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

The problem with artificial writing is writing alone doesn't accomplish much. Art is the expression itself, so AI art is the final product. But with writing, the goal is to often communicate something novel, in which case the AI wouldn't know, or transact and get something done, in which case the AI won't have the authority or access.

Take customer service chatbots. It really does not matter how "human" the bot is, if it isn't given any power, such as refunding an order. It can't really apologize either, without lying about its lack of empathy, creating a real life turing test between it and the customer.

Of course, businesses justify making refunds hard, so they often will prevent even humans that power, and arm them with templates and legal excuses anyway. Which explains why most customer service is sh*t. They're not there to help you, as much as they are paid to entertain the thought of helping you.

With code, at least the code is itself the end result, like the art. And it can be tested with execution and with unit testing. But most programming is still deciding what to name functions and what functions are needed. The implementation is just the technical task that "anyone" or "anybot" might do. It's the mindless part.

So like AI driving, the last mile of the problem may prove to be close to impossibly hard. Or at least, it isn't an AI problem that can be solved with current AI. It needs another breakthrough. Until then, even AI will continue to rely on human input. The furthest we'll get with AI coders is AI coders taking instructions from a coder, with the end result heavily dependent on what the human adds, not what the AI adds.



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

Search: