AI: Admin Life
Created by Anastasia Salter
Generated with ChatGPT 4.0, DALL-E, and Adobe Firefly
Design
"AI: Admin Life" was originally developed for the 2024 CCCC Fall Virtual Institute: Machine Writing and the Work of Rhetoric and Composition Program for a roundtable on Administration and generative AI facilitated by Sherry Rankins-Robertson, University of Central Florida; Aurora Matzke, Chapman University; Angela Clark-Oates, California State University, Sacramento; Kyle Jensen, Arizona State University; Priscila Santa Rosa, University of Central Florida; and myself, Anastasia Salter, University of Central Florida.

Given the levels of frustration and burn-out that frequently accompany such conversations, we decided to approach our initial discussion through play. To facilitate that dialogue, I developed this simple visual novel focused on the experiences of a hypothetical chair navigating a day while trying to balance the concerns of students, faculty, and upper administration. It is in conversation with the approach of a simple playable editorial Ian Bogost, Simon Ferrari, and Bobby Schweizer discuss in Newsgames: Journalism at Play: thus, it is intended more as a timely provocation than a true simulation.

While the emphasis of the play experience is on triaging the constant flow of questions and complaints to gain reputation with the various groups, there are other options in the interface that diverge from that emphasis on productivity. The player is able to explore the bookshelf repeatedly, finding new advice both on administration and generative AI each time, but at the cost of reputation with all factions. There is also a pencil cup that calls the player to their own writing - it cannot be selected. The overall design choices were all intentional, and were not generated by the system. Everything else, including the at times questionable humor, was generated based on these design patterns.
Generation
"AI: Admin Life" was primarily authored through conversations like this one, leveraging iterative dialogues with ChatGPT to shape the game's content and flow. The game is built in P5.js, a JavaScript library designed for web art that enables elements reminiscent of Adobe Flash games. Approximately 70 iterative prompts were used in combination with testing, writing, and rewriting code to build out all elements of the game's interactions. All the images were initially generated using DALL-E, followed by refining character variants in Adobe Firefly. Changes in color keywords, particularly for hair, were used to expand the range of characters based on ChatGPT's suggestions. This approach allowed for more rapid prototyping, although the resulting code is somewhat inefficient (over 1,100 lines of JavaScript for the game itself.) All the character names and exchanges were generated, allowing the resulting game to riff on the existing discourse around generative AI and higher education. Several sections of the game code were hand-authored to fix bugs and challenges in connecting visual content with the game's structure - for instance, all the mouseover zones are "hand-coded."

The images generated for the game's backgrounds with DALL-E used reference images from the University of Central Florida. Those were then recolored using a palette and aesthetics inspired by the early web, with a palette intentionally moving away from realism. Similarly, the images of the generated characters started with a single portrait of the designer (re-imagined as a game character), which in turn was refiltered through Adobe Firefly into different characters using the same underlying template structure. Other visual elements of the game, such as the book layout and interfaces, use P5.js's drawing features and were initially generated then heavily modified.
Extension
Participants at the CCCC Virtual Institute are invited to extend this generative view of our labor through their own contributions. During the writing session, you can contribute using the AI Admin: Scenario Submission form. Scenarios require multiple choices and should have an impact on one's reputation with at least one campus group depending on the player's decisions. While this input model is deliberately reductive, it reflects norms of the genre of visual novels, where significant weight is often given to everyday interactions.