The Attack of Machine Intelligence: Will AI Revolutionize the Game Industry?
Looking at how the use of AI has evolved in the gaming industry, and what it holds for (Web3) gaming.
Recent Surge
Years and years of development in the AI space have now led to a point where people are just seeing what machine intelligence of capable of. A topic that quickly became hot on Twitter through programs like Chat GPT, Midjourney, and Stable Diffusion. Users were able to generate stories, headlines, Twitter threads, full pages of copy, code, and generate game characters, worlds, and objects with just a simple set of prompts. Something which felt like a wake-up call to many, and led to the question: ‘will AI replace our jobs?’.
Resistance by Gaming Twitter
Now, the recent surge of AI applications wasn’t welcomed by all. In a thread that Jon Lai wrote, a GP at A16Z, many workers in the game industry shared their opinions on the matter: ‘AI doesn’t have a soul’ ‘AI can’t replace creativity’ and ‘AI is just stealing others their work’. And, with Lai’s positioning in the space, being closely tied to Web3 gaming, this topic quickly got intertwined and labeled as another ‘scam by the crypto bros’.
It is true, however, that Lai’s headline in this thread is very exaggerated. Hinting towards an almost complete replacement of human work and talent. And understandably feels like discrediting artists and game developers for their hard work.
History of AI in Gaming
AI: A collection of algorithms that may generate results without requiring explicit direction.
~ Clustox
Whilst recent developments and progress in the space feel really new and groundbreaking, AI has already been used in games for years. It is as old as the 1950s when it was used as the first program to play chess. The efforts that were made to create machine intelligence that was able to defeat humans in strategy games like chess and poker, was something that made AI research significantly progress. This led to greater adoption over time, and in the 1970s all kinds of arcade games featured AI systems. Games like Pacman and maze-based games.
Applications
Bots and NPCs
Most games already employ some kind of AI method. In many cases tied to programming NPC behavior, with one of the first applications being pathfinding. Pathfinding includes how NPC characters interact with other characters nearby, navigate terrain, and how decisions (e.g. questing paths) the player makes affect their behavior.
Through the years different training models have arisen, assisting in programming the NPC’s behavior.
Decision Trees
Decision trees are one of the most basic machine-learning methods to describe choices and consequences. The model below is a simple illustration that shows how an NPC reacts if they see another player in a hostile situation. Decision trees provide insights into how the player’s decisions affect gameplay. This model is often used in story-driven games, like Skyrim or Red Dead Redemption.
Neural Networks
Neural networks are more advanced, showing more similarities to human brains, where AIs are able to learn according to training data. Having enough input data, neural networks can model complex game scenarios. Even being self-adaptive and changing according to changes in the game’s environment. This can be how the weather, and terrain change, or how game agents react to a disaster.
Neural networks don’t tell the game agents what to do, they learn what to do.
Different from decision trees, neural networks can be trained using input data before the game is deployed, and actively learn when the game is live as well. In the latter, the AI becomes self-learning and improves whilst the game is being played.
It has to be noted that the application of neural networks in games is still very shallow, due to the frictions of control, performance, and debugging.
Genetic Algorithms
Genetic algorithms make up a system that is seeking to make the next best move. Using this learning model NPCs can adapt and defend themselves from repetitive tactics that the player might use. For the player encounters with NPCs are unpredictable, and different encounters will be able to provide novel experiences. The player performing the same action, will also not always lead to the same results.
The below video is made by rct.ai, the makers of the Web3 game The Delysium, which gives an excellent inside with their demo into how NPCs make decisions using this model. The same tech will also be applied in The Delysium game itself with AI ‘Metabeings’.
Reinforcement Learning
The last model, reinforcement learning, teaches itself by trial and error. A model which has been widely used in chess, where the machine played out millions of scenarios and learned from the outcomes. The outcomes that led to success/ rewards are then further iterated on, so the machine keeps learning. Reinforcement learning is used when NPCs have to make decisions in changing and unknown environments.
The below video displays a Web3 game called AI Arena, here the player trains a fighter controlled by an AI to fight automatically. The player teaches the AI how to move and attack, and receives data based on training models, which they can decide to adjust parameters on. This includes a lot of trial and error.
NPCs and Personal Bonding
AI systems becoming more advanced will contribute to NPCs being experienced as more realistic (human-like). Something that will lead to instances where the player can create a deeper emotional bond with in-game characters.
Recently, Goddess of Victory: NIKKE was a game that was reviewed by me. And the way the game tried to craft deeper relationships with the player stood out. It was speculated that these bonds with in-game characters had a heavy correlation with spending.
It is likely from this point, with AI becoming more advanced, that relations with machines become so human-like, that they will be able to replace the relationships people have in real life. Something games already kind of do, as people often play for the reason of escaping reality.
Procedural Generation
Procedural generation: the creation of data by computers.
~ Jessica van Brummelen and Bryan Chen
Procedural generation is often used in games to create level designs, 3D objects, animations, or NPC dialogue. In No Man’s Sky for example this type of machine intelligence was used to create 18 quintillion different planet types. Something which would be impossible to do so by manual work.
Similar to NPC programming models procedural generation also needs data input. Using the example of roguelites, games that need a high level of replayability, procedural generation is often used in level/dungeon design. So, a game designer can input a number of level designs, and neural networks are then able to continually update a small set of numbers, which leads to differences in the level. Creating infinite variations.
A system the Web3 game and RPG roguelite, The Beacon, will make use of in order to provide much more longevity to their game. However, this isn’t something new and is tech that has been used as early as the 1980s.
Opportunities for Smaller Teams
Whilst procedurally generated content isn’t anything new, the applications got much better over the last few years. The tech will enable smaller teams, and Indie studios to produce a large number of assets in way less time, therefore, cutting development costs heavily. It won’t be surprising to see an increasing amount of teams leaning into tech or see ‘AI-native’ studios arise in the future. Developments don’t go unnoticed by VCs, which was slightly touched upon by Joakim Achrén a Play Ventures partner in the Metacast.
Stable Diffusion
Stable Diffusion is also an example of AI that generates procedurally generated content. The machine uses a common crawl database with 5 billion image-text pairs and creates content on the basis of the user’s inputs (prompts).
Image Enhancement
Another application for AI is its use in upscaling photorealism. In the example below Intel’s AI system was used to recreate the LA and southern California landscape in GTAV in a hyper-realistic way.
The machine was trained by recognizing objects and materials such as trees, streets, buildings, and cars. Then it used footage from GTAV and The Cityscape Dataset (a large video database on street recordings) to further understand how these elements correspond in the real world. And later overlay that real-world look.
Remastering Classics
Another application using image enhancement is to upscale the textures of classic 3D games. This would allow the players to relive their favorite games but in a modern jacket. Something that was also used by the developers of Mass Effect when releasing their remastered version of the original game, Mass Effect Legendary Edition.
Limitations in Game Design
AI has been around for a long time and has been used in games for good reasons for decades. Whilst it is getting close to becoming a viable option for in-game art, it is still far from getting close to replacing full teams and game production.
Limits with NPCs
It is still really difficult to have NPCs that can create a realistic experience and automatically produce an engaging level of content for the player. AI-based NPCs are also often programmed to act the best way according to the player’s behavior, which can become predictable quickly.
Lack of Creativity
Creativity is defined as the tendency to generate or recognize ideas, alternatives, or possibilities that may be useful in solving problems, communicating with others, and entertaining ourselves and others.
~ California State University
Whether AI can be or is creative is a different debate, however, procedural generation systems are less creative than a human can be. Procedurally generated levels for example are way more repetitive than when a level is completely crafted by hand. Something the rogue lite Dead Cells is also criticized more often for.
Limited to Training Data
AI isn’t able to use context outside their training data, something which can create a rather robotic experience for the player. And feel less immersive than a handcrafted experience, with an added human touch. Machine intelligence is also not able to understand ethics, which can be problematic at times.
Further Controversy
Workers in the gaming community, primarily artists feel attacked by AI. Rightfully so, since it can accelerate a team’s creativity by a magnitude. And there is a good chance many artists will be replaced by machine intelligence. Especially now, when the gaming industry is experiencing a higher degree of layoffs, and it seems AI can be a tool to save costs.
‘Stealing Art’
A common argument is, that AI art is ‘stealing’ the work from others. Since programs like Stable Diffusion use the art of others in their data sets to create iterations, it is derivative work. However, have artists not been inspired by the work of others for decades and centuries? Teaching themselves using methods and techniques of the works of successful artists.
There’s also the argument of an artist using their own work as input. Does it make them a thief of their own work, or make them smart and speed up the process massively? The argument of ‘stealing art’ doesn’t hold ground.
High on Life
The game High on Life is a recent example of how controversial the topic is. The developers made use of Midjourney AI, a tool that uses procedural generation to produce a wide range of images with varying styles.
And we used it to come up with weird, funny ideas.
~ Justin Roiland, Squanch Games Founder
Furthermore, AI was used to prototype some character voices, having only a minor role in the game. Justin also noted that he sees AI as a way to make content creation ‘incredibly accessible’.
Many believed this is just the beginning of AI taking away the work of actually skilled artists, and this didn’t stay unnoticed on Twitter.
Adapting
Artists worrying about the security of their passion and job is a very valid concern. As it is likely more game studios will be downscaling in the upcoming months. Especially in a climate where it is getting harder to receive funding, something which is even more applicable for Web3 gaming studios. The best way for an artist to survive is to pivot and focus on the points AI isn’t good at, and use it as a tool. This is similar to how generic Twitter threads will quickly die off because of an influx of AI-generated Tweets.
Hostility towards AI in the artist’s community is way more significant than coders have shown. Explained well by the below Tweet, basically saying that the supply/demand function is way more favorable for coders than artists.
An Overview
The usage of AI in game development has benefits that are mostly related to creativity, streamlining processes, and saving time and budget.
Here’s an overview of the applications:
Image enhancement such as adding photorealism or upscaling pixels
Generating content such as story, level, character, and world design
Giving intelligence to NPCs
Assisting in quality assurance and balancing in-game complexities
Playing a basic role in coding
The Future
The growing tool stack of AI applications, and the kind of benefits this carries, will likely lead to greater adoption in game design. Especially on the content side, where teams will be able to cut costs and time in development. But also growing on the programming side.
This will create opportunities for game developers to produce games with more content in shorter development cycles. Leading to a higher output of games, and more competition, but only a small amount of games succeeding.
A higher output of games will especially be beneficial to the growth of Web3 games. Since this increases the chances of one of these games being picked up by a larger, traditional gaming audience, which in turn increases the overall awareness of the industry. Especially, since it could be the case where many of the larger Web3 games releasing in the next two years could have a bad product-market fit, and won’t succeed.
Sources
https://www.bartleby.com/essay/Advantages-And-Disadvantages-Of-AI-In-The-PCFCZ33WCU
https://www.gamedesigning.org/gaming/ai-in-gaming/
https://johnnyholland.org/2022/08/the-pros-and-cons-of-ai-being-used-in-gaming-development/#r
https://game.intel.com/ww/stories/how-intel-ai-drives-in-game-photorealism-
https://pixelplex.io/blog/how-ai-enhances-game-development/
https://readwrite.com/5-ways-artificial-intelligence-will-revolutionize-game-development/