0.9 C
New York
Monday, February 26, 2024

In GTA 6 what AI will we see?

In GTA 6 what AI will we see?

Since the 1950s, artificial intelligence (AI) has been a big deal in gaming industry, especially when Nim was invented. It got people frustrated and gained a lot of attention when engineers made a 23-kilogram machine at W. L. Maxon Corporation, and it kept beating humans at Nim regularly. That was AI taking its first steps into gaming. Fast forward to now, and Nim hasn’t lost its edge. Even in the present day, if you try your hand at it here, winning won’t come easy in the initial tries.

User interface of Nim (Archimedes-Labs)

Nowadays, most video games use a subset of AI called Generative Artificial Intelligence (GenAI). It brings about clever, responsive, or adaptive behaviors, marking a shift in game development. For example, in Left 4 Dead, the zombies get smarter and change how they act depending on what the player does. So, if the player keeps using the same tricks, the zombies catch on and mix things up, making the game more exciting and challenging.

Although many companies have gained significant reputation in the gaming industry for using GenAI, Rockstar Games stands out as one of the best players. They are famous for creating highly interactive games, with their latest release, Red Dead Redemption 2, which came out in 2018, opening new possibilities for how interactive an open world can be. As they are already making GTA 6, their newest game set to release in 2025, Rockstar Games continues to push the boundaries of gaming experiences.

When games had a boundary wall preventing our characters from exiting, gamers were excited about all the cool things that could happen beyond the boundaries. However, today’s gaming engines are putting an end to this limitation, and among them, Rockstar has overcome it by developing games featuring endless maps. This is particularly clear in the GTA series, starting from San Andreas, where you can drive cars along an infinite route that never ends.

This concept draws inspiration from the wave function collapse in science, a phenomenon observed in quantum mechanics. Interestingly, this scientific concept has found a practical application in the gaming world through procedural generation. The wave function collapse algorithm, which is originally a peculiar aspect of particle behavior, can be harnessed programmatically to generate diverse content in video games. By using a set of initial parameters and rules, this algorithm allows for the creation of dynamic and expansive game worlds, offering players a virtually limitless space to explore and interact with.

Generating endless Town From WaveFunctionCollapse (jaxry.github.io)

The above diagram illustrates the expansion of a town map using the WFC algorithm, where the only set rule is that roads (yellow path) must be connected. This is logical because, in the real world, roads we see follow the same behavior, and additional sets of parameters can also be passed to create a more realistic environment.

We observe a similar behavior, but with a more complex implementation in Red Dead Redemption 2, a Western-based game that does not contain modern city-like architecture but rather mountains and deserts. Even outside of the city maps, a WFC-like behavior is used to expand the game world, but containing random NPC moments in those regions. For instance, you might encounter a stranger out of nowhere in the middle of the jungle, stumble upon many abandoned houses, or even come across a pack of wolves in winter. This can be achieved through a set of rules and parameters to make the game more realistic.

Generating Mountains From WaveFunctionCollapse (jaxry.github.io)
Spaceship seen on an old house in the jungle in RDR2 Game

Although the GTA series already features more real-world locations compared to their other games, IGN has published several articles detailing about the potential map and locations in GTA 6. The trailer suggests that GTA 6 might have an even bigger map than their previous releases. With the WFC algorithm and a large set of rules and parameters, they can expand the outward regions of the map and incorporate more NPC moments, similar to what we’ve seen in the game Red Dead Redemption 2. It’s likely that we’ll witness the same approach with GTA 6, offering more detailed environments not only in the main region where the story unfolds but also outside of it, featuring individual NPC stories.

NPCs are like the eyes and ears of the game world. Developers frequently use biologically inspired movement algorithms such as particle swarm optimization, and firefly algorithms to create the perception of intelligent behavior in groups of non-player characters (NPCs). However, recent advancements in GenAI have shifted the focus towards making NPCs think more like real humans.

Replicas Studio recently developed a playable demo featuring Smart NPCs in the Matrix Awakens project. In this project, users can engage in conversation with NPCs, asking them a variety of questions. What makes it even more interesting is that each NPC have a name, age, professions, providing a whole separate life. Their responses are more intelligent compared to those in previous games. A YouTuber tries to convince an NPC that they are living in a simulation. This is their conversation:

Youtuber: You’re not living in a real world, this is all made up.

NPC (named Liam): What do you mean not real? I feel alive, and this city is amazing.

Youtuber: Ever wonder why there are no pigeons?

NPC (named Liam): Of course, pigeons exist. They’re all over the city, just like us in the left field.

Youtuber: Hey, Liam, what is today’s date?

NPC (named Liam): It’s a beautiful day in May, my friend. Let’s enjoy it.

Youtuber: Liam, today is July 18, 2023.

NPC (named Liam): Oh, I must have lost track of time.

Replica Studio did not open-source their work. However, recently, developers from Stanford University and Google DeepMind released a starter kit for building and customizing your own version of AI Town, a virtual town where AI characters live, chat, and socialize. They introduced generative agents that simulate believable human behavior in interactive applications. These agents are built upon a novel architecture that combines a large language model with mechanisms for synthesizing and retrieving relevant information. The architecture comprises three main components: the memory stream, a memory retrieval model, and a comprehensive record of the agent’s experiences.

Interactive Simulacra of Human Behavior Paper
Two-Agent Conversation in AI Town

Replica Studio started the trend of giving NPCs more reputation in a game, meeting gamers expectations may take some time. However, the recent AI Town creation is more focused on what gaming engines are currently doing, providing memory remembrance to NPCs. This involves remembering what they did yesterday and what they will be doing today.

Rockstar has started incorporating this feature in their games already. Red Dead Redemption contains numerous interactions between our character and NPCs, such as giving money to beggars, finding thieves, and earning rewards. The game also features interactions between NPCs, like an NPC building a house, which slowly builds throughout the game. If you somehow hurt that NPC, the house remains unbuilt forever. Additionally, there are scenarios where one NPC may start a fight with another to resolve a dispute, or a group of NPCs may go on a bear hunt, only to never find it. These stories involve the memory remembrance feature, allowing our main character to be a part of these narratives and influence their outcomes.

The finished house outside Valentine

The GTA 6 trailer indicates that beyond the main story, NPCs will play a important role. Not all NPCs will have their own whole separate lives, but many of them will have unique stories. What stands out is the inclusion of unique side missions for these NPCs, adding an extra layer of engagement for players. Each NPC’s journey will have its own twists and turns, providing gamers with the opportunity to observe and interact with the diverse and dynamic virtual world. This emphasis on individualized NPC experiences suggests that GTA 6 aims to create a more interactive environment, where players can explore not only the primary storyline and also explore the details of the lives and missions of the characters in the game world.

Rockstar has never focused on achieving realistic graphics but has worked to improve with every new release to be a competitive contender in the market.

In 2021, engineers from the Intelligent Systems Lab at Intel Labs introduces a new machine learning project called “Enhancing Photorealism Enhancement” that intends to enhance the photorealism of rendered images from the popular game GTA 5. Their approach involves training convolutional networks to enhance the photorealism of the rendered images using intermediate buffers produced by a conventional rendering engine. The method integrates learning-based approaches with conventional real-time rendering pipelines and is designed to be compatible with real-time ray tracing. They also demonstrate the effectiveness of their approach on other datasets such as Cityscapes, Mapillary Vistas, and KITTI.

YouTube — Intel ISL

The key components of their approach include an image enhancement network, G-buffer encoder network, and a modified HRNet architecture. The image enhancement network is trained with two objectives: an LPIPS loss and a perceptual discriminator to evaluate the realism of output images. Additionally, they propose a new strategy for sampling image patches during training and develop a new adversarial training objective to facilitate enhancements that are geometrically and semantically consistent with the content of the input image.

YouTube — Intel ISL

Although this approach is useful for making cutscenes more realistic, unlike other game engines, Rockstar has a habit of generating cutscenes with the same graphics as in the game to maintain consistency throughout. We won’t be seeing this in GTA 6, as creating cutscenes with more realistic graphics could have performance issues. Throughout the trailer of GTA 6, multiple cutscenes have been shown, suggesting the same method has been applied as Rockstar did in their previous games.

RAGE has gone through different versions since 2006, starting with an unexpected game, Rockstar Games Presents Table Tennis. It had major updates for GTA 4, the remasters of GTA 5 on PS4/Xbox One, and Red Dead Redemption 2. Rockstar is known for taking their time to make things perfect, so it’s quite likely they’ll give it another overhaul for GTA 6. This becomes even more interesting when you think about the advances in artificial intelligence (AI) and how they might influence the game.

opening scene of the GTA 6 trailer

According to thegamer.com, since Red Dead Redemption 2 came out, other games have tried to be the best. The Last of Us Part 2 stands out because it has tiny details, like Ellie checking for wounds. Tears of the Kingdom plays with physics and copied an ability from behind-the-scenes tools, showing they really get the game. Elden Ring is different but might be a new standard for open-world games, something GTA 6 needs to match. People still say Red Dead Redemption 2 has the best, most alive world. Also, there’s talk about how artificial intelligence (AI) is changing games, making them even more exciting.

It’s uncertain if RAGE can bring a city to life the way it did with the wilderness in Red Dead Redemption 2, and dealing with the hustle and bustle of GTA’s busy setting and faster gameplay will be an even bigger test. Yet, considering RAGE handled the demands of Red Dead Redemption 2 well, maybe a greater challenge is just what it needs. Grand Theft Auto 6 faces immense pressure to be the greatest game ever, or gamers might get really upset. However, thanks to RAGE, at least on a technical level, it might have the potential to meet those expectations.

Source link

Latest stories