Comment & Opinion

AI promises to make your game for you. But there’s a catch.

Procedural generation is nothing new in games, but can 'Generational AI' actually make a difference?

AI promises to make your game for you. But there’s a catch.

As AI increasingly surfaces in our lives, more and more people are pitching in to offer their thoughts on the matter. On one hand there's the evangelists for whom this is rapidly approaching a zero-sum game, where big promises are and expectations carved in stone. Whilst on the other there are sceptics, with questions unanswered and serious concerns to raise.

Now comes the advent of AI into video game production – including the concept of ‘generative AI’ being discussed heavily such as in this article by James Gwertzman and Jack Soslow – where entire portions of a video game can apparently be generated by artificial intelligence. So what are the issues and are there any advantages to this technology?

The benefits

As always the argument is that generative AI can dramatically decrease the amount of human time and work required whilst maintaining the same level of quality. The assumption being that the art or assets produced by an AI are just as useful and of the same quality as those made in the more traditional way.

Putting prompts into an AI engine is after all, far less time-consuming than design meetings, concepts and drawing, surely? After all, they might argue, the most important work is behind the scenes, with the game's workings and mechanical functions, not in what the audience actually sees, hears and plays.

The harms

Many AIs are trained specifically on publicly available information. There’s a reason why ‘trending on artstation’ is a key part of many prompts after all. This has raised the ire of many, who both dislike the notion of their hard work being used without permission (and mixed carelessly with other art) and the notion that the AI-positive individuals are on a quest to dismiss their talent and make them redundant.

Thus AI has seen backlash from many corners as much as it has seen proponents. For many creators, in its current incarnation at least, AI art-generation is repulsive both philosophically and ethically.

I, Robot

AI generation is arguably a more benevolent use of the technology. Rather than replace the artist wholesale, it seeks to aid, saving time on creating variations of different assets and allowing time-consuming and not particularly creative tasks to be more easily accomplished.

It’s worth noting that it actually resembles a much older technology that's been in use in games for decades: procedural generation. A good example, perhaps most famously, being the worlds of Minecraft which use the technique to generate random world ‘seeds’.

And Gwertzman and Soslow point to a great example. UE5 offers procedural tools for assets such as foliage. So when your audience demands gorgeous shadows on every branch, your artists can go and work on something more stimulating and expressive instead.

Guiding the machine

Procedural generation is not perfect of course, and, in fact, still requires design work. Minecraft includes numerous examples of hand-crafted features of terrain, villages, fortresses etc, that are made by the game’s team in a more traditional way. In fact, this year saw the Caves & Cliffs Pt. 2 update, touted as being a major overhaul of said procedural generation, being released.

The fact is that procedural generation and generative AI as we now know them, are limited in application. Given free rein they can make entirely incoherent layouts in traditional game design and thus are most often used producing random, simple level designs and details in order to keep things feeling fresh (and save the artist from repetitive strain of arm and brain).

Thus we are still very, very far from this sort of AI ever being capable of producing anything outside of elements for a game. In Gwertzman and Soslow’s article however they paint a comparison between Red Dead Redemption 2 and Microsoft Flight Simulator. They argue how generative AI can ‘create’ a game, in a similar way to how Microsoft partnered with blackshark.ai to produce the map for their flight simulator. But they miss a key difference in their genre and the way AI is used.

Microsoft Flight Simulator has always been well-known for utilising maps, satellite imagery and other data to construct huge swathes of terrain in order to build a realistic flight simulation world. However, on closer inspection these models rarely stand up to scrutiny. Yes, the current iteration of Flight simulator is perhaps its most impressive yet, but it hardly holds a candle to the hand-crafted nature of a world like Red Dead Redemption 2.

An AI may theoretically be able to generate numerically the same amount of assets and voice-lines (and this is a very big maybe) but it won’t be coherent and it certainly won’t be able to ship ‘as is’. The idea of an AI creating an entire game is, at this point, complete science-fiction.

Which raises the main problem with procedural generation, And that’s that it requires tinkering, and attention and constant testing in order to ensure that the generations are usable and coherent. It is not a limitless technology either, as in Minecraft’s most famous example of the ‘far lands’ where terrain generation completely breaks down into incoherent random terrain, can attest to.

Yes, AI can fill in a world map and make it look believable from the air. But having it construct characters, a world, a story, music and more all out of thin air is not about to happen any time soon.

Fixing the problems

To give credit where credit is due, Gwertzman and Soslow point to the most significant issue perhaps facing any AI generation technology at the moment: The legal ramifications. The fact is that while AI now benefits from ‘fair use’ this can only go so far. When ‘training’ moves from relatively helpless smaller artists to bigger corporations and companies, a crackdown on what’s it’s plucking from the internet and social media will be inevitable. As big as the promises may be, AI is reliant on human input, so when this is removed so is the likelihood of something useable being produced.

Not only that, but consider AI technology itself. At the moment many of these engines are free or relatively cheap to use, but if (or when) they become standard, their prices are likely to rise dramatically. It makes little sense for the builders of these engines to not want to monetise them, pricing out smaller developers that would likely be the ones who could benefit from the technology the most.

Ultimately, the stumbling blocks are far too many and it’s far too early to see where this technology might go, or even if it’ll be accepted by the industry and public.

AI in mobile development

And the world of mobile gaming has it's own specific issues with the technology. Mobile games are perhaps infamous for the amount of clones and derivative games that are put onto the marketplace. The best mobile games are the ones which defy this convention. Angry Birds for example, innovated and worked with the limitations and advantages a smartphone platform offered it. For a truly astounding game to stand out from the crowd it needs to not only have easily copied aspects such as gameplay features, but to have something unique. Iconic characters and art are not created by AI, but by artists and designers.

How do you ask an AI to tweak a model’s hair-style? Or alter other aspects? Do you go back to step one and generate a new image? In many cases, the time and effort it takes would be better spent working alongside a team of creatives who understand what the next best step would be and always have an eye on producing something workable.

The bigger issues, and smaller ones too

While Gwertzman and Soslow do point out the problems: Keeping a coherent style or even a similar design between generations… The issue of ‘training’ AI on samples and more… One particular 'bright side' they touch upon seems a step too far – the notion that AI prompt writing may become a marketable skill.

The fact that ‘anyone can do it’, is one of the main selling points of AI. To promise the dawn of the ‘skilled prompt creator’ smacks of finding one ray of creative sunshine in what otherwise appears to be a road to nowhere.

Meanwhile the claim is that coding or other mechanical aspects will not be affected. It seems that the main proponents of AI are not those in creative industries, insisting that everything they cannot do, can be done by AI, but everything they can do, cannot.

That’s not to critique everyone who is a proponent of AI technology of course, and Gwertzman and Soslow certainly don’t imply that artists are redundant. In many cases tools to place random effects, brushes in programs like CSP or Photoshop and more have all been received well by artists and designers whilst utilising tech similar to AI. But it’s important to recognise that such advancements support the traditional manner of development, allowing enough space for the expression and innovation which are crucial to game design.

Meanwhile, 2021's Grand Theft Auto: The Trilogy attempt at almost solely utilising AI upscaled textures was criticised heavily for its attempts to ‘improve’ the graphics of the original and instead fell flat compared to those available nearly 20 years earlier. The time saved didn’t magically result in an amazing product.

The conclusion?

AI has its limitations. Always destined to re-plough furrows already created, and reliant on being spoon fed content in order to churn out similar generations. Making games ‘in the style of’ can be a beneficial, artistic and financially sound decision of course. Many games on mobile and beyond recapture the feeling of playing long-dead franchises or niche genres that many other studios have left behind. But the fact is that AI relies on doing what has come before, and that puts studios that rely on it potentialy entering a creative dead-end.

However, are we close to the day when we’ll be able to type in a prompt and have a whole AAA game spat out for us? Not even close. So perhaps it’s time to think about how AI technology can support and not replace.


Staff Writer

Iwan is a Cardiff-based freelance writer, who joined the Pocket Gamer Biz site fresh-faced from University before moving to the Pocketgamer.com editorial team in November of 2023.