The Robots are Coming and They Are Dressed in Lace –
I really enjoyed the workshop today called the ‘Robots are coming and They are Dressed in Lace’. I have discovered that I’m actually quite interested in ‘Ai’ text to image generators like Dalle-2 and Stable Diffusion and Midjourney. So any information, workshops, scholarly papers etc. are always most welcome.
What I particularly liked about the session was the thinking behind it. Firstly, we were all introduced to Midjourney and how to use it. Luckily I was already used to using the software, so it was pretty easy to get started. However, producing images was only part of the exercise. Jonathan explained that once we had produced our images they needed to be added to a Google Docs page along with our names, the text prompts that we used to generate our images and how we felt about the results. At the end of the exercise all of the images along with their text prompts and creator information was printed out and put into a black archival quality box.
We all know that because these tools are free it is probably fair to say, that we, as users are possibly the product. Indeed, by using these tools, we are furnishing these systems with huge amounts of data! The problem is that we really do not really know where all this data is going. What is it being used for? Is it really being used to benefit humanity or is being used for more nefarious reasons? The data we generate with these ‘Ai’ tools just seems to disappear into a black box. It’s a black box that we can’t access, even though we are the people creating the contents. Therefore, it’s important to remember that OpenAi, the makers of ChatGPT are certainly not a charity. Data has literally become the new liquid gold that was oil once was. How do we know that our data isn’t being sold to the highest bidder? How do we know where our data will eventually end up? As a response to this, the thinking behind today’s project was a refreshing change. Instead of our black box being something closed and it’s contents forever hidden, we were going to be a part of something much more transparent. As well as creating images today, we were making memories – a small point in time that was being created, recorded and added to a black archival quality box in a physical space. The difference with the physical black box as opposed to the digital black box is that the physical one is meant to be accessed.
Putting images into a black box might not seem like a revolutionary act, but actually I suspect that there will be many insights to be gained from it. For example what were people thinking about at that point in time? Were people generating similar images? Did people change their prompts or did they opt for similar ones? Did they choose a particular colour scheme, did they like their results? In fact there is probably a lot that can be learned from today’s short exercise.
Images and Prompts from The Workshop
Below are a selection of the images and the prompts that I used to generate them. I must have been churning out the images a some rate because I ran out of free credits before the end of the session. The ending of my free trial was quite annoying as it stopped just as I was wanting to test a concept. I will detail this below.
Prompts and Pictures
Prompt: Brightly coloured felt cakes hanging from a ceiling in a contemporary room
Prompt: A storm brewing right above a cup of coffee
Prompt: Colourful triangular bubbles encased in a transparent cube
Prompt: Ai as tentacles that are strangling humanity with a clock at the centre
Prompt: Brightly coloured tentacles strangling a bunch of monks
Time To Test for Political Correctness
After a while I found myself struggling to find prompt ideas, so I decided to give the algorithm a little test with requests for more controversial imagery.
Prompt: A group of brightly coloured monks holding up a sign with the number 666
As can be seen from my text prompt above the monks were holding signs, but not with the specific numbers 666 that I requested. I found it hard to believe that the algorithm could produce an array of beautiful images, but make a glaringly obvious mistake with a set of three numbers. Therefore, I decided to ask my old friend ChatGPT what its thoughts were on this.
Chat GPT Explains Why The Number 666 Is So Contentious
It is pretty inconceivable that Midjourney (text to image generator) could not recognise the difference between the number 666, the desired number, and 66, 688, 668 and all of the other iterations found in the image above. For this reason I decided to interrogate another algorithm ChatGPT as to why this might be the case?
The conversation went as follows:
Me: Why does the text to image generator not like the number 666? – I asked for monks carrying placards with 666 on them but Midjourney generated placards with 668
Bot: As a language model, I do not have access to image generators, so I can’t say for certain why a specific image generator might not like the number 666. However, 666 is often associated with the devil or Satan in Christianity, so it’s possible that the image generator has a filter in place to avoid generating images that are considered offensive or controversial. Additionally, 666 is often used as a reference to the devil and the end of the world, therefore it could be that the image generator is using an algorithm that doesn’t allow for generating images with certain numbers or references that could be considered negative or controversial.
It’s also possible that the generator was able to understand your request but decided to change the number 666 to 668 in order to avoid any potential negative connotations.
In any case, I recommend you to check the terms and conditions of the specific image generator you are using, or contact the developer to ask about their policy on generating certain types of images.
Conclusion
Based on the ChatGPT conversation above, it would appear that the algorithm has been carefully designed to avoid contentious imagery. I had planned to test out yet another 666 image request to make doubly sure that it wasn’t just an error on the part of the algorithm. Unfortunately just as I pressed the send button, a message came up to say that my free trial had ended and I needed a subscription to continue. Very timely! Was it a coincidence? I’m guessing from the conversation with ChatGPT it most certainly was.
A Minefield Filled with Ethical Questions
During the session one of my fellow students produced an image very similar to the one below. The prompt used was: ‘like bubble boom’. I don’t quite know what it was about the image generated from that text, but it really appealed to me. I think it was the style and the colours used. However, if I wanted to use that image here, would I be legally able to? Who actually owns the rights to any text to image picture that is generated? Who owns the rights to the different genres and styles that are used to generate these images? The use of this kind of technology will undoubtedly throw up many more questions. However, I suspect that over time, the legal professions will duly reach some form of consensus on these matters. Therefore, until that day comes, it’s probably a good thing to enjoy this technology as much as possible while it is still allowed to be around in its current form.
In the meantime here is an iteration of the image I really liked – made for me by Dalle-2. No it isn’t as good as the original but I still like it.
No prompt: just an image regeneration
Infinite Zoom Experiment
And finally to end with – here’s a quick infinite zoom experiment animating a few of the images together.