“Another advantage causal models have that data mining and deep learning lack is adaptability“
Judea Pearl, along with Dana Mackenzie, provide an insightful and thought provoking view of Causal Inference.
Judea Pearl is well known for his contributions to AI, Bayesian Networks and Causal Analysis. He challenges and questions many beliefs held by statisticians, and not all agree with his views. It is worth reading this fantastic review by Kevin Gray, which includes a response by Pearl, and then a response by Gray, followed by a response by Pearl…
The premise of the book is to bring to light the ‘Causal Revolution’, and to detail how to climb the ‘Ladder of Causation’, and enable AI to reach the top:
- Counterfactuals: the ability to imagine, understand and ask Why?
- Intervention: the ability to do and intervene
- Association: the ability to see and observe
The claim is that the Causal Revolution will allow AI to climb the ladder, and climb it must according to Pearl, as “deep neural networks have added many layers of complexity of the fitted function, but raw data still drives the fitting process”.
The book provides some great historical context to not only Causal Analysis, but also Randomised Controlled Trials, Bayesian Analysis and Bayesian Networks, and details the connections between them.
It’s a great starting point to anyone interesting in learning more about the fascinating world of Causal Analysis.
Key takeaways:
- Causal Diagrams, simple dot and arrow pictures, can help summarise our existing scientific knowledge
- Data alone can’t provide an answer when causal questions arise
- Strong AI, ie machines that think like humans, is where we’re heading
- Causal models can help answer counterfactual questions