graphs, ontologies and semantic layers. Trusting your metric store being auto-generated by a tech trained on literary classics like “War and Peace” or “Dr. Suess” is a big stretch.

In addition, LLMs are a “black box” – it is impossible to reverse engineer their way of “thinking”, which reduces transparency and potentially deteriorates the trust in analytics result, a problem which the analytics space has already been fighting for a long time.

And how does ChatGPT affect illumex?

Illumex is a proprietary Generative AI engine that automatically maps and interprets metadata, and connects data and analytics layers without human intervention—making it the World’s first Active Semantic Layer trained on the industry-specific data corpus.

We are happy to leverage the hype around Generative AI. Our solution is designed to keep the controls over ontologies creation and metric generation that are automated, to mitigate the above mentioned risks. illumex’s augmented governance allows our customers to control their semantics and metric layer and to avoid the data model “shift” due to new trends.

The future is here, and now it is more evenly distributed.