I assigned my game design students to brainstorm with Chat GPT. Here’s what happened.
I have been teaching Foundations of Game Design at Northeastern University since 2016.
Up until this year, I’ve taught this in-class exercise in my Foundations of Game Design class: Brainstorm 100 ideas as fast as possible.
Given 45 minutes or so and the directive that only quantity mattered, groups of three-to-four students have constantly been able to brainstorm upwards of 100 ideas with relative ease. The lesson in the lesson is that ideas are easy to come by. It’s the next stages: vetting them, building on them, and finally executing them that takes real skill.
But what happens in a world where you can generate 100 game ideas not in 45 minutes, but in 10 seconds? Does the lesson still have value?
This year, instead of making them struggle to come up with their own ideas, I asked my students to experiment with different prompts on ChatGPT to get to 100 concepts quickly, and then spend the rest of class sorting through the content to pull out actionable concepts.
Here are my takeaways.
Most of what GPT-3 generates is bad. That’s good.
This should probably go without saying, but the vast majority of the ideas generated by the AI were poor. However, that’s not a problem, it’s a boon for the purposes of my lesson. First of all, most of the 100 ideas my students were formerly rapidly generating the hard way were also poor, unoriginal, or impossible to produce. Think, “What if Mario was a dog?”
So not much of a change on that front. The point of the exercise was always to generate a shotgun spatter of mostly bad ideas.
It’s probably worth talking about exactly how the ideas the machine spat out were “bad.” As a rule, they were derivative. “A board game where you play as the leader of one of several alien races in a fight to conquer a galaxy” was one of them.
This is so basic and derivative that it’s hard to even call it an idea. It’s more of a description of a genre of games that already exists and has for decades. It’s not that this type of game couldn’t be made, or couldn’t be fun. It’s just that it’s… nothing. If your creative director asked you for the next great game idea and that’s all you brought them, you’d be fired.
It thinks in words, not mechanisms
This surfaced a problem with GPT in general when it comes to game design. One would think that a super smart computer program would excel at dreaming up complex systems with interlocking parts. But Chat GPT is a Large Language Model (LLM) and was trained on billions and billions of words. Therefore, it also thinks in words. This is why when asked to generate 100s of ideas, it spat out thematic ideas like “A puzzle game using quantum time” and “A game about negotiating peace” and not “A third person 3d-game where the player must roll up all environmental objects into a ball that can attract larger objects the larger the ball becomes, because of its gravity.”
The above mentioned game of galactic intrigue is a passable, if spare, theme. But there is absolutely no indication of how the game is played. And when interrogated, GPT does not supply novel or even sensical mechanisms.
It vaporizes the blank page
However, let’s say you liked the idea of this war in the stars and wanted to flesh it out. Taking this as a staring point gives you much better purchase than starting with nothing at all. This is where human brainstorming really begins: taking a kernel and building on it.
Yes, you can ask Chat GPT to ideate on an idea further, and with great prompt writing you might be able to get it to do some useful work, but at this point it becomes more efficient to do the ideation yourself. The LLM excels at what the human is worst at: conquering a blank page. In fact, it does more than conquer it, it annihilates it, shock-and-awe style. Once this has been accomplished, the humans step in and do what they do best: hone and polish.
Using it feels “cold”
I asked my students if they liked brainstorming with Chat GPT. Many of them praised its speed, but some pointed out a deeper issue: its lack of humanity. They described the process as feeling “cheap” and “joyless.” And said (almost definitely correctly) that they could have come up with better ideas themselves given a little time. What, exactly, is the value of generating 100 bad ideas in seconds and then spending an hour plumbing those depths for something usable, when one could have spent that hour using another brainstorming technique to create something much more personal and novel?
I have to say, I was moved by this perspective and feel that it’s true at its core.
That said, sometimes you just need something to get you past a creative hump, and if a small injection of artificial intelligence gets you there, so be it.
The Ideas
At the end of class, I asked my students to think about the “best” ideas that Chat GPT generated, and to sleep on them. The next day, I asked them to turn the ideas into more fleshed-out concepts. Here are some of these:
From Eleanor L:
Witch Hunt
The idea that I chose was an RPG called “Witch Hunt”, where you play as a witch who must utilize dark magic in order to avoid being discovered captured by witch hunters as well as uncover some conspiracy. The people you talk to and the choices you make when interacting with them would affect which ending you get in order to encourage replayability. Your character would also have a measurement for morality, which would be affected by how much dark magic you choose to use. While using dark magic can make your character more powerful, the more you use it, the lower your morality becomes. I think it would be interesting if after your morality becomes too low, the dark magic begins to corrupt your sanity and changes what choices you can make when interacting with other characters. This gives players the options of either utilizing dark magic to its fullest extent despite its cost, avoiding using magic altogether by using stealth and attempting to blend in as a human, or somewhere in-between.
From Stephen W:
Time Travel Platformer
A puzzle platformer where the player wakes up as a patient a forgotten laboratory with the ability to switch 3 general states of time; past, present, future. Through this mechanic the player will progress through the laboratory while trying to find the exit and learn more about who they are from the things shown in their past and future. In order to progress the player must interact with their past and future times in order to change the puzzles in the present. I also like the ideas of each version has a unique ability so I’m thinking the past has increased strength and health, the present can slow down time for a short duration and the future has a fancy metal arm that can shoot a projectile. or hack into devices.
From Shanie H:
Parallel Universe
The idea I wanted to expand on was called “Parallel Universe”, in which you command multiple armies in various dimensions. Supply lines and chain of command goes through portals to these universes, so you need to be sure not to let creatures through portals or lose access to one. Each universe has a different strategy, and you only have access to a certain amount of “energy” to make moves, so you need to make sure that you’re prioritizing the right armies. Universes have different weather situations, and different armies/creatures, so you might need to allocate more ranged troops to one universe only to find yourself lacking in them somewhere else later on.
From Juhyun K:
Reincarnation
Basically the theme is reincarnation, and you will build up karma depending on what you do and when you die. You get reincarnated but certain interactions will be different depending on the karma points you have making it like a parallel universe. This could be maybe something like if you save an npc it’ll unlock a new patch to a new level and if you kill one of the npc maybe you get recruited to dark side or something and go to a different level making it like a multiple ending puzzle plat-former game
So what do you think of the ideas my students came up with using Chat GPT? Is it a worthwhile brainstorming tool? Would you use it? Have you used it?
Sam Liberty is serious games and gamification consultant and professor of game design at Northeastern University.