Welcome to my official RU-vid channel, Causal Python, with Alex Molak. Dive into the fascinating world of Causal AI, unraveling the complexities of Causal Inference and Discovery with Python.
My content simplifies these intricate topics, making them accessible whether you’re starting or advancing your knowledge. Here, I explore the intersections of causality, AI, machine learning, optimization, and decision-making, all through Python’s versatile capabilities.
This is also the home of The Causal Bandits Podcast. For more insightful discussions, check out my Causal Bandit Podcast Website.
For Collaboration and Business inquiries, please use the contact information below:
📩 Email: hello@causalpython.io
🔔 Elevate Your AI and Machine Learning Skills to New Heights! Subscribe now for clear, accessible insights into Causal AI and Machine Learning. www.youtube.com/@CausalPython/?sub_confirmation=1
Such an inspiring interview! I really loved the musical ending: what song was that? Also, who is the person mentioned by Judea who said that causality is ascientific, as you can follow the chain of causal links down to the big bang? Thanks!
Hi Carlo, I am glad you liked it! Thank you for sharing. The song is "Shalom Aleichem" -- a traditional 16th/17th century Jewish song that is traditionally sung on Shabbat evening. The person that called finding causes of effects the "cocktail party chatter" was Donald Rubin. Here's Alex Vasilescu's blog post that briefly discusses this story: www.aiacceleratorinstitute.com/causal-explanations-in-computer-graphics-computer-vision-and-machine-learning/
❓Should we build a Causal Experts Network to connect you with other like-minded people in causality? ❗Share your thoughts in the survey: bit.ly/3RM8ziz
❗Should we build a Causal Experts Network to connect you with other like-minded people in causality? ❗Share your thoughts in the survey: bit.ly/3RM8ziz
❗Should we build a Causal Experts Network to connect you with other like-minded people in causality? ❗Share your thoughts in the survey: bit.ly/3RM8ziz
❗Should we build a Causal Experts Network to connect you with other like-minded people in causality? ❗Share your thoughts in the survey: bit.ly/3RM8ziz
❗Should we build a Causal Experts Network to connect you with other like-minded people in causality? ❗Share your thoughts in the survey: bit.ly/3RM8ziz
❗Should we build a Causal Experts Network to connect you with other like-minded people in causality? ❗Share your thoughts in the survey: bit.ly/3RM8ziz
❗Should we build a Causal Experts Network to connect you with other like-minded people in causality? ❗Share your thoughts in the survey: bit.ly/3RM8ziz
❗Should we build a Causal Experts Network to connect you with other like-minded people in causality? ❗Share your thoughts in the survey: bit.ly/3RM8ziz
Very good question, @DistortedV12 Generally speaking, the environments will be characterized by the same underlying causal structure/mechanism with respect to the variables of interest, but they will have different joint distributions/covariate shifts (e.g. exchangeable, but not iid data). One example would be different hospitals that have slightly different admission or treatment administration rules or are operating in socio-economically different areas. Any particular paper might define slightly different set of assumptions regarding the data generating process or distribution properties, but that's the general idea.
Bernhard quoted a pioneering ethologist Konrad Lorenz, saying that "thinking is nothing more than acting in an imagined space". The intuition here is that thinking is a mental simulation that we carry out in the imagined space (produced by our minds) that perhaps also involves us acting in this imagined space. Does this answer your question?
only problem is that if you dig into the latent space, you will find out that even in such space, you have regime changes, restating causal arrow in opposite directions.
only thing is to achieve causality it is expensive, even simply using observational data. often times, association plus domain knowledge is good enough. we don't need to principle every step, like some perfect algorithm. close enough is good enough, especially in a world of limited data. we don't need a hammer for everything. different tools for different situation works too.
I feel like with the sora commentary, why not fine tune a physically valid version of it? We've been doing this in the language domain to get at factuality, and can surely be done here if the output is rendered as 3D. Just use your strongest physics simulator to provide feedback or do some kind of self play like in alphago.
Thank you for the comment @DistortedV12 Very good points. I believe combining symbolic representations (like simulators) with generative models can be a promising direction. There are some interesting works in this area, and hopefully we'll see more interest in the community in this kind of ideas.
You are getting more and more big names. I wouldn't be surprised if Imbens will come on soon. I personally would like to see more people using causality with multivariate high stakes settings involving temporal classification. Maybe finance again?
