When you look at the following video, you can intuitively reason about causal relations, such as what is moving the bat, what is causing the ball’s direction to change, what would happen if the ball flew a bit higher or lower, and so on.
Machine learning, however, lacks sorely in figuring out causal relations. And bridging this gap has perplexed artificial intelligence scientists for years. My latest article on TechTalks unpacks machine learning’s struggles with causality, drawing ideas from a recent paper by researchers at the Max Planck Institute for Intelligent Systems, the Montreal Institute for Learning Algorithms (Mila), and Google Research.
We still don’t know how to create causal machine learning systems, but the new paper brings together interesting ideas from different—and often conflicting—schools of thoughts and provides interesting directions for future research.
Read the full article here.
In case you’re interested on other perspectives on the challenges of machine learning, see these posts: