[13years]

Fundamentally, I think AI alignment is an ML problem

I disagree and think this is one of the great fallacies. The current predominant view of alignment theory is that we need to align the AI with humanities "values". This makes it a human problem and all the discussions you see on alignment issues could be described as human behavioral issues.

Most critically, I think AI alignment is a solvable problem.

I strongly disagree and I know I'm in a significant minority on this point of view. However, at least under the current premise for which alignment theory is based, it is a paradox. A logical contradiction. It will not be solvable.

My case for this I've expressed in detail at the following - https://dakara.substack.com/p/ai-singularity-the-hubris-trap