Generative AI – is the future of video games

Generative AI is everywhere! It's on every news site, in every social media conversation. The use of new powerful artificial intelligence tools capable of creating text, voice, or graphics with a simple prompt has opened up numerous exciting applications. You can use it for generating text, composing music notes, writing accompanying lyrics, coding for vector graphics, and animation.

However, amid all the loud revelations, carefully orchestrated press events, technical demos, and social media buzz, it's challenging to truly grasp the current state of affairs and whether it will indeed impact the video game industry. Does generative AI usher in a new era of tools that will make game development easier, more accessible, and affordable? Will creating AAA games now take weeks or even days instead of years? Are we about to receive the source code and assets for games like [new title], just by asking ChatGPT? Behind all these bold statements, well-planned press events, technical demonstrations, and social media hype, it's difficult to gauge the level of this technology and whether it will genuinely influence the gaming industry. Is generative AI the beginning of a new era of tools that will simplify, democratize, and cheapen game development? Will it now take weeks or even days to create AAA games? Will we get the source code and assets for titles like [new title] by simply typing a request to ChatGPT? The answer and the reality associated with it are much more nuanced than a catchy phrase circulating on Twitter. Risking breaking the internet's credo, I'd like to delve into what these technologies entail, how they work, the challenges surrounding them, and their readiness for production in the context of how game development traditionally operates.

In this article, we tackle the significant question of generative AI that continues to dominate our news feeds, emphasizing that the journey for generative AI to become a useful tool in the gaming industry will be a slow and gradual process.

Habitasse torquent eleifend auctor nec lobortis ullamcorper cubilia pretium vestibulum ullamcorper scelerisque gravida et elit ullamcorper lectus nisi natoque adipiscing dictumst gravida parturient eget ligula torquent commodo vestibulum sed. Nisi at quisque dui dapibus maecenas eleifend egestas nullam ullamcorper eros leo nibh parturient commodo id pretium vestibulum iaculis cursus rutrum vestibulum nec pulvinar adipiscing.

Review of Generative AI

Alright, let's start with the basics: when we talk about generative AI, what are we really referring to? This trendy term has dominated our news feeds for much of 2022 and 2023, and I need to break it down for you. So, here we go...

Generative AI is a simplified and sought-after term in the market for any AI system, typically created using machine learning, that can produce a certain artifact as its main function. GPT generates text, Stable Diffusion and DALL-E create images, GitHub Copilot writes code, and Speechify generates voice sounds. Got the idea? In each case, they are provided with input data, and something is created from this prompt. While we usually assume that an AI system generates answers to problems, it rarely produces results that we, as humans, can practically and creatively use. It's called 'Generative Artificial Intelligence' because it generates artifacts that are interesting for the user.

The term 'Generative AI' emerged thanks to the work of Ian Goodfellow and his colleagues in 2014 on the use of Generative Adversarial Networks (GANs), which can be applied to generate images. GANs stand alongside another form of machine learning known as Variational Autoencoders (VAE), which emerged from the research of Diederik P. Kingma and Max Welling in 2013 and has since proven its usefulness for generating images and processing natural language. GAN and VAE have become two driving methodologies of what we now call generative AI. Despite differences in their approaches to training, they share a process in which they aim to understand the fundamental properties and features of the analyzed data and preserve them in a compressed format—this concept is known as latent space. If a system can process the latent space sufficiently, it can subsequently attempt to decode it in new and interesting ways that still reflect the properties found in the latent space. Thus, the system can create images very similar to those analyzed during training.

Shifting Market Dynamics

The majority of this mystique revolves around the aforementioned surge in overall output and productivity. Major applications of generative AI, such as text generation, speech-to-text conversion, text-to-speech, and text-to-image, all in some form of development for several decades, have seen significant growth due to three key elements:

  1. Advances in deep learning AI, particularly in the development and training of large-scale convolutional neural networks.
  2. Access to large datasets that allow a better understanding of the underlying feature space and the creation of a broader spectrum of responses that better align with our expectations.
  3. The sheer technological power provided by cloud and edge computing today. Training large-scale systems in the cloud and deploying them to be easily accessible on desktops, laptops, phones, and any other devices that can quickly connect to the internet.

Of course, there's a fourth fundamental element - money! Over the last 10 years, we've witnessed a massive shift in how research and development in AI are funded. The share of AI research conducted in corporate labs was relatively minimal. Today, as evident from the owners of many AI systems, corporations take on the majority of the work on a scale that was previously unimaginable due to the enormous amount of money allocated to it. This has led to a sort of rat race, as companies ranging from the largest corporations to the smallest startups strive to make their bold statements as quickly as possible. In 2021, venture capitalists invested around $70 billion, and in 2022, approximately $46 billion. Considering that other speculative trends, such as Web3, NFT, and VR, are mostly fizzling out, and recent headlines about developments like GPT-4, it's safe to say that investments in generative AI will only continue to grow.

All of this has led to the emergence of new platforms, awkward false starts, strange refutations or reworks, as well as buzz, enthusiasm, and quite understandable apathy and disappointment in the state of generative AI. With the financial incentive, whether the desire to be the first to bring customers to market or simply to attract even more investors, we see new tools, systems, features, technological demonstrations, etc., related to AI almost daily on social media or press releases. Capitalism is working at high speeds, and only time will reveal which of these companies will stay afloat when the dust settles.

Generative AI for Games

Now, painting a broader picture, we can understand the significance of all this in the context of games.

There is tremendous potential in advancing games through the use of generative AI methods. Using AI to create textures and sprites, generate animations for specific characters, write descriptions for quest journals or story bibles. Generating plotlines for role-playing games. Creating real-time conversations with non-playable characters that exist in the world, are relevant, and react to player actions. All of this is entirely achievable, and in many ways, developers of all shapes and sizes can start using these tools today. Indie developers can use Mid Journey to kickstart the ideation process, helping create mood boards and concept art. And a programmer can use generative programming tools, such as GitHub Co-pilot, to assist in writing code for a new feature. We are entering a new era where generative AI is capable of changing the approach developers take to game creation. This idea has been enthusiastically embraced by everyone, from the smallest startups to the largest corporations, such as Microsoft, Google, and Nvidia. Recent announcements like Nvidia ACE platform for NPC creation, Inworld AI's Origins demo available on Steam, have not only caught the attention of developers but also players.

But regardless of their desire to capture this part of the market, the path to success in this area is not a straight line. Creating tools that are safe to use, meet the needs of developers, and do not lead to public embarrassment or even legal disputes is an ongoing effort.

We'll discuss some of the challenges reflecting the current state of affairs a bit later. But it's worth noting that there are numerous companies trying to address the issues I'll talk about. AI Dungeon creators Latitude, as well as Hidden Door, are working on language models for creating narratives in games. Ubisoft is experimenting with text generation for scriptwriting. Inworld and Convai are exploring the possibilities of creating more realistic avatars using workflows connected to game engines. Meanwhile, Unity is exploring the integration of generative AI into its workflows, and Roblox already has tools that game creators can use. The state of generative AI for games will undergo significant changes in 2024 and beyond.

Technological Challenges

First and foremost, let's start by checking reality: despite significant improvements in generative AI systems, the technology often doesn't possess the capabilities and power as frequently advertised. While many hype merchants claim that generative AI will revolutionize the gaming industry, it will take several months, if not years, before reliable, stable, practical, and seamless technologies emerge.

Unlike NFTs, AI does have the potential to significantly alter game development. However, these technologies and their associated tools and integrations are evolving at different speeds worldwide, as various companies tackle serious challenges in making these technologies more acceptable. Some have gained an advantage and achieved real success in this area. Still, it's only the beginning of a long journey for generative AI to become more suitable for practical and commercial use.

A substantial amount of work needs to be done to move from showcasing cool technologies to creating developer-ready toolsets. It's crucial that, as new demonstrations progress, a reality check begins. Certainly, a tool can generate a cool dialogue, an interesting texture, or even a fully-fledged non-playable character or game level. But will it do so with high quality in 100% of cases, every time? If there are remaining issues and risks requiring human intervention, it diminishes their effectiveness when marketed as a user-independent problem-solving solution. Additionally, it's worth emphasizing that many generative AI tools need to address the actual challenges faced by game developers, working in the same manner as game studios. It's important to recognize that games, even from the smallest indie studios to the largest AAA ones, are complex productions where multiple individuals often work on the same project in various capacities. Generative AI should address the real challenges game developers encounter and contribute to increased productivity, rather than seeking to replace them.

Concluding Remarks

In this article, we aimed to paint a broader picture of the current stage of development of generative AI, considering that it will garner increasing attention in the future from the global AI and gaming communities. As mentioned earlier, we observe the evolving landscape of many of these technologies, with some companies gearing up to introduce their products as valuable tools for game developers.