We're already at the point where abstractions need to exist because devices (such as a modern x64 CPU) are so incredibly advanced. Even knowing assembly is a high level abstraction over an incredibly advanced closed-source chip. This article is more about someone afraid of new technology than it is about someone lamenting that ML is inaccessible or "unknowable", when all that matters is the observed behavior for day to day usage.
>> This article is more about someone afraid of new technology
This is a predictable response but just doesn't hold up. The path through abstractions for a modern CPU/GPU is more straightforward and linear than AI, and you don't have to go very deep to be back to 50-yr-old principles.
>> when all that matters is the observed behavior for day to day usage.
The entire point of the post is that observed behaviour is NOT what matters, vs understanding how we get to these outcomes.
