The Future of Artificial Intelligence in Video Games

The power and influence of artificial intelligence is inescapable; it’s used within our homes, cars, phones, and computers. Because of this ubiquitous presence of AI in our lives, it’s easy to imagine that with their myriad hypothetical elements and their graphically, thematically, and sonically evolved interfaces, video games must also boast highly evolved AI.

While that isn’t the case—yet—there is truly exciting progress happening in gaming every day, turning it into a mammoth industry poised to explode even further from its already staggering size: 2.6 billion people playing in concert¹ in a $160 billion size market². And make no mistake about it, AI will play an increasingly important role during this transformation, informing how games look, sound, and play.

Below, we explore some of the key ways in which AI is currently being applied in video games, and we’ll also look into the significant potential for future transformation through advancements inside and outside the game console.

The graphics, interfaces, and nuanced layers that overlap in complex 3-D games like Minecraft, Grand Theft Auto, and God of War make it seem as though complex algorithms are at play, but the reality is that these virtual journeys are still very much powered by the same core principle that made playing old-school favorites like Snake or Space Invaders or Pac-Man so compulsively addictive: pathfinding. Let’s face it, that’s not exactly the pinnacle of self-learning.

The algorithms driving games today can’t really be classified as learning intelligence because video games are not currently capable of developing behaviors on their own and doing things they weren’t programmed to do. They are driven by “if-then” scenarios, which is by design: games aren’t supposed to be unpredictable or unbeatable. Today’s “if-then” scenarios are, in reality, quite simple:

So can AI inside video games become self-learning? Can video games adapt as we play them? Can AI teach itself to create imagery? Yes, maybe, and definitely. All of these endeavors are already on the horizon. But they are enormously costly.

Cost and control play a huge part in why many video game developers are hesitant to build advanced AI into their games. It’s not only cost-prohibitive, it also can create a loss of control in the overall player experience. Games are by nature designed with predictable outcomes in mind, even if they seem layered and complex.

This limits the use of AI in video games today to maximizing how long we play and how good of a time we have while doing it.

So how is AI used in games today? Says Eric Nesser, a global director of 3-D and interactive for a multinational advertising company, “Today, video games use AI to enhance computer graphics and to create ‘smart’ opponents for a player to battle. The most practical, most active application of AI today is to combine it with technology like real-time ray tracing for photorealistic lighting, shadows, and reflections.”

Because of AI, a game like Grand Theft Auto 5 can look stunningly photorealistic. AI can modify imagery within the game through a convolutional network that takes images of actual cityscapes frame by frame and superimposes them to produce things like smoother asphalt, shinier cars, and lush green backgrounds that really do make it look like you’re speeding through California.

With the push to create smart opponents through AI, there is a leveling off driven by an always-connected world and the existence of 2.6 billion gamers playing simultaneously. This means the emphasis remains on multiplayer games that bring real players together, with the creation of an AI-powered opponent taking a back seat. Nevertheless, says Nesser, “it’s certainly something we’ll see oscillate in the other direction in the future—trends like this always do.” He adds, “With 5G and broadband, the short-term focus will be on playing with real people.”

“In the future,” says Nesser, “we can expect Generative Adversarial Networks (GANs)—pitting two neural networks against each other to produce new video game content. GANs can implement style transfer, use photographs or video feeds to replace in-game assets with something that looks hyper-realistic, and even help create game assets that have never been seen before, including new game mechanics or game rules to make games evergreen, perhaps even help create new games.”

Adds Nesser, “To me, to watch GANs work in the realm of content creation is like watching a computer ‘dream.’ With very little direction or ‘steering,’ GANs always produce interesting results, even if they’re not always necessarily usable ones. But with a human at the helm who knows how to augment their own abilities with those of a GAN, it’s sort of like being able to paint with a paintbrush from Fantasia. It’s like working with an entity that sort of has a life of its own.”

In a TED Talk on the transformative power of video games, Herman Narula argues that the really important transformation video games will bring will come from the staggering amount of people who today are playing in concert. This shared reality, he argues, will result in unprecedented technological advancements, myriad new jobs and opportunities, and of course, ethical and business challenges posed by questions on how information is gathered, centralized, and used.

The thinking is: If billions of people are playing video games at any given moment, then millions will end up having jobs in gaming, whether those jobs are in research, development, engineering, art creation, story-building, or marketing. This makes the job-growth potential so big, says Narula, that “it could be the earliest income potential for kids.”

No matter how we look at it, video games will be one of the biggest job creators of the future.

While the use of AI inside video game development can and is creating enticing new virtual worlds, and numerous jobs alongside it, what researchers, scientists, and developers are also doing is using video games to help AI learn and problem solve. And in the process, they’re aiming to move the needle forward in important ways toward real-world efficiencies across industries.

For DeepMind, a deep learning research lab, the hope is that once success is achieved in the virtual space, it can then be effectively translated into highly capable AI across various verticals in the real world, including spaces like manufacturing and autonomous vehicles.

Raia Hadsell, a research scientist at DeepMind, uses “reinforcement learning”—an extreme trial-and-error type of machine learning—to teach AI to problem solve, first by playing simple games like Pong, then increasingly complex ones like Dota 2 and StarCraft II.

In an interview with the New York Times, Greg Brockman, who oversees research at a similar lab called OpenAI, said: “Games have always been a benchmark for AI. If you can’t solve games, you can’t expect to solve anything else.”

Which brings us to another thing altogether.

Janelle Shane, an optics research scientist, AI researcher, and writer, challenges us to reframe our thinking in terms of AI and unpredictability, which has traditionally created barriers not just for the public’s perception of the usefulness and benevolence of artificial intelligence but for researchers as well, albeit in very different ways. Shane asks us to think not in terms of how or why AI may be dangerous or unpredictable on its own (it really can’t be, as we are truly far from self-thinking, free-will machines) but in terms of giving AI “the right problem to solve.”

It is a reminder that artificial intelligence can only be as evolved, efficient, unbiased, and useful as the people behind it.

¹Narula, Herman, “The Transformative Power of Video Games,” YouTube Video, 2019.arrow_upward

²Reuters, “Report: Gaming Revenue to Top $159B in 2020,” May 2020.arrow_upwardReturn to footnote reference

Return to footnote reference