Barriers Tumbling: AI and the Future of Game Development

My stepfather is one of the smartest people I know – an engineer and a mathematician, a polyglot descendent of a Scottish poet, and a whiz with computers long before they were in every household. I’ve written before about how he first introduced me to games; to no small extent, I owe my career to him.

When I started to talk about OpenAI’s ChatGPT, he was skeptical and made it his mission to see how fast he could stump it. It didn’t take long. He grilled it on an Irish copyright dispute from the sixth century; in its responses, ChatGPT provided quite a bit of inaccurate information while stating it as factual.

My stepdad’s criticisms are valid. ChatGPT, despite the buzz around it, has plenty of flaws. By now, the caveats are well-known – if asked questions that require factual analysis, biases in the data set it was trained on lead it to make authoritative statements that can be misleading or even entirely untrue.

Much like information technology that’s come before it – searching Google for answers, or quick consultations of sometimes-misleading Wikipedia pages – there’s the potential for users to make bad decisions if a user doesn’t approach the technology with a healthy degree of skepticism.

That said, in the two short months since BonusXP shut down and I started to make active use of AI tools, several have proven to be invaluable in my work. Game developers, particularly indie or solo developers who ignore the competitive advantages of leveraging AI for the prototype phase of their projects, are in for a rude awakening over the next few years.

Comedy Value

When I first tinkered with ChatGPT, my motivations were half curiosity, and half wanting to see what ridiculous things I could get it to say. For some reason, I fell down the rabbit hole of asking it to generate a wide variety of reviews for classic movies or books as if Gordon Ramsay, the well-known TV personality chef, was the writer.

Not only could ChatGPT produce completely plausible reviews, in a style that mirrored Ramsay’s signature grumpy affectations, but it could tweak the reviews based on subsequent prompts – “Make the review sound more like Gordon,” “Make this review more negative,” etc. I could give it feedback on the elements I didn’t like, and it would adjust on the fly. 

At some point – again, purely for laughs – I typed in “Give me a level idea for the game Orcs Must Die, but with a Gordon Ramsay theme.” ChatGPT produced a wall of text, almost a full design document – an overview of the idea, then a breakdown of the details, including setting, custom traps, special enemies, and unique mechanics.

This exchange, and the subsequent refinement of the idea through additional prompts, were revelatory for me. Though the initial output had all kinds of problems, when viewed as the product of a junior designer’s first draft write-up with the original marching orders, it was more than reasonable. 

“Gordon Ramsay” + “Orcs Must Die!” = GENIUS. The game you never knew you wanted!

Good game developers iterate; the first ideas are never the ones we keep. Good designers also know that suggesting initially ridiculous ideas is part of the “secret sauce” for making great prototypes – freeing your mind from the constraints of existing games and the demands of the schedule.

While I knew to some extent the output was just a party trick – ChatGPT mixing everything in its data set about Orcs Must Die and Gordon Ramsay in a blender and vomiting out some words – it wasn’t unlike the process of brainstorming meetings.

A Nonjudgemental Coding Buddy

Before my time at BonusXP, when I was first consulting for a couple of years, prototyping my own solo game was slow going. I’m reasonably technical for a producer, but I’m no programmer – and I’m definitely not an artist.

When I’d get stuck on a tricky bug, I’d often burn several hours figuring out the problem. Whether it was some UI sorting issue in Unity, or a bit of C# script giving me fits, I knew I was in for either an hour watching video tutorials or diving down the rabbit holes of forum posts looking for the one nugget of information I needed.

In my new effort at version two of my prototype, using AI tools has shaved all those hours down to minutes. The data set for ChatGPT may be thoroughly inadequate for obscure questions about sixth-century copyright disputes, but it’s proven to be stellar for Unity and general C# questions.

Asking ChatGPT very specific questions about bugs and problems immediately produces results – and often produces code as well. As long as the scope of my questions is narrow enough, I can say “I need a C# routine for this” and eight times out of ten, it produces a useable bit of code. In the rare times when it hasn’t worked, I can describe the errors I encountered and ChatGPT is usually able to revise the code snippets and fix its own bugs.

On the art side, when I was working on my original prototype I paid for concept art and a few ingame pieces on contract, but I’d often just grab a free asset off the Unity asset store as a stand-in or clumsily cobble together my own. 

For finished pieces, working with an artist was fantastic. For rapid iteration on a prototype, it was hard to get exactly what I wanted without substantial back-and-forth discussion (which cost money and time for both the artists and me).

Though in the second version of the prototype, I’ve leveraged AI tools for both prototype art and music, the results are of more questionable value than ChatGPT helping fix a bug. I don’t feel my prompts have produced anything I’d consider shippable content.

That said, using AI art tools like Dall-E 2 and Midjourney, it’s been ridiculously fast to get an asset that’s representational in a game-ready format for a prototype. It also gives me a great starting point to talk to a human artist about exactly what I need once the game is funded and I’m ready to make real assets – “I want this, but more like this, and with this changed, in this format.”

A prototype image – the peak of Mount Everest. Not remotely a shippable game asset, but good enough for what I needed at the time. I grabbed a screenshot from Google Earth (so this couldn’t ship anyway for legal reasons), then ran it through an AI “oil painting” filter.

The Ethics of It All

In the end, I’ll replace all of the prototype AI-generated art with work done by a human artist – first, because I’ll get something fundamentally better, and secondly, because there are a ton of legal, ethical, marketability, and copyright issues around using AI art in a shipped project.

Many artists I know are justifiably concerned about how AI will affect their future job prospects. I share those concerns – as a former game designer, a chill went down my spine when ChatGPT produced a viable, albeit silly, design document. And the worry that actors are about to be replaced by AI versions of themselves is a core driver behind the SAG-AFTRA strike of 2023.

For the time being, I think the job prospects of game industry artists are safe. Early experiments with purely AI-generated games, while impressive, still suggest there’s a need for human guidance, iteration, and polish for a game to come together coherently – even if it’s a smart human artist using the right prompts in an AI tool for their first draft work.

That said, the legal and ethical issues around the use of AI assets are evolving. The base technology is improving in leaps and bounds with each iteration, more rapidly than anyone expected.

The Coming Glut of Games

Over the last decade, the number of games released daily on Steam has been steadily rising. It’s easy to understand why; the barrier to entry in game development has dropped.

When I first started in the industry, making and releasing games required specialized, technical knowledge. Today, any enterprising high school or college student can teach herself Unity, leverage the asset store, and (if motivated enough) release a game.

The rise of AI tools is going to accelerate this trend, lowering the barrier to entry to game development even further. The first games that leverage AI in one form or another have already been released.

Overall, I welcome this trend toward development democratization. I’m excited to see what greater diversity brings to the game industry, and I love indie games as much, if not more, than the latest triple-A releases.

However…

If you’re a developer, and you’re not thinking about how to use AI tools to help accelerate your process and make your game better – watch out. Because everyone else, from the big studios to the one-person garage developer, will be leveraging the heck out of it.

We are never late with the lamb sauce! New blog content appears on ScreeGames.com every Tuesday. Follow us on Facebook or LinkedIn.

Leave a Comment

Your email address will not be published. Required fields are marked *