Hacker News Reader

Comment by ythl | Hacker News Reader

AI is still having trouble competing with us in terms of general intelligence and creative ability. I will be very surprised if AI can ever generate a movie, song, video game, or book completely on its own that can actually compete with human created entertainment.

View on HN

Your comment reminds me of a scene in the movie iRobot, when Will Smith's character asks a robot if he can write a symphony or paint a masterpiece.

The robot replies "Can you?"

Social Capital is what makes the generation of content valuable. I have no doubt there are millions if not billions of people who could write poetry or songs alive right now who have never done so and even if they could, nobody would read. Ideas are cheap. It's execution that matters and by execution everyone means: can you work with a team of influential and able human beings to make this idea valuable for others. A computer outputting content will never do that. The only thing AI would have going for it in that sense is that maybe people will dazzled by the initial idea that the computer is really really smart but even so there are a great many really really smart human beings right now living and producing ideas that you likely couldn't care less about. So I don't think computers are going to get far without friends in this regard. We have systems in place for helping us make friends. They don't.

Here's the state of things as far as AI-written films: https://www.youtube.com/watch?v=LY7x2Ihqjmc&ab_channel=ArsTe...

Generating a screenplay from a bunch of existing screenplays is about as good as generating a poem from a bunch of existing poems. I think there was a story the other day showcasing AI-generated poems, and all of them pretty much sucked. They might be able to pass as "modern art", but they were a far cry from poems like "O Captain, My Captain" or "The Highwayman".

Generating poems from a bunch of existing poems is pretty much what humans do also. Art is very derivative, and necessarily so.

To be clear, I'm not saying this example is good. (Unless you particularly enjoy absurdism).

Humans do exactly that though - an artist isn't born in a vacuum.

> Humans do exactly that though

There's a weak and a strong version of that claim. To say that humans use existing works in creating their own is easily provable. But what I don't think is easily provable is that the new work is entirely the product of existing works, with the brain contributing nothing new beyond an AI-like algorithm.

Humans also have another pretty large corpus of input data: their entire life experiences.

Why? The current state of the art certainly appears to be a very long way off from that, but there don't appear to be any fundamental limitations that would make it impossible forever.

Deep learning program to generate image from text https://github.com/paarthneekhara/text-to-image.

Art and music generation using tensorflow https://github.com/tensorflow/magenta.

I think the technology is already there, all we need is lots of training examples.

Reply on news.ycombinator.com