Hacker News new | past | comments | ask | show | jobs | submit login

The statistical approach is painfully naive and doesn't work - as is obvious from the example.

It's like feeding a net with the complete works of Shakespeare and expecting it to produce a genius-level original play. It's simply not going to happen.

The issue is not with the statistical but with the parameters around the output and the organization of training data.

I think that your assumption is that the genius of Shakespeare's plays can be statistically reproduced through sufficiently-clever organisation. That is not obviously true to me.

Some things are just art, capable of being truly understood only by a creature with a head and heart, arms & legs, love & hate, emotions, experiences ­— in short, a man.

Art is about perception not creation. If autonomously created art produces the same perception and evocative response, it sufficiently passes that test.

An autonomous agent doesn't need to understand the underlying emotion, it just needs to mimic it.

A autonomous car doesn't know why it shouldn't hit a child that jumps into its path, or why it is making any decisions at all despite those being some of the most important and fundamental to humans on earth. It just needs to reproduce the actions of a human that does understand those things.

Yes, I believe that artificial creativity will produce art will be indistinguishable from that made by humans.

So you're saying humans are not just cleverly organised statistical automata? If so, science would disagree with that position.

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