Responding to Humans in the Loop by Ge and Experimental Creative Writing with the Vectorized World by Allison Parrish.
Humans in the Loop
I don’t think I totally agree with the idea that neural style transfer couldn’t be used to make good art (even as a “Big Red Button” system), but I do agree with the overall sentiment expressed. It reminds me of the thesis expressed in this video, Smoother Animation ≠ Better Animation. Basically: people on the internet have been passing animated videos into AI black boxes to artificially increase the framerate (using interpolation) and then sharing them online. They claim the end result is “smoother,” and thus immediately more desirable. The video says this much better, but not only are there many reasons why you wouldn’t always want a smooth animation, you are completely flouting artistic intent by taking humans out of the equation.
What about timing? What about ease? What about intentionally adjusting framerate to mimic certain art styles? If you stick something into an AI black box without any way to modify or input your intent, you can’t expect to get something meaningful from it. While I don’t think there is anything wrong with animation with a higher frame rate or using interpolation software (and similarly I don’t think there is anything wrong with using style transfer as a tool), I do think (as is implied in Humans in the Loop) that problems arise when you apply AI to an existing work of art and call it finished.
To be honest, I’m actually all for 100% automation when humans aren’t getting much out of doing the work in the first place, and I even think that completely automated systems can produce fun & good art (not totally intended to be art, but this simulated subreddit is hilarious and beautiful). It is not like AI can’t also be created with intention. But ultimately, meaning in artwork is always imbued by humans. I agree that AI is meant to empower us, not replace us (though honestly… sometimes I feel like that wouldn’t be so bad).
Experimental Creative Writing with the Vectorized World
Ahhh, vector representations of language. Brings me back my not-so-distant past, when I thought that a concentration in AI would be a good idea… I very much enjoyed this presentation, and especially loved learning about her research on phonetic vector representations. Poetry is a great example of the thing that we’ve been talking about — something that humans do, that can be helped made by AI, but ultimately humans are the ones that read meaning into it. I think that in many ways, the tools she has created are an excellent example of how we can use AI with intention. I really enjoyed hearing the “random walk through phonetic similarity space.” Felt like the AI-generated version of Skwerl.