Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

The issue with automating software is that you first need to specify exactly what you want, and what should happen in all the edge cases. The best way to do that is with a domain-specific language, which is what programming languages are.

I don't think the issue is that there isn't a tool that understands natural language, it's that even in natural language you need to specify everything before it works.

Who knows, though, maybe we'll see a piece of AI that can make educated guesses from rough descriptions and make it work.



> maybe we'll see a piece of AI that can make educated guesses from rough descriptions and make it work

Right. If we can build a strong AI, then we can certainly automate all stages of software development, by definition.

The article's argument essentially boils down to I have certain assumptions about the limitations of AI, and I've decided software development is one of the things that will always be beyond AI. It isn't a technical argument. It does a poor job of giving a solid justification. These sorts of arguments don't have a great history. From the article:

> Creating software is a human collaborative activity. Humans are good at working with other humans and with finding solutions. Computers can only automate using rules, but software projects have to work out the rules as they go along.

We already have AI that can do a surprisingly good job of writing English prose. A few years ago I'm sure people were making exactly the same kinds of arguments about how writing can't be ever automated because computers only ever follow rules.

edit In reality I imagine we'll see incremental progress, as we've seen with GitHub Copilot. I'm not saying we can expect to see software development automated overnight, but the idea that it's impossible even in principle for it to be automated, is not a position I've ever seen a good defence for.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: