Amit Sharma, Principal Researcher at Microsoft Research and co-creator of the DoWhy library, discusses the future of agentic systems and their impact on complex human tasks. He highlights the challenges in current frameworks, particularly around verification, while emphasizing innovative approaches in causal modeling. The conversation touches on integrating large language models to improve decision-making and the creation of the Duy library for causal inference, aiming to enhance accessibility for newcomers in the field. This engaging dialogue showcases the intersection of causality and AI in shaping robust systems.
01:10:10
forum Ask episode
web_stories AI Snips
view_agenda Chapters
menu_book Books
auto_awesome Transcript
info_circle Episode notes
insights INSIGHT
LLMs & Causality
LLMs revealed that causal reasoning relies heavily on domain knowledge, not just algorithms.
LLMs learn domain knowledge, enabling reasoning even without extensive data, challenging prior assumptions.
insights INSIGHT
Causality in Agents
Causality becomes crucial for agents interacting in real-world scenarios where actions and rewards matter.
Questions about agent usefulness, rewards, and blame require causal abstractions.
insights INSIGHT
Agents Mimicking Causal Reasoning
Agents in diverse environments can learn optimal actions mimicking true causal agents, even without world models.
Observing these agents' actions can potentially reveal their world model.
Get the Snipd Podcast app to discover more snips from this episode
Published in 1949, '1984' is a cautionary tale by George Orwell that explores the dangers of totalitarianism. The novel is set in a dystopian future where the world is divided into three super-states, with the protagonist Winston Smith living in Oceania, ruled by the mysterious and omnipotent leader Big Brother. Winston works at the Ministry of Truth, where he rewrites historical records to conform to the Party's ever-changing narrative. He begins an illicit love affair with Julia and starts to rebel against the Party, but they are eventually caught and subjected to brutal torture and indoctrination. The novel highlights themes of government surveillance, manipulation of language and history, and the suppression of individual freedom and independent thought.
Causality
Models, Reasoning, and Inference
Judea Pearl
This book provides a detailed analysis of causality, transforming it from a nebulous concept into a mathematical theory. It applies to various fields such as statistics, artificial intelligence, economics, philosophy, cognitive science, and the health and social sciences. Judea Pearl presents simple mathematical tools for studying causal connections and statistical associations, making it a valuable resource for students and professionals in multiple disciplines. The book has led to a paradigmatic change in how causality is treated in several scientific fields and has been cited in over 5,000 scientific publications.
The idea of agentic systems taking over more complex human tasks is compelling.
New "production-grade" frameworks to build agentic systems pop up, suggesting that we're close to achieving full automation of these challenging multi-step tasks.
But is the underlying agentic technology itself ready for production?
And if not, can LLM-based systems help us making better decisions?
Recent new developments in the DoWhy/PyWhy ecosystem might bring some answers.
Will they—combined with new methods for validating causal models now available in DoWhy—impact the way we build and interact with causal models in industry?
*About The Guest* Amit Sharma is a Principal Researcher at Microsoft Research and one of the original creators of the open-source Python library DoWhy, considered the "scikit-learn of causal inference." He holds a PhD in Computer Science from Cornell University. His research focuses on causality and its intersection with LLM-based and agentic systems. Amit deeply cares about the social impact of machine learning systems and sees causality as one of the main drivers of more useful and robust systems.
*About The Host* Aleksander (Alex) Molak is an independent machine learning researcher, educator, entrepreneur and a best-selling author in the area of causality (https://amzn.to/3QhsRz4 ).
Causal Bandits Podcast 
