Thread by @timothyvienne

What's the one #STEM-y book to read this year? Check out @yudapearl THE BOOK OF WHY amazon.ae/dp/0141982411/…

Why, you might ask? 😉 His work helps scientists learn about the 🌎 & avoid flawed conclusions 👉 can help you do the same! A THREAD... 🧵: (1/)
#amreading #datascience

@yudapearl set out to create true #AI in computer science in the 80s, but couldn't find a way to make machines learn from data. It didn't matter how good the data was, machines couldn't identify *relevant* vs. *irrelevant* patterns. (2/)

For example, @Google famously invented an #AI algorithm that learned to identify cats in @YouTube videos. That's cool, but how does a machine know when we want to identify a cat because we want to adopt one *or* b/c we are searching for our cat that ran away 😿? (3/)

Data doesn't distinguish between these two stories, but stories (causal stories) are how humans process the world. @yudapearl argued that we need *causal diagrams* to really learn about the world from #data. What is a causal diagram? (4/)

It's just a bunch of factors with arrows between them. As in cold virus ➡️ sniffles, or hungry tummy ➡️ eating food. To humans this is obvious, but in #datascience, it's not. Data are dumb--they don't tell you about causes, just patterns. With just cutesy diagrams, we can (5/)

know ahead of time *what we can learn from data before we even collect it.* So for the #COVID19 vaccine -- if we know what factors contribute to people getting COVID, we can make sure we collect that data so we can do nuanced comparisons over time and see how the vaccines (6/)

fare in the real world, not just in a lab. It's a fascinating book that even includes some history (and only a bit of math, most of which you can skip). I encourage you all to check it out as a stretch read! (7/)