human.c

Since 2023, we have all heard about AI making programming obsolete. But have we ever asked ourselves, what programming really is about?

It was always about translating fuzzy human intention into something concrete enough that a machine can execute it. It is more like (research + translation). Everytime you write code, you are making a small bet about how the system works. AI can now place a lot of those bets and run the tests faster than we can. And just because the guessing part was small enough to hold in your head while you typed, we never called it research.

But can an AI, sit with a failure long enough to ask itself "maybe my idea about the system was totally wrong"? Nope. Instead it'd just patch the symptom and move on.

While working on something, only human can reliably change the assumptions or the whole representation which the small "bets" depended upon. All the major AI labs, try to hide that "representation choice" and "result interpretation" being generative acts, are highly unstable, and require an exogenous medium ("humans") to create any tangible effect. Here "effect" refers to anything new or something which would make you gasp a "Aha!, that's elegant!".

Exploratory and combinatorial creativity is possible given the dense & quick feedback nature of the field (albeit, with a lot of inference time compute), but coming up with something transformational is still in our hands. While your favourite AI keeps puking out the same post hoc repair narrative, while being stuck in the same exploratory basin for hours, you could have moved on to explore rest of the landscape. AI increasingly automates search inside spaces humans define. The remaining frontier of programming is deciding what space should be searched, what evidence matters, and when the space itself must change.

Fred Brooks had described about Essential Difficulty and Accidental Difficulty. AI is perhaps the ultimate tool for the latter, and the former lays completely on our shoulders. HackerNews has a good thread on it.

Perhaps the programmer of the future is more like a scientist.

And at the end? the Human programmer still stands.