The recent explosion and domestication of tools to facilitate computational creativity raises a significant number of critical and creative questions. With tools like GPT-3, DALL-E, Midjourney, Deep Dream, and Artbreeder just a click away, the contemporary capacity to co-create with machine systems represents an entirely different set of relations. Clearly, it is vital to think critically about what is at stake in such co-creations – to investigate questions of bias, model training, “art,” and ownership, to name but a few. In this workshop, we are inspired by the philosophy of the AI Anarchies Autumn School, which proposes to push back against “a general lack of imagination about what our lives with AI might be” and acknowledges “that our tableau of algorithmic timelines, natural language processing bots, and bustling data marketplaces is the world we now live in.” (https://www.holo.mg/dossiers/ai-anarchies/ emphasis original) With these orientations in mind, we believe that one of the richest ways to unlock important critical questions, to produce fruitful pathways for socially engaged investigation, and to flex our imaginations in the world we now live in is through direct experimentation with the tools.
What do we want to create? What do we want to test with and within these tools? How can the process of testing, creation, and critical making work together? This workshop approaches our current AI-saturated environment as a given and asks both “What do we make of it?” and also “What can we make with it?”
This hands-on workshop provides a semi-structured approach to experimenting with two specific tools that allow us to co-create with the machine – Zac Whalen’s text-based Markov Chain generator, and the AI image generator Midjourney. We encourage participants to take whatever critical path they choose through the workshop’s process. Some may choose to test the tools, to see where we might expose the gaps, omissions, and biases. Some may choose to work with, rather than against, the tools to see what critical and creative possibilities they offer. Many will try both. Our objective is to provide a structure within which participants can experiment with their own questions about computational creativity from the variety of standpoints that the HASTAC community comprises.
For the first half of the workshop, participants will be taught how to use the Markov chain and will be guided through a few prototype experiments before being set loose to develop their own texts. We expect there to be significant collaboration on these texts as participants share their discoveries, help each other to develop new queries, and work independently or in teams as they choose. For the second half of the workshop, we begin by walking the users through the basics of Midjourney prompting, and introduce them to Shane McGeehan’s prompter tool. We then invite the participants to use some or all of the text they’ve generated with the Markov chain as a prompt for image generation. The ultimate output, we hope, is a text/image combination that participants can share on HASTAC Commons, perhaps with a critical paragraph about the process of its creation. There will, of course, be plenty of time for general experimentation with the tools and for organic collaborations to emerge.
Why Markov chain?
While not technically an AI tool, a Markov chain is a rich site for exploration and experimentation. Markov chains allow you train them quickly on specific data, so each participant can train the tool on texts of their choosing. As a result, Markov chains are much more interpretable than a conventional black box AI, so it will be easier for participants to understand how and why the tool is behaving as it does, thus producing much more fruitful avenues for iteration.
Why Midjourney
There are not such clearcut advantages to any image generation tool. We’ve selected Midjourney because it provides an explicitly social environment for creation (via Discord), and because there are some excellent materials that help people get up to speed quickly. Importantly though, it has been quite responsive to user feedback – particularly on questions of bias in its outputs, so provides an interesting case when experimenting with critical making.
Information for participants: This workshop presumes no previous experience with either tool. Participants are asked to bring a laptop or tablet; both tools (along with our collaboration environment, Padlet) can be accessed through a web browser, so no software downloads are required and all tools are platform agnostic. However, participants must have a Google account in order to use the Markov chain and a Discord ID to use Midjourney. There are no workarounds. We will provide a selection of prepared texts for the Markov portion of the workshop. However, participants are also encouraged to bring their own text files (.txt) that they think would be interesting to use to train the tool. Although not necessary, things will be quicker (and possibly more fun) if these texts are already in Google Drive before we begin. Note: while participants can use the Markov chain as much as they like, trial accounts for Midjourney are, unfortunately, limited. We suggest participants not experiment with Midjourney before the session, unless you can sign up for a second Discord ID to restore your account limits.