DALL-E 2: Emerging Content Category Breakdown
DALL-E 2: Emerging Content Category Breakdown
One interesting thing I’m observing about DALL-E 2 is its potential to create new kinds of content in emerging categories. If you treat this phenomenon at the surface level only, I think you’ll miss the big picture. Which is why, to help break down this working theory I have, I’ve created this visual diagram through the analogy of an iceberg:
Distasteful Stuff
We often talk about the risks of multimodal AI models to generate offensive, disturbing, or even horrific content. In general, for all the models, I think this is a very big risk. However, I think we may just see a lot of, “distasteful” stuff in general.
What do I mean by distasteful? Well, this is subjective, right? I personally don’t like cursive fonts, I find them distasteful. I used to hate maximalist design, but last week, I put out a video talking about how I find it refreshing now. On the other hand, there’s definitely a spectrum here on what counts as distasteful. Will people find ways to circumvent safety controls to get these models to portray horrific content? Yes, I think it’s a certainty. To be clear, I would also say that that stuff, is deeply distasteful.
At the same time, I would say the vast majority of art is distasteful - at least to you! There’s tons of fan art on the internet which I would argue isn’t any good, and different types of people just gravitate towards different kinds of art and taste that I would never check out personally. Different DALL-E 2 users will generate different kinds of art which I think will certainly disagree with your definition of, “tasteful” even if they don’t mean to.
The point I’m really trying to make here, I think the surface area for content you may find distasteful may be a lot greater than even the horrific, polarizing stuff (although that stuff may have a greater impact on you). And this category of “distasteful content” will be a greater amount of content you consume from image generation models in the future, I think more so than people realize today.
It does get complicated though - I would say by adding the keyword, “digital art” to your prompt, you do get some nicer, more artistic completions. So, aesthetically, maybe we could see some artistically pleasing content overall instead of distasteful stuff, but it really depends on the multimodal artist themselves far more than the actual prompt they use.
Stuff you’ve never seen before
A pure joy of using DALL-E 2 is the ability to instantly generate content we’ve never seen before. This could be content that defies a physical reality or just the unwritten arbitrary rules of society. I’m including some of my own fun examples below:
I think humans deeply crave novelty - this type of content is fun for us to see because we are fascinated in how something would actually play out in our physical and socially constructed world. There is a potential for misuse here too, but I actually think this category is all about tapping into human curiosity.
Stuff you can’t unsee
Finally, every once in a while, we’ll see content that is not only something we’ve never seen before, but for whatever reason, it connects with us as a collective and sticks with us. It could be an image or something which remains ingrained in our memory, connects with us for a personal reason, or defies our view of constructed reality enough to have an impact on us. To be clear, I’m not talking about disturbing, offensive, or traumatizing content here.
I’ve seen some examples on the internet before of content you can’t unsee:
I’ve even tried creating some DALL-E 2 content you can’t unsee of my own:
Now, to be clear, I don’t think, “stuff you can’t unsee” is necessarily a bad thing. Great art solicits a reaction from someone, and can be polarizing. I also think it’s interesting to see content creators find these interesting societal tid bits to focus on. To some extent, maybe this is what art is about.
The point that I think is interesting about this type of content - I think it will just show up in our feeds, so, we won’t really have a say over when we come across it or not. Perhaps, we may become desensitized to it. But I think it’s an important dimension to think about content in the future. In my view, pretty much anything you can imagine and visualize, can be created with DALL-E 2, so, I think this category of content will only grow. It’s definitely worth keeping an eye on.