Refering to his work related to malaria. 

Author argues, instead of merely correlating fever with malaria, machines need the ability to reason that malaria causes fever. 

Once this kind of causal framework is in place, machines will gain the ability to ask counterfactual questions. 

These counterfactual questions allow the automated modification of causal framework depending on the given intervention. 

Example application of causal reasoning in AI

We need to think about how truly intelligent machines would think. 

The KEY is to replace reasoning by association (Bayesian Network) with causal reasoning. 

University of Michigan - Ann Arbor

He thinks that current artificial intelligence research is stuck in probabilistic associations. 

He expresses today's (in 2018) artificial intelligence is powered up versions of what machines could do three decades ago. He defines the machines as "finding hidden regularities in large sets of data"

He is not impressed by the current achievements in deep learning as it only amounts to just intensive curve fitting. 

Learn Before

Related

Learn After