No industry has been left untouched by the looming presence of artificial intelligence – and gaming is no exception. But the rise of generative AI tools like ChatGPT and Midjourney have somewhat obfuscated the fact that AI has been influencing development in one way or another since at least the 1980s.

There is a broad consensus that AI is “new” to the space, says Tommy Thompson, director of consultancy firm AI and Games, speaking on a panel as part of our new GI Sprint series about how to make games faster and cheaper. But the idea that AI is a new innovation, and that it’s only now beginning to threaten jobs, is far from true.

Thompson, who works with developers on how to integrate AI into their pipelines, has seen this “silent revolution” brewing in the industry since the late 2000s. Back then, developers began incorporating machine learning into their workflows. But AI is a field of computer science that stretches back to the 1950s, and the games industry has been a beneficiary for decades.

Sean Cooper, technical director at Didimo, explains: “I can think of countless automations and AI that we’ve used since the ’80s. And now we’re starting to talk about it, and starting to worry about it. But it’s always been there. And it puts people out of jobs. Things like, we don’t need a tester to test the game, or we need less people to test the game in order to get to [our] goal.”

Cooper, whose company builds tools to automate character creation and populate NPCs in game worlds, adds that he can’t see why people are so worried right now – because AI and automation has always been a fixture, in his experience. He says that it’s always allowed developers to be more efficient, and has also been a boon for studios that lack the financial firepower to hire staff to raise the quality of their output.

Thompson adds that the use of sophisticated AI tools has traditionally happened behind closed doors in AAA spaces. So these systems wouldn’t have previously been widely publicised as they would have been considered mundane everyday tooling.

It’s only now that AI has spread like wildfire through the public consciousness that those outside these walls are waking up to the potential benefits and risks. How then, in this new age where AI is in vogue, can game developers make the most of it to improve their processes and ultimately build games at a lower cost?

You can watch the full podcast below, download it here, or find it on the podcasting platform of your choice.

Watch on YouTube

Don’t just chase the latest craze

That there’s an AI boom is undeniable, and it’s predictably led to an explosion of tools and services that purport to offer AI in some way, shape, or form. This ‘productisation’ means it may be difficult as a studio to determine which tools are best suited – and whether generative AI is really all it’s cracked up to be.

It would be wise, our panellists agree, to filter away the buzz and try to hone in on the substance.

“The problem is that we continue to have this conversation around AI in video games that is often being led by people who don’t work in games, and are often trying to sell an AI product,” says Thompson. “And that is not only warping the conversation, particularly among the playerbase and the general public, but it’s also warping those perceptions internally within the industry.

“That’s certainly the experience I’m having chatting with a lot of studios. Because they’re hearing the soundbites about GPT and everything else and they’re like ‘Can we do that in our games?’ and I’m like, ‘Well, yeah – but no’.”

Whether the use of some of these tools is even legal, given the copyright concerns, is another issue altogether, he adds.

“There is no one silver bullet that will create your game for you. I would be surprised if we see that in our lifetime”Lucie Migné

Cooper also points out that if you simply add “AI” to a product name, it’s likely to get more investment – even if it’s just simple automation. That’s because people with the purse strings are excited by innovation and tend to hype up new ideas. In turn, that leads to a proliferation of AI products and services that developers may see as cash-spinners. Whether or not these are truly AI products doesn’t matter to some extent.

Meanwhile, although generative AI is leading the conversation, studios may find that adopting services in this space leads to new problems they may not have anticipated – particularly around using these tools effectively – not to mention the potential legal pitfalls of not knowing which data some models were trained on.

Instead of chasing tools that may represent a false promise, developers should instead integrate tried and tested tools like automation or machine learning into workflows. That’s Lucie Migné’s advice, as senior producer at Might Build and Test – an AI and automation studio focused on quality assurance.

“They will probably find, like we have at Keywords [Might Build and Test’s parent company] recently, that the tools as they stand right now do not necessarily live up to their complete promise,” she says. “And the cost of learning that might be high for a small team.”

Lean into open source

As the breadth of AI tools proliferates, how can studios without the resources to build in-house tooling from scratch get a slice of the action? Developers – especially indies – may want to tap into the ripe pool of open source as a starting point, says Cooper.

“If you look at indie games, they’re small budgets, and if you bring in AI systems – especially free open source stuff – their quality bar is going to get substantially raised, because the missing people that they cannot hire are now filled in by these generative AI systems,” he explains.

But failing to lay the groundwork for the adoption of new tools or training staff to properly use them will lead to unintended consequences, adds Migné. Many in the open source community also warn that the total cost of open source isn’t zero – despite the fact no money changes hands. IBM, for example, explains that running costs need to be considered, while maintaining open source software is often a hard endeavour that requires volunteers to dedicate their time to, say, fixing bugs or adding new features.

“We’ve got to keep in mind that the general perception of [generative] AI tools in the public eye is that [they] are free to use – and it’s certainly not the case,” says Migné. “It might also skew the perception in the industry that because they are free, widely accessible, easy to use, that they are a bit of a panacea for all of their production problems – their missing people, their missing skillsets… pretty much like a fast-forward on experience.”

Identify the best workloads to automate

It’s difficult to know where to start when automating workflows, says Lauren Maslen, director of production at Mighty Build and Test, because many parts of the industry appear ripe for automation. But this may not work out in practice.

Some good candidates are any areas in which you have traditionally generated content, she says, highlighting NPCs and characters, instances where you have plenty of duplication in environments or even in coding and engineering. For her company, however, a lot of what they do is look at the QA process.

“If you’re looking at any game of any scale, then you’re dealing with sheets and sheets of test cases. You’re relying on exact execution of these for a distributed team reporting up the chain to make sure regressions are caught early – all of those sorts of things. And that again is an area where you’ve got this repetition of work, [which] can be a place for automation or AI to fit neatly into the puzzle.”

“The problem is that this conversation around AI in video games is often being led by people who don’t work in games, and are trying to sell an AI product”Tommy Thompson

On that note, Thompson adds it’s important to identify early on exactly which parts of your workflows are being wasted on the workforce – “taking a point to stop and think about the parts of your process that are quite mundane and monotonous.”

“So when we talk about QA, for example, the QA tester has to run into the door 500 times at 700 different angles to make sure you can’t run through the door – automate that,” he says. “But things like, ‘I need someone to go through a quality pass through the level once every couple of days to make sure the UVs are correct, and the textures are displaying correctly in the game and all the visual fidelity’s there’ – that’s something we can do with image recognition, certainly, but I think we can often do it much faster using a human.”

Know how far to go and which tools to use

When undergoing your journey into adopting AI tools or introducing elements of automation, it’s critical to find the right tools for the right purpose – and to do your homework before investing the time and resources.

“Not all AI or automation tools out there are useful for the same usage,” warns Migné. “If you’re trying to make a bot bang into a door, you’re not going to use that tool to create a repetitive meeting agenda. Identify those areas that you do want to automate, and then do your research into what’s available there – and use the right tool for the right purpose.

“There is no one silver bullet out there that will create your game for you. It does not exist yet. I would be surprised if we see that in our lifetime.”

“Instead of blindly jumping onto a bandwagon, the sector as a whole need to be intentional as to what gets automated”Aleena China

Dr Aleena Chia, lecturer at Goldsmiths, University of London studies game developers and how they automate processes. She says that different members across the team may see automation through a different lens and that, ultimately, it’s important to know how far you should go – and when you’re going too far. She uses the example of generating individual trees versus creating a forest environment; AI isn’t capable of big-picture thinking or conceptualisation, and you need to be careful to ensure these elements of content generation aren’t outsourced to machines.

“So I think that instead of blindly or just jumping onto a bandwagon, managers and studios – and the sector as a whole – need to be intentional as to what gets automated.”

Don’t forget that gaming is an art form

Ultimately, studios shouldn’t lose sight of their signature style or USP as they’re leaning into automation and AI tools.

Chia says that when deciding what to automate, it’s a trap to think purely in terms of how much more efficiently a machine can perform a task. Instead, studios must take a big picture approach and think about how each discipline – from narrative designers to artists and programmers – may consider automation.

What’s going to happen, Chia continues, is there’s going to be a moving target as to what is deemed creative enough to not be automated, and what is “meat and potatoes”. As a whole, how we appreciate games and what consumers look for will also change with time and as the overall picture of automation moves forward – especially if gamers value knowing their game offers a human touch.

“I don’t think that we can put all the pieces in place right now, but certainly we can have an idea of what games will look like at different levels,” Chia says, signalling that the industry will only see more automation and AI adoption in the near future. How far this goes, however, remains to be seen.

“I was at GDC and at at roundtable an artist came and said, ‘Actually, texturing is not as mindless as you might imagine, there are a lot of aspects of texturing that tell the story as well’,” she adds. “So while there is consensus [that you can automate] things like running into a door or game testing, when it comes to more nuance of creative processes, I don’t think we are yet seeing a consensus as to what should and shouldn’t be automated.”



Leave a Reply

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