Generative game development is bullshit
In the sense that ChatGPT is Bullshit, meaning unconcerned with either the process or 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. That said, as with Flash, this technological revolution brings with it a pile of baggage.
What I write here applies to the scope of hobbyist game development, or those making games without financial incentive, rather than the professional industry. The hobbyist scene necessarily feeds the professional one, and there are plenty of people like me with fingers in both pies, but the two environments remain different and separate.
I also, as much as possible, want to avoid digging into the ethics of the problem. Many traditional game developers believe that use of these tools should be forbidden on ethical grounds, while a lot of vibe developers assert that said ethical issues don’t actually exist, because humans also learn by looking at prior art. I don’t really agree with either of those positions, but I’m also not going to present a happy medium where none exists. The ethics of generative tools built on large datasets is the topic for entire fields of research, not one essay.
It’s a lot like Flash
Flash lowered the bar, and reduced moderation coming from mainstream publishers, but also the game development community at large. Fringe or taboo topics don’t easily attract collaborators, so the number of games on said topics was low. While it was possible to create edgy, offensive, unpleasant games before, the cost to do so and the necessary onboarding into more moderate communities acted as a brake, so that by the time one had the capability to make something untoward, they had likely been dissuaded by the time they were looking at distributing to an audience.
By reducing the time, energy and knowledge requirements to publishing, we opened the gate for less conscientious games to reach large audiences. This is the flipside of hearing from new voices; some of those voices are necessarily going to be bad. LLMs and other generative tools do this again, where now not even a passing interest in computers is required to build games. On the one hand, I think it’s great to be able to make games without a programmer in the loop. On the other, game creation being so easy means that a lot of unpleasant stuff from people acting unthoughtfully is coming.
Reversion to mean
Some of the newcomers 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 a chatbot 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? If you search something on Google and read the generated summary, should you have closed your eyes instead?
However, there is a contingent among the newcomers that simply do not care. These are the people for whom it’s them and “their” 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.
Many reclaim the term “slop”, valorising quantity of output over everything else. 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.
Generating games like this 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 creative process. 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 process is the art, and contains as much or more value than the resulting artifact.
A game made traditionally might fall right in the middle of any statistical model of games, but this is as unlikely as the existence of a true median voter: there are too many axes. Really, we should expect any thoughtfully made game to exhibit something outside of the norm, which is valuable for communicating something novel. Generated games don’t have this quality, being fundamentally derived from the work of others. They reinforce whatever status quo there is, in blithe unawareness, just to feel like they made something.
Intent
So, what do we do about the bullshit? Maggie Appleton foresaw the dark forest we’re now living in, and ends her article with several good suggestions for navigating the slopocalypse we’re now living in. The rest of her writing is good too. I think we should call out slop when we see it, and have conversations about where value actually comes from in art. The companies producing these tools seem to be more or less all-in on commercialisation, evident in the recent trailer for Unity Spark proposing that we should create generic, sloppified, “perfect” versions of the games we “always wanted to make” and publish them to their platforms.
In Games: Agency as Art, Nguyen makes the case for games as “artifacts […] intentionally constructed for engendering […] aesthetic experiences in their audiences.” Well, the artifact is almost free now, so it’s the intentional construction which carries the value. A common refrain among vibe developers is that the value of everything now drops to zero because products can be reverse-engineered as just a function of LLM credits. It’s getting harder to argue against the technical position of the argument, but the gulf between building something because you can versus something because you care is a world of difference, and maybe that’s all there needs to be.