The Future of Generative AI in Games Development

29/04/2024

Matthew Gough – April 2024 – Bolton University

Videogames have always been at the forefront of pushing technological advancement (Xun et al., 2022). As such, it was only a matter of days from the public launch of ChatGPT (OpenAI, 2023a) before the use of Generative Artificial Intelligence (GAI) found its way into the Games Development scene. Countless videos of developers using AI to enhance or create their products were uploaded to the content distribution platform YouTube. Indeed, one video from Matt Wolfe (2023) demonstrates an AI only process for building a game which aggregates text-based GAI with Image generation to create something where the only role played by Matt himself is assembling the pieces.

With so much early adoption then, it’s easy to imagine that in a few short years, GAI will be generating most of the content in videogames, that jobs would be dissolved in favour of using AI solutions. The reality of the situation, however, is not so straightforward (Pennefather, 2023). In a recent interview John Riccitiello, CEO of Unity, a prominent name in Games Development, talks about the role of GAI, not in terms of replacing staff, but rather augmenting them, allowing them to work “2 or 3 times faster than before” (Growth Masterminds, 2023). Rather than relying on GAI to solve the problems traditionally solved by designers, programmers, artists, etc., Riccitiello describes a vision for Unity where GAI based tools are baked into the platform, allowing for seamless communication in a creative space, while retaining the core functionality that does not require the complexity of AI intervention (Growth Masterminds).

The tools being alluded to are already being heavily utilised. In programming, it’s a trivial matter to ask ChatGPT to solve simple coding problems, yet already new technologies are emerging that integrate GAI at its core. The new Integrated Development Environment (IDE) Zed (Zed Industries, 2024) boasts Github Copilot (Github Inc., 2024), providing GAI content directly at the point of use. This functionality is also present via extension in an ever-increasing list of IDEs. Clearly, GAI is making an impact in a big way, only a year after hitting the global stage.

the-future-of-generative-ai-auto-complete

Figure 1: Autocomplete has been helping coders for years, and is only gaining in intelligence with tools like GAI (Image copyright: EgoProblems Games, 2024)

This is not a change to shy away from. Due to the exponential rise in popularity of GAI to solve any number of problems presented by the population (Hu, 2023), there is no way to avoid its significant impact even in its infancy. Therefore, the challenge becomes “How do we embrace it effectively?”. Attempts at writing long form applications in ChatGPT are frequently far from perfect, but mundane tasks, such as correcting coding errors, suggesting solutions to common programming problems, and implementing commonly used algorithms are all rendered trivial by this new technology. To be effective though, it is imperative that the user be skilled enough to critically evaluate the generated snippets. Without the base level of skill to appraise generated code, the likelihood of producing something successful and performant is low. In this sense, in situations where GAI is employed in programming, the user takes on the role of editor, becoming responsible for critically appraising the content.

This practice is not dissimilar to the already popular technique of code review, whereby code submitted by one programmer to a codebase is appraised by another coder (Bacchelli and Bird, 2013) often before it is allowed to enter the main repository for the source codebase. By using GAI along with a review process such as this already standard practice, GAI generated code has the potential to increase productivity and reduce the need to perform repetitive mundane tasks, freeing up the programmer for creative thought, developing more complex solutions for the software.

A more controversial use of GAI is in the field of art. Not just limited to text generation, image generation is also a widely used tool growing in popularity. The software Dall-E (OpenAI, 2023b) provides a convenient way to ask for a 2D image in a conversational manner, and it will respond with several generated pictures for the user to employ. This has caused much debate regarding the copyright ownership of the content, and whether the data it is trained on is legally obtained (Feldman, 2023). If such controversies are resolved, the potential for AI generated art in videogames in undeniable. Wrapping textures, Concept art, Simple backdrop pieces, all of these are within the power of these tools. Thankfully for industry professionals, the work produced is rarely game-ready at the point of delivery. For smaller companies that cannot afford artists however, this tool might be just enough to gain them traction and a foothold in this competitive industry. For larger companies, a similar appraisal process and modification will still be required by artists. This once again creates a supervisory role for less intricate tasks, while freeing up creatives to work on the bigger picture or develop complex, critical assets.

the-future-of-generative-ai-tiling-texture

Figure 2: Power and Limitation in one image. The prompt was "draw a tiling texture for a cartoon style brick wall that has multi-coloured bricks that each represent a mood". As you can see, the bricks are colourful and represent moods. However it does not tile and would not be fit for purpose. (Image generated using Microsoft Copilot powered by Dall-E 3)

Why are these larger tasks best left to a human? Riccitiello believes that the difference comes from the human ability to see the bigger picture, and to iterate (Growth Masterminds, 2023). This combined with the limited way in which an AI can assess data – through pixel analysis and text pattern matching and language processing – means that while GAI holds information and can use it as directed, it lacks the ability to contextualise that information and develop ideas according to a grand vision (Growth Masterminds).

So, to conclude, what future does generative AI hold in Game Development? Well, as with another paintbrush in an artist’s satchel, it is another tool at our disposal. GAI is going nowhere, and the uses we have for it are only multiplying. So, it’s time to embrace the tools, and harness them as a springboard enabling us to push boundaries and build even better games.

the-future-of-generative-ai-showing-gai

Figure 3: GAI representing GAI in graphical form. (Image generated using Microsoft Copilot powered by Dall-E 3)

References

Bacchelli A. and Bird C., 2013, "Expectations, outcomes, and challenges of modern code review," 2013
35th International Conference on Software Engineering (ICSE), San Francisco, CA, USA, pp. 712-721

Feldman E., 2023, Are A.I. Image Generators Violating Copyright Laws?, Daily Correspondent,
Smithsonian Magazine
GitHub Inc., 2024, GithubCopilot [Large language model]

Growth Masterminds, 2023, Generative AI in Games with Unity CEO John Riccitiello, 02/06/2023

Lee J., Eom S-H. and Lee J., 2023, EMPOWERING GAME DESIGNERS WITH GENERATIVE AI, IADIS International Journal on Computer Science and Information Systems, Vol. 18, No. 2, pp. 213-230

Hu K., 2023, ChatGPT sets record for fastest-growing user base - analyst note, Reuters

OpenAI, 2023a, ChatGPT [Large language model]

OpenAI, 2023b, Dall-E [Image Generation Software]

Pennefather, P., 2023, AI and the Future of Creative Work, Creative Prototyping with Generative AI

Wolfe M., 2023, Using AI To Build A Game From Scratch (NO Experience), https://www.youtube.com/watch?v=IyKKhxYJ4U4

Xun X., Baoxing X., Chenglin M., Rongjian Y., Jie X., Rong X., Feng H., 2022, Factors influencing technological innovation efficiency in the Chinese video game industry: Applying the meta-frontier approach, Technological Forecasting and Social Change, Volume 178

Zed Industries, 2024, Zed [Integrated Development Environment]