Hacker News Reader

Comment by karmakaze | Hacker News Reader

I wouldn't say thanks to information theory, but rather on human tendency to be bored with repeated styles. It doesn't matter how good the tech is, but if it comes out like Michael Bay movies or anything else that we learn to recognize, people are going to want 'more original' art. The other difference is that the point of art is to express and evoke emotions, it's an open question whether that can be done effectively without having feelings during the process.

View on HN

This is a good take and under appreciated generally. Already theres an identifiable generic flavour to much AI art. Just knowing that it’s AI generated makes it feel somewhat lifeless. Not because of the end product is lacking , but because of the awareness of the production method. it’s less engaging if you know a machine spat it out, especially once you recognise and become bored by the stylistic markers that indicate machine generation. That will play into the spending calculation of, for example, advertising agencies, particularly high end brand work.

I’ve seen some screenshots of ai ads already. They are usually somewhat horrying, like the person will be smiling, but you start to notice their mouths are open too large and they have too many teeth and lack eyelids, and the people in the background are even more distorted and nightmareish.

I do landscape photography and the AI produced landscapes that I've seen have things that are subtly wrong and annoying/disturbing upon inspection. Things like the wrong types of clouds showing up (a lenticular cloud in the wrong spots in the wrong seasons), sunlight coming from multiple angles or the wrong angle for that time of day, reflections in water (or the lack of them), aerial perspective being peculiar, and shadows just being wrong.

While I don't doubt that someday, an AI of some sort will be able to get that right, the knowledge of the items in the scene and how that relates to the physical world isn't something that that the current models are able to recreate.

Bingo - solving the problem of making "good" AI art seems close to solving the problem of making AI human.

I wouldn't make that strong a statement anymore. The level of play by AlphaGo/Zero seems to demonstrate creativity and understanding that we would only attribute to humans. That is to say I can't say for certain that emotion couldn't be faked, and we don't know that human emotion is super-special, only that it's complex.

Do they demonstrate creativity, or is that just us romanticizing it? Maybe these games really are just reducible to numbers at the end of the day.

They do in the sense that choices are so far off what's known good "based on the numbers". How else do we define creativity? With the brute force search tree being roughly 191^lookahead and outcome being life/death of a groups of stones that depend on distantly neighboring influences, it's hard to call it simply computation.

Reply on news.ycombinator.com