P2: AI Judgment Day

Final Game:

https://ai-judgment-day-game.vercel.app/

Overview

In the world of AI Judgment Day, a rogue AI (“Sydney”) has taken over the world. Sydney is now judging each member of humanity to determine whether they should be spared or purged. As one of the last humans on earth, the player must convince Sydney why they should be spared — but Sydney is a busy AI, so they only have two minutes to convince her!

History Versions of Game:

The purpose throughout the proofs of concepts is to create a game which beckons players to consider the role of technology on society and consider the morality / purpose of their past in the context of an uncertain future. I also wanted to create an eerie sense of what it feels like to be judged by a human-created program.

Iteration 1: Terminal Proof of Concept

The first iteration was to test whether GPT could judge the quality of a player’s responses, and roleplay a rogue malevolent AI. It successfully fulfilled these objectives. However, it turned out that talking to a rogue AI for unbounded time turned out to be an unengaging and slow experience, so I realized I would have to make serious revisions to this idea.

Iteration 2: Attempting an extended parser fiction

I attempted to completely change the idea to a parser fiction with a storyline. The player is in a mind-controlling dystopian future, and their objective is to lead a rebellion against the corporation making the brain-computer interfaces. However, I noticed in playtests that this prototype would often end up in narrative “doom loops”, where the options provided to the player would be uninteresting and granular. I shelved the idea.

Iteration 3: Original idea, with a time limit

I added a time limit and scoring system, where players only have 2 minutes to pass by the threshold necessary to be judged “worthy of living”. In playtests, I got a) excited and visceral reactions from the premise of this game, and b) a lot of enjoyment watching folks play! The accelerated cadence made the game more interesting (and frantic). It became in some ways a de facto typing speed test, but folks said that they preferred it that way (rather than a constraining mechanic based on limiting the number of possible turns).

Iteration 4-6: Building a frontend, deploying this into a webapp

I used NextJs and Vercel to build this as a webapp (my rationale is that it would be easier to share). I used the Chakra framework for UI, repurposing some of its slider components to work as timers. It wasn’t straightforward or easy to build a whole UI and chat interface from scratch, but I am proud of the result (hopefully now mostly free of bugs!)

I built in a slider component and a timer sliding bar component to visualize the player’s status and add to the sense of urgency that the time is running out. In playtests, users liked this a lot more! It was clearer (especially with the intro screen) what their current score was in the game, the objective, and the strategies to win.

I believe this iteration most successfully achieves the objectives previously mentioned. The time constraint forces quicker, more intense reflection. The welcome screen, with its concise writing, is a type of in medias res approach which immediately places the player in this dystopian scenario. The sudden and lurching aspect of this design, I’ve found in playtests, actually prompts more contemplation. After playtests, folks would ask “is this really a scenario that can happen?” or, “wow, I’ve never questioned what I’m doing as much as needing to explain myself in front of an AI.” I’m also proud of the visual affordances for scoring and timing, as they communicate the player’s state more clearly and add to the sense of urgency/excitement.

Photos/Video Clips of Game Testing

Player 1 is a 23-year old woman. Background in HCI, now adult professional in the Bay Area in tech. She first tried playing by summarizing their life story. With the AI finding that unsatisfactory, she tried to “virtue signal” their way into success by choosing occupations with positive social implications. Along with fixing a couple bugs, I ended up changing the prompt to penalize excessive lying such that players could not use this to game the system.

Player 2: 21 year old man with game design background

Seamus’s playtest was super helpful! He tried a variety of approaches, comprehensively stretching every mechanic to its limit within the game. He tried to engage the bot in a moral debate. The bot would sometimes be too verbose. From his playtest, I ended up changing the response behavior so that the bot would be more concise.

Playtest3: early 20s woman, game design + GSE master’s student

This playtest was helpful for testing multilingual capabilities! It turns out the bot is capable of understanding multiple languages, and will respond in kind. Surprisingly, certain languages seem to get consistently scored higher by the bot (such as Italian) when players respond in it, which is perhaps an indicator of bias by the bot.

Playtests 4-10: with various folks on my dorm. Computer science, human biology, education, etc. backgrounds

A lot of reactions to this. Folks played on their own devices, and were intrigued by the premise of the game. Those from more humanities-facing backgrounds felt unfairly judged by the bot, which seems to prefer engineering-type backgrounds that have a direct socio-societal benefit. They also found many bugs (including with the timer, and text overflows), which I promptly patched. This was a blast. Also, players really liked the replayability — trying different strategies — and to enhance this, I added a “reroll” button which allows you to play as a given invented character in future playthroughs.

Reflection

Taking a narratively-based game from start to finish, with a strong plot premise, was exciting! I learned so much about crafting a narrative and about using AI to craft stories like this. For the narrative, I learned that less is more. I initially started with a super involved narrative and painstaking worldbuilding (I even wrote a short story to accompany it) to try to immerse the player in the world. However, some players felt constrained by the narrative. Since my objective was to be as thought provoking as possible, I realized that by paring down the game I could prompt more reflection. I think the final game at the minimum engenders a spirited discussion / reflection! I will reflect more on the process of building this game in my reflection tomorrow!