The Impact of AI in Game Development

Artificial intelligence and video games have gone hand-in-hand since their inception. The concept of machine learning human traits was first popularized in Karel Čapek’s 1920 play, Rossum’s Universal Robots (Rossumovi Univerzální Roboti), which was also the origin of the word “robot.” Decades later, scientists began creating models of reason that relied on mathematical probability called artificial intelligence – essentially the robot's brain. The difference between an assembly line robot and those with artificial intelligence is that AI grants the machine free will. It is not programmed to do one action but instead to perform the most statistically probable task to succeed. With the advent of computers in the mid-20th century, the technology became more viable and was quickly used commercially.

The world’s first introduction to AI in video games came in Atari’s 1972 classic, Pong. The timeless volleyball-like game pitted a player-controlled vertical bar against an AI-controlled one, vying to deflect the “ball”, a single white pixel, past their opponent. As simple as it seems, this was the first time a team of engineers could teach a computer to replicate a human player. This all took place in an era when Cadillacs still sported whitewalls, telephones were connected by wire, and George McGovern was running for president against Richard Nixon amid Watergate. We now have electric vehicles, smartphones, and a whopping 59 new sovereign states. It makes one wonder what AI has been up to all these years.

Artificial intelligence is based primarily on ontology. It learns to differentiate concepts such as dog and cat while associating others like wolf and coyote. In Pong, the only differentiation needed was a goal or no goal, but first-person shooters and sporting titles quickly pushed AI to its limits. Games in the 80’s and early 90’s were kneecapped by technological restraints, but with the release of the Playstation, Nintendo 64, and Sega Dreamcast, so much became possible. A massive breakthrough in AI was EA’s The World is Not Enough, based on the famous James Bond film. For the first time on console, enemies in multiplayer deathmatches could react to danger, flank players, and camp in tactical locations to poach an easy kill. Though it had nothing on the likes of modern titles, this use of AI revolutionized the gaming world.

The World is not Enough (2000) let players square off against a whopping 3 AI enemies in deathmatches, capture the flag, and free for all modes, paving the way for a new generation of first person shooters.

As consoles and PCs advanced, so did the AI characters they hosted. It would be easy to create an opponent who always wins, but engineers have strived to make non-player characters as lifelike as possible. For instance, developers can now adjust how intelligent a computer opponent is. This is evident in games like Madden, where a stud linebacker like Fred Warner will read a counter and meet the running back in the gap, but a rookie like Junior Colson might need to be in position and make a tackle. If given a chance, AI would win every game against a human. Still, developers have found ways to make opponents more realistic and susceptible to mistakes, adding to our overall enjoyment of their games.

In the early 2020s, machine learning turned artificial intelligence into a real game changer. In layman's terms, engineers found a way to make robots teach themselves. This created exponential growth across all of AI and has played a crucial role in game development ever since. In addition to creating more intelligent allies and opponents, AI now plays a huge role behind the scenes, researching, drafting dialogue, and generating plot lines. Some AI platforms like CoPilot or the famous GPT4 can even write and review coding!

Using AI as a game-builder has its ups and downs. On the bright side, it allows for rapid idea and setting generation, but it will always lack that human touch. Using wisely helps developers reduce the cost of think tanks and brain-trusts. With the proper prompts and input, AI can spit out a handful of prospective stories and characters, and project managers can reap and edit at their will. As factions and characters become more developed, AI can better simulate their motivations and desires and develop great content. Some might argue that this content is not original, but in reality, AI works just like the human brain and generates ideas based on the sum of all it has read and learned in its lifetime.

If writing and coding aren’t impressive enough, developers now employ AI tools to assist their artists and level designers. Realistic image generation has existed for a few years but was not very useful until the implementation of “generative adversarial networks,” or GANs. A GAN pits two technologies against each other. The first is the image generator; users input a description of a character, landscape, or interior, and the computer generates a set number of images based on the input. These images are then run through a discrimination network, a technology used to spot fake pictures by scrutinizing the shadows' angles, corridors' depth, and, in some cases, simple logical fallacies like a European Marmot in an otherwise American biome. Developers can tweak how intensely the discrimination network vets the images before dishing them back to the users. This is especially nice when creating fantasy games, as not everything is meant to be hyper-realistic. Though AI cannot translate words into art as masterfully as a human, these images are often used in a new, pre-concept-art stage and give sketch artists some things to glance at for inspiration in the earliest stages of development.

Generative adversarial networks such as Wasserstein and DCGAN have robust discrimination networks capable of determining which image of the Eiffel Tower is real, and which is AI-generated. Can your eyes do the same?

Outside the creative rooms, AI also replaces human game testers to sniff out bugs and glitches during development. Though it sounds like a dream job, testing games is arduous and can run up massive bills for developers. Testers aren’t just there to say whether a game is fun but must also ensure every mission, cutscene, and dialogue line is functioning correctly. Imagine the man-hours needed to test every possible mission and storyline of a title like Skyrim multiple times over! With AI, developers can train an artificial player using “Reinforcement Learning” (think: operant conditioning of a computer) to push the game’s limits while reporting back any bugs. This process is faster, more thorough, and can save tons of money in late development.

By the end of Rossum’s Universal Robots, the robots have outgrown the need for human guidance. They kill off all but one man, Alquist, before realizing they lack the formula to create more of their kind. The play ends dramatically, with Alquist reasoning with two young robots until they develop a capacity for love and creativity. With this, the need for humans officially ceases, and the curtains fall. Fortunately, R.U.R. is an act of fiction, and even the most advanced AI systems in the world haven’t crested this final hill yet. This is why game projects in our era still need human oversight to ensure they feel natural.

A great video game can draw emotional and physical reactions from its player base. Whether this is jumping out of your seat while launching a buzzer-beater in NBA 2K, or shedding a tear when your favorite protagonist takes their last breath after hours of gameplay together, only a cohesive human team can create these feelings. AI doesn’t know what it means to have fun or feel upset. Moreover, artificial intelligence still lacks a long-term vision for the project. It can’t manage human resources or reliably check for plot or character consistency throughout a large project. Video games are works of art and will always require human input and oversight to be immersive and memorable.

AI has been at the core of many legal discussions in recent years. On the one hand, it has revolutionized how law schools teach students and how lawyers research cases. On the other hand, it has opened up one nasty can of worms about intellectual property rights. If an AI-generated sketch becomes the basis for a game’s main character, who has the rights to this character? Should the developers reap the rewards since they prompted this output, or should the developers of the AI platform own it since their machine created it?

Paid services have started mitigating this problem with lengthy user agreements about intellectual property rights. Still, most GPT-powered services are free, leaving the ownership debate open. One of the more vocal camps argues that generative AI is just a fancy calculator: it requires user input and then calculates an output. Calculator companies don’t get a cut of a mathematician's salary, so why should AI developers? The other side of the aisle argues that the hours of human input required to train their AI system means they should have a stake in anything it creates because this technology is so new and surrounded by murky legal waters. It is always best to consult with a professional or at least read the fine print of any agreements before using AI-generated content in a project to avoid charges of plagiarism or theft of intellectual property.

AI has become a great tool for economizing and speeding up the development process, but at the end of the day, it is still that: a tool. When managed properly, it cuts back on busy work for game developers and frees up their time for creative output. Taking the stress off of artists and storyboard writers gives them the space needed for their ideas to flourish, gives smaller studios a chance to compete with the fully-staffed, financially-backed powerhouses of the industry, and will help create countless classic titles to come. AI is a great way to turn an idea into a proof-of-concept, but is not capable of turning the parts into a whole. For help turning your game into a masterpiece, consider reaching out to Argentics. Our veteran teams of artists, programmers, animators, and project managers have the oversight to work with games at any stage of development, and can add the needed human touch to any AI-backed project, making it truly unique.