Pedro Domingos

Forget all. We need to understand the brain .

I wish you'd write out "Machine Learning". Every time I see ML I think, "Maximum Likelihood". Have pity on us old guys.

Connectionist is not necessarily differentiable, e.g. any local learning based algos like SOMs. Those are likely to eat the world in the long run.

If I had to put a stake in some "brain-like" algorithm, it would involve learning in terms of a differentiable or "correlational" causal model. We exist not in perception space, but in causal model space. Bio seems to learn by correlation, but perceive by inference.

As long as we don't realize that INTELIGENCE is our greatest asset and to learn to use it, we will continue finding local minima

We started w/scalar computers & found múltiple partial solutions. We developed superscalar, vectorial, parallel computers but still stuck within local minima. Now qubits. Complexity has been increasing accordingly. Will we ever find global minimum? NO The core issue: Intelligence

So symbolists reconfiguring themselves as connectionists are really contortionists?