Some of you might know I have been writing a book about my dating experiences.
Its moving along thanks to some great friends who have done such a great job editing, structuring and shaping the book. But one thing I turned my attention to a while ago, is the illustrations.
I did pay for an artist out of my own money but wasn’t quite happy with every single illustration for each chapter, so only had about half done. The rest I’m talking to another artist about but recently been quite impressed with the AI art generators like DALL-E 2, Midjourney, Nightcafe.
The generated works are strange and abstract enough to fit with what I’m looking for in the book. Not only that, the ownership and copyright seems to be working out (from what I read using DALL-E 2).
(c) Copyright. OpenAI will not assert copyright over Content generated by the API for you or your end users.
I certainly seen the AI bias in some of the images generated. For example if I don’t say what gender or race the person is, the AI defaults to male and white. Its only when I deliberately say Black male / female it then switches. I would also say the images of black women are not as fully thought out as white women. Because I’m generating pictures of dating, it always defaults to straight dating unless I add something to the query. Likewise the women are always thin never curvy unless specified. Actually a few times, I got women who were pregnant. Of course every single time I make a query, it takes credit (money) making it costly to really test its bias, sure someones already on this.
The big question I have is, if I was to use DALL-E for illustrations in my book, what would that say or mean for my stance around AI, bias and data use? To be honest, I’m actually thinking about generating the front & back covers in full colour, rather than the in book illustrations.
Maybe I should be less worried about this? Or even better I was thinking about ways to not just make clear it’s AI generated but show the process of selection or something similar?