How Generative AI Can Charge Web3 Game Development
Exploring the Technology’s Applications In Various Areas Of The Game Life Cycle
Recent developments in AI technology have made waves across the games industry and beyond. VCs and industry observers alike are making increasingly bullish predictions on the impact that AI will have on the games industry. Generative AI, specifically, has driven tremendous hype, allowing creatives to bring new ideas to life at incredible speed. This has been borne out in the rapid adoption of these new technologies. To wit, OpenAI’s ChatGPT reached its first million users in less than one week and crossed the 100 million user mark after only 2 months – far faster than previous growth success stories like Netflix, Twitter, Facebook, or Instagram.
Time taken to reach 1m users for the most popular web applications | Source
As games industry veterans, we recognize the potential that this powerful new set of tools and techniques hold for reshaping the game development process – much of which we will explore in greater detail in this piece. However, as builders in the blockchain gaming space, we also understand all too well the perils of being overzealous and getting caught up in the excitement of a new wave of venture funding, consumer interest, and speculative prognostication. As such, it’s crucial that we bring a critical lens to the discussion around AI in game development and understand in detail what is possible, what is plausible, and what is unlikely to occur in the near term. After all, anything that affects game development broadly will also impact blockchain gaming, and as such we must have a clear-eyed view of the downstream implications of these new technologies.
Before exploring specific applications to game development, however, let’s first take a moment to unpack what we mean when we use the term “generative AI” so that we can better understand what this technology is capable of.
What is Generative AI?
Where Generative AI sits within AI | Source
Generative AI is a form of deep learning able to generate new content (images, text, audio, etc.) by training a neural network with existing content. This relies on machine learning algorithms that are able to recognize underlying patterns in the training data (the existing content, or model inputs) and create all new content without human influence.
While it isn’t hard to imagine developers using AI tools to create a wide range of content – from text and images to (eventually) entire game worlds – a pragmatic examination of the technologies’ strengths and weaknesses reveals more actionable takeaways for forward-thinking game developers.
Applications to Game Development
Generative AI tools are already being tested and implemented across the games industry. Most early implementations are offshoots of the technology’s strengths, which largely center around generating text or images from prompts and first-pass content creation:
Prompt-Driven Text & Image Generation: Tools like ChatGPT take textual requests as inputs, meaning that natural language prompts, character dialogue, narrative exposition and even code snippets can be used to create new content. These various forms of text can be utilized to generate 2D images across a wide variety of styles, and through further fine-tuning can be effective at consistently replicating visual themes, allowing for rapid iteration at little to no marginal cost.
First-pass content generation: Generative AI helps creatives get past the blank page and enables quick brainstorming of game concepts by efficiently visualizing ideas, outlining plots, creating characters, and more. Rather than create a finished product, the AI helps to kickstart the creative process for developers to then layer on their own ideas.
With these building blocks in mind, let’s delve deeper into how they can be combined with existing game development tools and practices to aid developers, accelerate game creation, supplement key studio functions, and even unlock entirely new types of gameplay.
At the earliest stages of game development – ideation and prototyping – generative AI can be a tremendous boon for efficiency. AI can help with brainstorming concepts, creating rough drafts, and honing in on an art style or color palette. Once early concepts have been outlined, AI can also assist with generating a game’s fiction, rapidly filling the gaps in new game worlds with characters, locations, and history. Given the emphasis on worldbuilding in many NFT projects as it relates to procedural NFTs and PFPs, this presents a meaningful opportunity for web3 builders to create new community experiences with customized lore and generative fiction.
Prior to the rise of generative AI tools, developers have sought to cut down on the costs involved in worldbuilding by relying on techniques like procedural content generation, or proc-gen. This method has been in use for years now to create “randomly” generated maps, with Minecraft perhaps being the most noteworthy example. Procedural generation has begun to appear in web3 gaming as well: the team at 0xPARC has already explored using procedural generation in fully on-chain games to create what they refer to as “autonomous worlds”.
An example of a procedurally generated map | Source
However, proc-gen has important differences to generative AI: it relies on human-made algorithms to reproduce existing content in a variety of ways, rather than creating new content based on training data. That said, the efficiencies enabled by procedural generation can now be taken one step further: instead of being used to create worlds, proc-gen can be used to create training data for generative AI models. Using the two techniques in tandem will enable more reliable outputs for game developers without the need to rely on reuse of existing assets over and over again.
Beyond the creation of virtual worlds themselves, one of the most commonly cited (and most hotly contested) applications of AI to game development is in the area of non-player characters (NPCs) and chatbots. Companies operating in this space use AI to power NPCs with goals, personalities, and agency. Inworld AI and Charisma are two prominent examples here.
The hype around this use case ties in with the rising interest in populating virtual worlds and “metaverses” with believable characters for players to interact with. Rather than hiring a team of writers and narrative designers to flesh out an intricate virtual world, these tools can drastically reduce development costs (in both time and dollars) and allow teams to focus their efforts elsewhere. In the world of web3 games, these characters can become digital assets that players are able to carry with them from one virtual world to the next. Developers like Alethea AI (AI-powered characters) and Altered State Machine (“Non Fungible Intelligence”) are already pioneering in this space.
A sampling of startups working on AI Characters | Source
Engineering & Writing Code
Just as generative AI can create believable text strings, it can also produce working code that can aid in early prototyping. For novice programmers and non-technical creators, AI tools like GitHub’s Copilot or Replit’s Ghostwriter can accelerate the speed at which a game idea can go from zero to one. It should be noted, however, that generative AI’s penchant for inaccuracy and consistent bias as a result of its training data can at times be subtle, rendering it difficult to use effectively (at least, as presently constructed) for amateur engineers.
With that said, these tools can also be a tremendous unlock for experienced developers, allowing them to work faster and more efficiently than ever before while potentially revealing solutions that they may never have thought of. For example, web-based projects (an area in which many web3 teams are building) can benefit immensely from the efficiencies AI tools bring to building around APIs. A further application of AI coding tools to web3 gaming can be found in boilerplate smart contract development, which benefits from the large amount of Solidity code already available via GitHub and Copilot. Of course, it should be noted that this code would still require careful auditing.
Additional use of AI tooling as it relates to writing code is taking place in the field of analytics, where generative AI tools can be used to democratize data analysis across organizations. Specialized tools like Borealis AI enable non-technical developers to easily compose database queries without prior coding knowledge, or to rapidly summarize and understand complex datasets. The same will be possible for on-chain data queries of web3 games and dapps like Nami is building.
New Gameplay Experiences
AI Dungeon | Source
Beyond improving the game development pipeline, generative AI can even be used to create entirely new, hyper-personalized gameplay experiences. The most famous example of this is AI Dungeon, an early innovator in designing games with AI-driven storytelling. Initially a text-only experience driven by the older GPT-3 model, the game has now grown to incorporate image generation via Stable Diffusion, bringing a new layer of immersion to its hyper-personalized role-playing game experiences. Similar types of interactive fiction experiences can be brought to web3 communities via integrations to Discord servers or web interfaces.
Blockchain gaming studios have also begun to utilize AI in novel ways, wrapping complex algorithms in NFTs (though this is less related to generative AI, specifically). AI Arena is one such example where players “can purchase, train and battle AI-enabled NFTs in a PvP platform fighting game.” Generative AI tools can also be applied to character creation, as studios like Spellbrush have already proven. It’s not hard to imagine a future where games could provide an end-to-end AI-driven experience allowing players to create characters, give them backstories and personalities, mint them as NFTs, and then bring them from game to game as companions or competitors, all the while learning and adapting to each experience.
AI Arena | Source
Marketing & Community Management
Outside of game production, generative AI has immediate applications to other critical disciplines such as marketing, public relations, and community management. One example of this is in player communication, where developers can use AI to generate blog posts, social media updates, and other important messages to players or clients. We have already seen this benefit web3 teams in speeding up communication and providing increased transparency. This can help with the aforementioned “blank page” problem, but it can also aid in improving writing quality in general among team members – particularly those for whom English may not be the native language.
Additionally, in the community-driven world of blockchain games where much of the dialogue takes place over Discord and Twitter, generative AI tools can help to ease the burden of frequent updates. Eventually we’ll see the emergence of AI-driven chatbots trained on internal company data (FAQs, internal wikis, patch notes, etc.), available to answer player questions 24/7, provide timely updates to the community, or even help with onboarding new employees or outsource vendors.
The Way Forward
Generative AI is a powerful tool that has the potential to reshape the game development process. The examples articulated above represent just a small sample of the potential applications across the game development pipeline. However, current iterations of the technology also suffer from issues such as genericness, inaccuracy and lack of real-time information.
In the future, we expect to see more domain-specific applications (e.g. art, narrative, coding, etc.) emerge to help optimize game production. It’s clear from the wealth of tools popping up in the generative art space (Scenario, Ludo, etc.) that this will be the area most impacted in the near term, but we will also be monitoring how game engines like Unreal and Unity seek to incorporate these tools. At Nami, we are also particularly interested to see how tools develop on the analytics side. Ultimately, any tools that improve the game development process will also meaningfully impact the web3 gaming space, so we will be watching closely for opportunities to better serve developers, studios, and the gaming industry.
Thanks for reading Nami! Subscribe for free to receive new posts.
Generative AI can be used for a variety of tasks, such as creating levels, characters, textures, and even entire game worlds. With AI, game developers can quickly create assets that would take days or weeks to create manually. In addition, assets created by artificial intelligence are often more realistic than those created manually servreality.com