The Fifth Elephant
The Fifth Elephant

The Fifth Elephant 2025 Winter Edition Call for Proposals

On AI disrupting analytics - make a submission

All submissions

Prasanna Bhogale

@pbhogale

Your Causal Parrot might be lying to you

Submitted Nov 12, 2025

Type of submission: 30 mins talk

The talk will start with an introduction / motivation about causal inference in general. Humans have excellent causal models of the world, so associational analytics suffice in human businesses because the causal model of the business is implicitly known to everyone involved, thats why - despite almost no one bothering with causal inference - most BI departments are doing just fine. On the other hand, there are situations where knowing the precise flow of causality determines what intervention is necessary and what form it takes. With this background, the talk will introduce Pearl’s ladder of causality, which will form the backbone of the rest of the talk. All of this should take ~ 10 minutes.

LLMs are wonderful at associational reasoning, but they are working off the average wisdom of the internet and will pretend to be causal (because humans habve filled the internet with casually causal language) but lack both the implicit model of your business that your mind possesses, as well as the tools to correct themselves. There are benchmarks that show that LLMs get progressively worse as one climbs the ladder of causation. Up to this point, I should be at the 15-17 minute mark. Then, the talk will quickly walk through a toy example of causal analysis in R using the wonderful ecosystem there up to the second rung of the ladder. ~10 more minutes. Leaving - ideally - 3-4 minutes for questions.

Main takeaways :
(a) The language of causality.. what is the ladder, what are interventions, counterfactuals, DAGs.
(b) Causal inference is wonderful, worth learning, and LLMs can’t do it out of the box so your agent will need explicit causal tools to reason about interventions and counterfactuals.

Target Audience : this is probably data scientists and data science managers who are dealing with non trivial and novel problems and want to see if new tools might be brought to bear on them.

Bio : I am a lapsed physicist with several years of experience in various Data Science and BI contexts, mostly in Germany. Now I am the founder of Romulan AI - building the causal layer for your LLM first business.

All submissions

Comments

{{ gettext('Login to leave a comment') }}

{{ gettext('Post a comment…') }}
{{ gettext('New comment') }}
{{ formTitle }}

{{ errorMsg }}

{{ gettext('No comments posted yet') }}

Hosted by

The Fifth Elephant
The Fifth Elephant
Jumpstart better data engineering and AI futures