|
|
|
|
|
by bitwize
434 days ago
|
|
> Can Gen-AI, moot the need for code? No, because if you read your SICP you will come across the aphorism that "programs must be written for people to read, and only incidentally for machines to execute." Relatedly is an idea I often quote against "low/no code tooling" that by the time you have an idea of what you want done specific enough for a computer to execute it, whatever symbols you use to express that idea -- be it through text, diagrams, special notation, sounds, etc. -- will be isomorphic to constructs in some programming language. Relatedly, Gerald Sussman once wrote that he sought a language in which to discuss ideas with his friends, both human and electronic. Code is a notation, like mathematical notation and musical notation. It stands outside prose because it expresses an idea for a procedure to be done by machine, specific enough to be unambiguously executable by said machine. No matter how hard you proompt, there's always going to be some vagueness and nuance in your English-language expression of the idea. To nail down the procedure unambiguously, you have to evaluate the idea in terms of code (or a sufficiently code-like notation as makes no difference). Even if you are working with a human-level (or greater) intelligence, it will be much easier for you and it to discuss some algorithm in terms of code than in an English-language description, at least if your mutual goal is a runnable version of the algorithm. Gen-AI will just make our electronic friends worthy of being called people; we will still need a programming language to adequately share our ideas with them. |
|
Now tell that to your compiler, which turns instructions in a relatively high-level language into machine-language programs that no human will ever read.
AI is just the next logical stage in the same evolutionary journey. Your programs will be easier to read than they were, because they will be written in English. Your code, on the other hand, will matter as much as your compiler's x86 or ARM output does now: not at all, except in vanishingly-rare circumstances.