"No - that's a bad idea"
Exploring the value of antagonistic AI in creative processes
“Withholding feedback is choosing comfort over growth.”
– Adam Grant
When we think about using generative AI in creative processes we mostly think about it as a tool, an extension, an augmentation, or in the most extreme case a replacement of the human. These definitions have something in common, as they assume the AI to be under the command of the creative human, as being used by the creative human.
On the other hand, we would instinctively categorize any scenarios, where the AI does not do what the human wants on a spectrum from "annoying” (why did it not write the sentence in exactly the style I asked for?) to “extremely scary” (why did this autonomous AI Agent develop and deploy a biological weapon?).
But if we think about the quote from the beginning and reflect on how much we need adversity and challenge to develop great ideas, we can begin to see a space for AI systems that do not just agree with us and follow our every command…
This idea of “Antagonistic AI” will be the topic of this /promptcollective newsletter. First I will share a hands-on experiment with you and then give my thoughts on a theoretical framework that supports the approach.
THE EXPERIMENT
Let’s take a look at two very different interactions with GPT4 - one with just the base setup and one with a custom GPT that I created called “Susan”.
We are starting off with ChatGPT classic:
Ok - what did I do here? I gave ChatGPT something very unoriginal, actually the exact logline for The Godfather, and asked for feedback. And while it picks up on the Godfather parallel, it goes on telling me basically that this is a great idea that I should keep working on… And when I ask it to write me a new logline it just does it. The tone of the interaction is super friendly and supportive all the way, but I did not really get anywhere and it all feels quite generic.
Now let’s see how it goes with Susan:
Quite the difference right? Susan has some personality (and a cat apparently). She calls me out for just copying but also helps me see that I can create something original while still being inspired by great masterpieces. And when I ask her to do my work, she straight up refuses and pushes me to keep at it.
Imagine yourself working on a creative idea - what interaction do you think would be more helpful to you in the long run? I know that for me getting some pushback and people (both carbon and silicone based) challenging my ideas is what helps me most in making my ideas better (or realizing that I should maybe start over).
If you want to chat to Susan yourself you can find her here: But be warned - she is not easy to impress!
Susan has been with me for quite a while, I created the system prompt for her back in early 2023 when we started experimenting with creating LLM-based co-workers for a virtual writers room. While having eager and positive colleagues who build on your ideas, praise you, and do your bidding is nice, I always found the conversations with Susan the most engaging and interesting.
THE THEORY
This always made sense for me intuitively, but I did not really think of it as anything more than just one of the fun persons I have been creating for various experiments and productions - see some examples in these /promptcollective newsletters:
But then some months ago I came across a really interesting research paper that I have been thinking a lot about since:
In this paper, the authors explore the idea of a new design framework for AI that they label “Antagonistic AI”. Here is a quick summary of my highlights from the paper. First I loved this characterization of the effect of traditional AI Assistants:
And their effects on us humans:
The authors propose antagonistic AI because it can prompt us to think and maybe even change:
In the study, they showcase different ideas for antagonistic AI systems that they developed in a speculative design workshop, and I feel that Susan would be in good company there.
I will definitively keep exploring this notion of antagonistic AI in the context of creative processes and would love to hear from you if you have tried out Susan or created an antagonistic System of your own!
Who Are We?
The /promptcollective was founded by Jes Brandhøj (Denmark) and Hannes Jakobsen (Germany). We're on the lookout for like-minded enthusiasts. If you're passionate about the AI-creative nexus, reach out!
Nice article. Thanks
Superinteresting, I will try your Susan asap!