Do you agree with Judea that learning from data is not everything? [D]

Link: [Judea Pearl, 2011 ACM Turing Award Recipient](https://youtu.be/XExyqAYDnvw?si=BVyX-oEetFslAvZq&t=8285) (2:18:05) Quote: >There is a limitation to that which people not everybody understand. I already mentioned a limitation that you have a hierarchy here and going from correlation to causation and from causation from causation to explanation or to imagination. It's hard for people especially in machine learning to grasp that wall the limitation of one layer where one layer ends and the other one begins. Why? Because of two things. Machine learning school of thought has two paradigms that they love everybody love. Number one tabula raza I don't want to get any opinion I don't want to get any preconceived knowledge I want to derive everything by myself let the computer learn it and you find the word learning overused .. The other handcuff is let's do it the way that the brain does it. So if it looks like neurons interacting, it's good. If it looks like knowledge coming from rule system, it's bad because it's man-made .. Now there's limitation to that. We can prove today that you cannot do certain things by looking at data and data only. It's not a matter of opinion. It's a matter of mathematical proof that you cannot you can look at people who take aspirin all day and people whether or not they have headache all day and you cannot prove that the aspirin is what causes the headache. In particular, Judea states: **"It's not a matter of opinion. It's a matter of mathematical proof"**. So we have formal proof that there are fundamental limits of learning from data. Judea later in the interview states we have solutions to problems faced by the machine learning community; nonetheless they are not adopted because of hype. **Discussion.** Do you agree with Judea?