Why would AI tools designed specifically for processes we call creative, not be creative?

Firstly, interesting creative action often happens when things get broken, or they glitch. Art happens when things break. Not all art of course, but cool stuff can happen when things don’t work as they aren’t meant to. However, this is a real problem for generative tools because everyone is thinking about ways to stop them breaking. AI ethics is huge and has its sights set firmly on the development of generative models. I published a paper about these risks last year. So it’s easy to get them to do silly, uncanny, weird stuff. It isn’t easy to truly break them. Also, breaking things often requires some skill to undo what has been done. Again, difficult. There is the HUGE problem of ethics also. It doesn’t feel good that these models rely on human data and labour to produce this content. And they’re going to keep needing content if we don’t want innovation to stagnate, right? They evidence mode collapse. That’s proof enough (for me) that they’re somewhat, maybe accidentally, predatory. They’re anti-creative just for that reason, I’d argue.

Secondly, generative models often function is isolation. Click in, generate, move the image out. Move the text block out. Copy and paste. But often creative action requires margins, buffers, zones of movement. The capacity to move things in and out, play around, break (!) things. Like imagine you could copy and paste in ideas and images into a workspace for a generative model and it keeps building and changing. You can keep things on the sideline. It isn’t just prompt and done. It becomes a process. That’s where cool things happen.

Finally, creative processes often need the ability to view shift. To look at something a little differently. Usefully for traditional mediums, you can do this by physically moving, getting closer to further away. Now of course you can do this with generating content (zoom in), but in my experience it isn’t the same. It’s asking a lot of your brain to see this flat generated image in a new way, right?

What I am really trying to say is that creative processes need flexibility. You need to be able to move stuff about and break it. Generating an image just doesn’t give you this in my opinion. I could be wrong, but I’d really like to see gen-AI research that pushes towards flexible interactions, like a conversation.