I had originally been quite impressed with AlphaStar, but this post showed me its actual limitations. It also gave me a good concrete example about what a lack of causal reasoning means in practice, and afterwards when I’ve seen posts about impressive-looking AI systems, I’ve asked myself “does this system fail to exhibit causal reasoning in the same way that AlphaStar did?”.
I had originally been quite impressed with AlphaStar, but this post showed me its actual limitations. It also gave me a good concrete example about what a lack of causal reasoning means in practice, and afterwards when I’ve seen posts about impressive-looking AI systems, I’ve asked myself “does this system fail to exhibit causal reasoning in the same way that AlphaStar did?”.