Generative game development is bullshit
DRAFT
In the sense that ChatGPT is Bullshit, meaning unconcerned with the process and unconcerned with the result. That is not to say LLMs in particular can’t be useful: in the hands of experts they are a force multiplier, and in the hands of novices they are genies in lamps. Game development is undergoing a bar-lowering on an even larger scale than what Flash brought to the web in the 2000s. This is a good thing! I want more games in the world, from as many new voices as possible. We should all be dismantling gatekeeping in art.
To scope my writing a bit, I want to say that what I write here really applies to hobbyist game development, or those that make games without a financial incentive, rather than the industry at large. The industry has its own problems, many of which existed long before the recent generative tech boom which muddy the picture. The hobbyist scene necessarily feeds the professional one, and there are plenty of people like me with fingers in each pie, but the two environments remain different and separate.
Lowering the bar to entry means bringing new people into the game development community, new people that haven’t learned the ropes like the rest of us, and in this case new people that possibly never will. Pre-LLM game developers fundamentally had to engage with the wider community and study hard to get anywhere, because there simply was no other way to build a game than to learn to write code yourself or to team up with someone that did. I think this had a moderating effect on newcomers, where over time they would get accustomed to the culture. Now that coding agents can write the code for you, as well as handle the engine and image models can make the graphics, that route in is now optional, and we have lost that necessary moderation.
I think the lack of that moderation is now driving a wedge through the hobbyist scene, which impacts both the LLM-free (I’ll refer to them as ‘traditional’) developers and the new crop of LLM-assisted (‘vibe’) developers. Many people are firmly planting themselves in one camp or the other, and encountering echo chambers which push them further to one side or the other. Online spaces like development Discord servers or subreddits become vocally in support of one side or the other, and flame those that disagree. This necessarily drives people with more moderate opinions to engage less, which fuels the vicious cycle of taking ever more extreme positions. This polarisation can’t be healthy for the artistic community, and isn’t going to benefit anyone except owners of platforms with engagement algorithms. I don’t know how we consolidate this, but I think it’s important to try.
Among the newcomers, I see several attitudes. Some are genuinely interested in how game development has worked historically, and use their tools to learn about games technology and techniques. I find it hard to see a problem with this: they want to learn, they want to do things right, and they’re trying to take the fastest route to get there. They’re using generative tools for the same reason we’ve always used libraries, game engines and templates; let them. As with anything, attitudes will be on a spectrum and I find it hard to accept any kind of hard, blanket position on how things ought to be done. Computers are complicated, and it’s easy to be at risk of making a real programmers argument. If you use AI to debug something once, is that over the line? If you scribbled some graphs on a napkin, uploaded a photo thereof to your model of choice and had it write the shader, are you less creative for not knowing GLSL? I don’t wholly agree with the position that LLMs are the next step on the ladder of high-level languages, but I have to admit it holds some water. I don’t think I am less creative for not implementing a function by hand that I’ve implemented a dozen times before over the last decade. I think it’s probably up to the individual to draw their own line on this, but we should be able to collectively accept that the acceptable amount of use of generated content in a game can be north of zero without disqualifying it as a game, or disqualifying its creator.
However, there is a contingent among the newcomers that simply do not care. These are the people for whom it’s them and Claude against the world, for whom the old ways are an embarrassment and a relic. These are the people who are proud to have never looked at the code, who generate game design documents from a couple of sentences and brag about their token burn rate. They hand off every decision to the model, but nonetheless post the output to traditional game development communities and react negatively when their slop is identified as such. Barely in the foothills of Mt. Stupid, these people are also likely to think that “coding is solved” and that personal models tuned to individual tastes will be the future of gaming. This is generative bullshit, and it is dangerous.
Bullshitting game development in this sense is dangerous because it is masquerading as game development without being anything of the sort. It produces, as with all generative tools, plausible results. The output is something that could plausibly be a game developed by a game developer, without actually having gone through any kind of artistic process. This does the same kind of damage to a community as sharing generated writing in a fiction community, or generated images in one for digital painting. In any artwork made by humans, even a toddler’s finger painting, the resulting artifact is a consequence of a series of decisions made by the artist, to evoke a feeling or communicate some meaning to the audience. This is not the case with generated art, because all of those decisions are deferred to a biased statistical model of the training data. The volume and scale at which such output can be produced eclipses that of traditional games, so if traditional spaces don’t take a hard stance against this material then they risk being obliterated under a tsunami of generic content by people that aren’t even aware of the existing issues they’re exacerbating. This necessary hard stance hardens the opinions of the anti-generation group, the vibe developers perceive this as hate, and the cycle deepens.
The positions of the companies producing and marketing these products is, of course, not helping. Unity in particular recently announced Unity Spark. In the video two collaborators build the game one of them “always wanted to make”. Since the collaborators are presumably not already professional developers, they are hobbyists. The resulting game after a punchily-edited build montage is one of the most generic, cheap-looking slop games I’ve ever seen. In principle, sticking to my line about lowering the barrier to entry, I don’t think there’s a problem with a very high-level collaborative tool for non-technical users to throw together a prototype, or for example a parent to make a customised game with their child. There are some great use-cases for generative tooling in building games, but although Spark’s tagline starts with “Build it”, it continues with “Perfect it” and “Publish it”. I don’t know a single game developer that would look at this trailer and consider the game shown publishable, let alone perfect. Unity know this too, so we have to ask: who they are actually targeting with this tool? Who is going to build their “dream game” out of a handful of off the shelf assets on a closed platform? I could name a few candidates, but among them are probably not people with an interest in how anything actually works.
So, what do we do about the bullshit? I don’t really know. I can understand the exclusionary stance taken by traditional spaces, they don’t really have a choice, they feel like they’re under siege and there’s an element of truth to it. Maggie Appleton foresaw the dark forest we’re now living in, and ends her article with several good suggestions for how to navigate the slopocalypse we’re now living in. If you’re in a position to call out bullshit when you see it, maybe do? Although it risks deepening the cycle for that person, letting it slide probably deepens it too. If you’re in traditional spaces, I think nuanced positions are likely to be healthier for the space at large than blanket rejection of generative tools. You never know who watches what you say online, and it would be sad to alienate the newbie lurker who wants to learn but feels pre-emptively excluded because they generated some wobbly pixel art. Having the conversation is better than cutting it off.