Since the birth of Pong in 1972, digital games have been steadily evolving along with computer technology.
From 8-bit consoles enabling classics Pac-Man and Super Mario Bros in the 80s, to the 3D revolution of the 90s with first person shooters (FPS) like Wolfenstein 3D, Doom, and Quake, followed by the rise of online multiplayer games during the internet boom, and most recently ultra-realistic graphics driven by advancements in graphical processing.
Today’s evolution of AI for gaming is no exception. AI in video games is becoming increasingly vivid with human ingenuity using this tool in numerous ways, making computer games better and easier to develop.
And still, there is so much more to come.
Let’s explore!
AI That Pwned People
An imperturbable focus, capacity to retain all relevant information, utilization of powerful mathematical foresight, and split-second reflex response – these are the traits of a perfect gamer… and a computer has them all.
Deep Blue’s narrow victory over Gary Kasparov, the undisputed chess world champion of the early 90s, marked the pinnacle of our belief in human superiority over AI in gaming. Since then, computers have steadily advanced in strategy games, reaching what seems like complete dominance, exemplified by Google’s DeepMind project AlphaStar and OpenAI’s Five, prevailing over us in complex real-time multiplayer games. They even succeed when constrained by limited information and artificially delayed response times.
At the center of many of these AI gaming successes lies the AI branch of reinforcement learning. It draws parallels with reinforcement in psychology, wherein an agent’s interactions with its environment, with positive rewards for success and negative consequences for failure, enable the agent to make correct decisions when faced with similar situations. This process is repeated continuously, as computers do not fatigue, allowing them to encounter an infinite number of situations and derive insights from them.
DeepMind has advanced to the extent of publishing algorithms capable of learning virtually any game, such as hide-and-seek, without prior knowledge of the game’s rules. This highlights the impressive generalization and abstraction capabilities of deep learning neural networks.
As this technology becomes more simplified and accessible, following the trend of commoditization seen with useful technologies, we can anticipate the emergence of adaptive-level AI players. These AI gaming opponents will intuitively adjust their gameplay to provide users with challenging opponents of nearly equal skill level, facilitating improvement with minimal frustration.
AI and Character Behavior
The gaming audience has come a long way from being content with the banal left-and-right movement of sprites in Super Mario Bros. Over time, increasingly more complex algorithms have been developed to enable intricate behaviors.
While sophisticated NPC behavior might appear magical, in most cases, it relies on predetermined scenarios with added elements of randomness. Actions such as gunfire from a patrolling guard who detects your presence, a car stopping to avoid hitting a pedestrian, or a monster attempting to attack you are typically dictated by a meticulously designed network of if-else statements, influenced by a multitude of environmental cues and decision-inducing variables.
Drawing inspiration from Tesla’s recent endeavor to condense 300,000 lines of complex code into one AI model, a similar approach can be applied to NPC logic. By adapting reinforcement learning techniques to instruct NPCs on behavior, for example negative consequences when a car veers off road or runs over pedestrians, we may achieve more organic and diverse behaviors.
Conversations with NPCs can benefit from the ongoing Generative AI revolution. As people become accustomed to conversing with LLM agents, embedding them to advance narratives or plots in games is becoming increasingly feasible.
Developers of CyberPunk by CD Projekt Red have experimented with unscripted NPC dialogues, and although the concept was temporarily shelved by the development team it was later made available through fan mods.
It is only a matter of time before NPC technology fully adopts and integrates artificial intelligence in video games methods as the standard go-to approach.
AI in Game Design and Development
Rogue’s groundbreaking procedurally-generated dungeons were primarily built on a foundation of random numbers combined with geometric constraints to create navigable environments. Naturally, ensuring both entry and exit points in the dungeon is a fundamental constraint. However, in contemporary game development, any structural element, item, or embellishment is typically limited to decisions made explicitly during the development phase.
This does not need to be the case.
Recent advancements like OpenAI’s Sora demonstrate that we no longer require models and rigs to craft convincing 3D animations. Extending this to computer games, we may eventually compose nuanced worlds and narratives purely in human language and generate entire games from such input. Making modifications to any element could become just as straightforward. A recent proof of concept, codenamed Genie by Google’s DeepMind, exhibits basic capabilities that were once unimaginable.
After two years in the Generative AI revolution, pair-coding models have emerged as immensely powerful tools for developers. For instance, GitHub’s CoPilot is proficient in all common programming languages, including those frequently used in AI and games development, and is well-versed in specific frameworks such as Unity and Unreal Engine.
A shorter time to market yields more games, while automatically generated content ensures greater enjoyment with each game. Gamers benefit from both the AI and video games – a double-win scenario.
Conclusion
The time is now. The AI revolution is in full swing, transforming the landscape around us. As AAA computer games continue to evolve, the artificial intelligence in video games evolution will inevitably trickle down to casual gaming as well.
Infinity Labs R&D Launches a New AI Specialist Training Track
Based on a deep understanding of the national-strategic need for AI specialists, the Israel Innovation Authority has entrusted us with training the next generation of AI experts and advancing the local industry. As an experienced R&D company that has developed products in artificial intelligence and successfully integrated thousands of graduates into the high-tech industry, we take pride in responding to the challenge by launching a dedicated career track for AI developers.
The Artificial Intelligence career track is a groundbreaking path that will transform you from university graduates into the future leaders of the Artificial Intelligence revolution. Throughout the training program, you will learn the technologies that drive artificial intelligence (such as Machine Learning and Deep Learning), develop advanced models, and implement them.
The high-tech industry in Israel longs for AI trailblazers who can integrate artificial intelligence technology and propel existing products forward. Every company understand that this is inevitable and that those who fail to embed AI technologies might find themselves falling behind.
Upon completion of the training program, you will be integrated into R&D teams as artificial intelligence experts, as the pioneers who will build and lead the AI systems of the companies.
Our business model is based on partnership. The training program is provided at our expense for suitable candidates. Only when you start working in a sought after position, upon successful completion of training program, do we see a return on our investment and can maintain a profit model. Our success is dependent on your success!
This model is valid for two years; upon completion, 95% of our graduates are placed within the company.
