# AI in gaming

> Source: https://aiwiki.ai/wiki/ai_in_gaming
> Updated: 2026-10-30
> Fact-checked: 2026-10-30
> Categories: AI Tools & Products, Artificial Intelligence
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "AI in gaming." aiwiki.ai, 30 Oct 2026. https://aiwiki.ai/wiki/ai_in_gaming
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**AI in gaming** refers to the use of [artificial intelligence](https://aiwiki.ai/wiki/artificial_intelligence) techniques to create intelligent behaviors, generate content, and enhance player experiences in video games. From the simple ghost behaviors in Pac-Man to modern [large language model](https://aiwiki.ai/wiki/large_language_model)-powered NPCs, game AI has evolved through decades of research and engineering. The field spans both traditional rule-based systems (finite state machines, behavior trees, pathfinding algorithms) and newer [machine learning](https://aiwiki.ai/wiki/machine_learning) approaches including [reinforcement learning](https://aiwiki.ai/wiki/reinforcement_learning) and [neural networks](https://aiwiki.ai/wiki/neural_network).

## History of game AI

The history of AI in video games stretches back to the earliest days of the medium. Each era introduced new techniques that expanded what game characters and systems could do.

### Game-playing machines before video games (1940s-1990s)

Games were a testbed for machine intelligence before video games existed. For the Westinghouse pavilion at the 1940 New York World's Fair, physicist Edward Condon designed the Nimatron, an electromechanical machine that played the mathematical game Nim [52]. Condon, Gereld L. Tawney and Willard A. Derr received U.S. patent 2,215,544 for a "Machine to play game of nim" on September 24, 1940 [51]. In 1959 Arthur Samuel of IBM published "Some Studies in Machine Learning Using the Game of Checkers," which reported that "a computer can be programmed so that it will learn to play a better game of checkers than can be played by the person who wrote the program" [53].

Later board-game programs combined search with learning. Gerald Tesauro's TD-Gammon learned backgammon through [temporal-difference learning](https://aiwiki.ai/wiki/temporal_difference_learning) from games against itself; in a 1995 paper Tesauro reported that backgammon expert Bill Robertie judged version 2.1 to play "at a strong master level that is extremely close (within a few hundredths of a point) to equaling the world's best human players" [55]. In May 1997, IBM's [Deep Blue](https://aiwiki.ai/wiki/deep_blue) beat world chess champion Garry Kasparov 3.5-2.5 in a six-game rematch in New York, which IBM describes as the first time a computer system defeated a reigning world chess champion in a match under standard tournament time controls; Kasparov had won their first match, in 1996, 4-2 [54]. The checkers program Chinook, from Jonathan Schaeffer's group at the University of Alberta, became the World Man-Machine Checkers Champion [57], and in 2007 the group reported in *Science* that checkers is solved: perfect play by both sides leads to a draw [56].

### Early arcade era (1970s-1980s)

The earliest video game AI consisted of simple, hardcoded patterns. Space Invaders (1978) used fixed movement patterns for alien formations. In **Pac-Man** (1980), developed by Namco and designed by Toru Iwatani, each of the four ghosts follows its own targeting rule while chasing Pac-Man. Blinky (red) targets Pac-Man's current tile; Pinky (pink) targets a tile four tiles ahead of Pac-Man in the direction he is moving; Inky (light blue) uses the most complex rule, which depends on both Pac-Man's position and direction and Blinky's position; and Clyde (orange) chases Pac-Man when he is more than eight tiles away but heads for his home corner when he is closer. All four ghosts also switch between chase, scatter, and frightened modes. These simple rules give the ghosts distinct "personalities" [1].

### Console and PC era (1990s-2000s)

As hardware improved, game AI became more sophisticated. Real-time strategy games such as StarCraft (1998) and Age of Empires II (1999) required AI opponents that could manage economies, build armies, and execute strategies.

In the 2000s, shooters became known for more sophisticated enemy AI. Bungie's Damian Isla described the AI of Halo 2 (2004) as "a hierarchical finite state machine (HFSM) or a behavior tree, or even more specifically, a behavior DAG (directed acyclic graph)," with a core of "on the order of 50 different behaviors" [4]. F.E.A.R. (2005) became known for enemies that take cover, blind fire, flush the player out with grenades and coordinate with squad members. Its characters planned sequences of actions with a planning system that Monolith Productions' Jeff Orkin called goal-oriented action planning (GOAP) [2].

### Modern era (2010s-present)

Middle-earth: Shadow of Mordor (2014) introduced the "Nemesis System," in which enemy orcs' rank, names and appearance change in response to the player's actions and the game remembers those actions. Warner Bros. received a U.S. patent covering the system, including features added in the 2017 sequel Shadow of War, in February 2021 [71].

## Traditional game AI techniques

Despite advances in machine learning, traditional AI techniques remain the backbone of game development because of their reliability, [interpretability](https://aiwiki.ai/wiki/interpretability), and low computational cost.

### Finite state machines (FSMs)

A finite state machine is the simplest and most widely used game AI technique. An NPC exists in one of several predefined states (such as "patrolling," "chasing," "attacking," or "fleeing") and transitions between them based on conditions like player proximity, health level, or environmental triggers. FSMs are easy to implement, debug, and understand, which is why they remain popular for simpler AI behaviors. Pac-Man's ghosts, which switch between chase, scatter, and frightened modes, are a classic example of FSM-driven AI [1][3]. Writing in 2006, F.E.A.R. developer Jeff Orkin said that nearly every game with any AI at all uses some form of FSM to control character behavior [2].

The main limitation of FSMs is that they become unwieldy as complexity grows. An NPC with dozens of possible states and transitions requires a tangled web of logic that is difficult to maintain.

### Behavior trees

Behavior trees address the scalability problems of FSMs by organizing AI logic into a hierarchical, modular tree structure. Each node in the tree represents an action, condition, or control flow element (such as a sequence or selector). The tree is evaluated from the root each frame, and the AI executes whichever branch of behavior is appropriate.

Halo 2's AI, described by Bungie's Damian Isla at GDC 2005, is an early documented example: Isla described its core as a behavior tree, or more precisely a behavior DAG, in which "decision schemes" such as prioritized lists and sequences choose among child behaviors [4]. Behavior trees later became a standard tool, and Unreal Engine includes a Behavior Tree asset for creating AI for non-player characters [70]. Their modularity makes them easy to extend: a designer can add new behaviors by attaching new branches without rewriting existing logic.

### A* pathfinding

The [A* algorithm](https://aiwiki.ai/wiki/a) is the most widely used pathfinding algorithm in games. Published by Peter Hart, Nils Nilsson, and Bertram Raphael in 1968, A* finds the shortest path between two points on a graph by combining the actual cost of reaching a node with a [heuristic](https://aiwiki.ai/wiki/heuristic) estimate of the remaining distance [5]. Orkin wrote in 2006 that A* and finite state machines would "undoubtedly" be the two most common answers if game developers were asked to list the AI techniques used in games, and that nearly every game with AI uses A* to plan paths [2]. Variants such as Jump Point Search and Hierarchical Pathfinding A* are also used.

Modern games typically precompute navigation meshes ("navmeshes") that represent walkable surfaces, and A* operates on these meshes to guide characters through complex 3D environments.

### Monte Carlo tree search (MCTS)

[Monte Carlo tree search](https://aiwiki.ai/wiki/monte_carlo_tree_search) is a heuristic search algorithm that uses random simulations ("rollouts") to evaluate potential moves in a game tree. At each step, MCTS selects a promising node, simulates random play from that point to a terminal state, and then backpropagates the result to update the node's value. Over many iterations, the algorithm converges on strong moves.

In a 2008 paper, Guillaume Chaslot and colleagues proposed MCTS as a framework for game AI that could be applied to classic board games such as Go, modern board games such as Settlers of Catan, and video games, with experiments in the Spring real-time strategy game [6]. Creative Assembly used MCTS in the campaign AI of the strategy game Total War: Rome II (2013) [68]. The technique gained worldwide attention as a core component of [DeepMind](https://aiwiki.ai/wiki/google_deepmind)'s [AlphaGo](https://aiwiki.ai/wiki/alphago) system, which combined MCTS with deep neural networks to defeat professional Go players [7].

### Utility AI and goal-oriented action planning

Beyond FSMs and behavior trees, two additional AI architectures are widely used in modern games:

**Utility AI** assigns numerical scores to possible actions based on current game state, then selects the highest-scoring action. For example, an NPC might evaluate "attack" (score: 0.7), "flee" (score: 0.3), and "heal" (score: 0.9), then choose to heal. The Sims franchise is one of the most prominent examples of utility-based AI: according to a chapter by David "Rez" Graham in the book *Game AI Pro*, motives such as hunger are scored through hand-tuned response curves, and actions are grouped into buckets by motive so that Sims always attempt to address their most urgent desire [69].

**Goal-oriented action planning (GOAP)** allows NPCs to define goals (such as "eliminate threat" or "find cover") and dynamically plan sequences of actions to achieve them. In F.E.A.R., the characters' finite state machine has only three states (Goto, Animate, and UseSmartObject), and a planner that Orkin compared to the academic STRIPS planning system uses A* to search for sequences of actions as well as paths. The resulting soldiers use cover tactically, coordinating with squad members to lay suppression fire while others advance [2].

## Machine learning in games

While traditional techniques still dominate commercial game AI, machine learning has produced some of the most celebrated achievements in AI research, using games as testbeds.

### AlphaGo, AlphaZero and MuZero

In October 2015, DeepMind's AlphaGo defeated the European Go champion Fan Hui by 5 games to 0 on a full-sized 19x19 board. The 2016 *Nature* paper describing the system called this "the first time that a computer program has defeated a human professional player in the full-sized game of Go, a feat previously thought to be at least a decade away" [7]. In March 2016, AlphaGo defeated Lee Sedol, one of the world's top Go players, 4-1 in Seoul; DeepMind says the match was watched by over 200 million people worldwide [47]. The system combined deep [convolutional neural networks](https://aiwiki.ai/wiki/convolutional_neural_network) with MCTS to evaluate board positions and select moves [7].

In December 2017, DeepMind described [AlphaZero](https://aiwiki.ai/wiki/alphazero), a more general system that learned to play chess, shogi, and Go entirely through self-play, starting with no domain knowledge beyond the rules. The December 2017 arXiv preprint said that AlphaZero "achieved within 24 hours a superhuman level of play" in all three games [48]. The peer-reviewed version, published in *Science* in December 2018, makes no 24-hour claim. It reports that, during training, AlphaZero first outperformed Stockfish in chess after 4 hours and Elmo in shogi after 2 hours, but needed 30 hours to outperform the version of AlphaGo that had beaten Lee Sedol, with Go training lasting about 13 days in total [8]. Both papers report that AlphaZero convincingly defeated a world-champion program in each game [8][48].

DeepMind's [MuZero](https://aiwiki.ai/wiki/muzero), first described in 2019 and published in *Nature* in December 2020, learned its own model of the game instead of being given the rules. It matched AlphaZero's superhuman performance in Go, chess, and shogi and set a new state of the art on a benchmark of 57 Atari games [62].

### AlphaStar (StarCraft II)

DeepMind's [AlphaStar](https://aiwiki.ai/wiki/alphastar) tackled StarCraft II, a real-time strategy game with imperfect information, a vast action space, and long-term strategic planning requirements. In test matches in December 2018, announced in January 2019, AlphaStar beat the professional players Dario "TLO" Wünsch and Grzegorz "MaNa" Komincz 5-0 each, in Protoss-versus-Protoss games on a single ladder map [49]. In October 2019, DeepMind reported in *Nature* that AlphaStar, playing in online games against human players on Battle.net, had been rated at Grandmaster level for all three StarCraft races and above 99.8% of officially ranked human players. The system used [supervised learning](https://aiwiki.ai/wiki/supervised_learning) from human replays and multi-agent reinforcement learning [9].

### OpenAI Five (Dota 2)

[OpenAI Five](https://aiwiki.ai/wiki/openai_five) was a system that played Dota 2, a complex multiplayer online battle arena game, controlling all five heroes of a team with replicas of a single neural network [10]. On April 13, 2019, at the OpenAI Five Finals event, it beat OG, the reigning Dota 2 world champions, 2-0 in a best-of-three match [10][50]. During OpenAI Five Arena, a public online event held from April 18 to 21, 2019, it played 7,257 games against 3,193 human teams and won 99.4% of them, although OpenAI counted games that human players abandoned as wins; these made up 3,140 of its 7,215 wins [10]. The system was trained using large-scale distributed reinforcement learning; OpenAI said the Finals version had experienced about 45,000 years of Dota self-play over 10 real-time months [50].

### Poker and Diplomacy

Games of hidden information and negotiation were later milestones. Libratus, developed by Noam Brown and Tuomas Sandholm at Carnegie Mellon University, defeated four top professionals at heads-up no-limit Texas hold'em in a 120,000-hand competition [58]; the 20-day "Brains Vs. Artificial Intelligence" event at Rivers Casino in Pittsburgh ended on January 30, 2017 [59]. Their successor [Pluribus](https://aiwiki.ai/wiki/pluribus), described in *Science* in 2019, was stronger than top human professionals in six-player no-limit Texas hold'em [60]. Meta AI's [CICERO](https://aiwiki.ai/wiki/cicero), described in *Science* in 2022, combined a language model with planning and reinforcement learning to play Diplomacy, a seven-player game built on negotiation. Across 40 games of an anonymous online Diplomacy league, it achieved more than double the average score of the human players and ranked in the top 10% of participants who played more than one game [61].

### Video game-playing agents

Researchers and hobbyists have also trained agents on video games, from classic consoles to modern racing simulators. Notable examples include:

| Game | AI approach | Achievement |
|---|---|---|
| Super Mario World | Neuroevolution (MarI/O by SethBling, 2015) | An evolutionary algorithm that started out knowing nothing about the game learned to complete one level after about a day of learning and 34 generations [66] |
| Atari 2600 games (49 titles) | [Deep Q-Networks](https://aiwiki.ai/wiki/dqn) (DQN) by DeepMind | Reached a level comparable to a professional human games tester across 49 games using the same algorithm, network architecture and hyperparameters (*Nature*, 2015) [63] |
| Atari 2600 games (57 titles) | [Agent57](https://aiwiki.ai/wiki/agent57) by DeepMind | First deep reinforcement learning agent to outperform the standard human benchmark on all 57 Atari games (2020) [64] |
| Minecraft | OpenAI VPT (Video PreTraining) | Learned from about 70,000 hours of unlabeled online video of people playing Minecraft; the authors reported the first computer agents able to craft diamond tools [65] |
| Gran Turismo Sport | GT Sophy by Sony AI | Won a head-to-head competition against four of the world's best Gran Turismo drivers (*Nature*, 2022) [22] |

GT Sophy was developed by Sony AI with Polyphony Digital and described in a 2022 *Nature* paper [22]. On November 1, 2023, Sony AI released Gran Turismo Sophy 2.0 to all Gran Turismo 7 players on PlayStation 5 in the game's Spec II update (patch 1.40), letting players of all levels race against it [67].

| AI system | Game | Developer | Year | Achievement |
|-----------|------|-----------|------|-------------|
| [Deep Blue](https://aiwiki.ai/wiki/deep_blue) | Chess | IBM | 1997 | Beat world champion Garry Kasparov 3.5-2.5 [54] |
| Chinook | Checkers | University of Alberta | 2007 | Checkers solved: perfect play leads to a draw [56] |
| [AlphaGo](https://aiwiki.ai/wiki/alphago) | Go | [DeepMind](https://aiwiki.ai/wiki/google_deepmind) | 2016 | Defeated Lee Sedol 4-1 [47] |
| Libratus | Heads-up no-limit Texas hold'em | Carnegie Mellon University | 2017 | Beat four top professionals over 120,000 hands [58] |
| [AlphaZero](https://aiwiki.ai/wiki/alphazero) | Chess, Shogi, Go | [DeepMind](https://aiwiki.ai/wiki/google_deepmind) | 2017-2018 | Superhuman play in three games from self-play; the 2017 preprint claimed "within 24 hours," while the 2018 *Science* paper reports 30 hours to pass AlphaGo Lee in Go [8][48] |
| [AlphaStar](https://aiwiki.ai/wiki/alphastar) | StarCraft II | [DeepMind](https://aiwiki.ai/wiki/google_deepmind) | 2019 | Grandmaster level, above 99.8% of officially ranked players [9] |
| [OpenAI Five](https://aiwiki.ai/wiki/openai_five) | Dota 2 | [OpenAI](https://aiwiki.ai/wiki/openai) | 2019 | Defeated world champions OG 2-0 [10] |
| [Pluribus](https://aiwiki.ai/wiki/pluribus) | Six-player no-limit Texas hold'em | Carnegie Mellon University and Facebook AI | 2019 | Stronger than top human professionals [60] |
| [Agent57](https://aiwiki.ai/wiki/agent57) | 57 Atari games | [DeepMind](https://aiwiki.ai/wiki/google_deepmind) | 2020 | Above the human benchmark on all 57 games [64] |
| [MuZero](https://aiwiki.ai/wiki/muzero) | Go, chess, shogi, Atari | [DeepMind](https://aiwiki.ai/wiki/google_deepmind) | 2019-2020 | Matched AlphaZero without being given the rules [62] |
| GT Sophy | Gran Turismo Sport | Sony AI | 2022 | Beat four of the world's best Gran Turismo drivers [22] |
| [CICERO](https://aiwiki.ai/wiki/cicero) | Diplomacy | Meta AI | 2022 | More than double the average human score in 40 online games [61] |

## Modern AI applications in game development

Beyond playing games, AI is also used to build them, at several stages of development.

| Application area | Description | Example tools and implementations |
|-----------------|-------------|-----------------------------------|
| Procedural content generation | Algorithmic creation of levels, maps, quests, items, and entire game worlds | No Man's Sky (more than 18 quintillion possible planets) [23], Minecraft (terrain generated as players explore) [104] |
| NPC behavior | Dynamic, adaptive character behaviors that respond to player actions | Halo 2 (behavior trees) [4], Shadow of Mordor (Nemesis System) [71] |
| Playtesting and QA | Automated bots that explore games to find bugs, balance issues, and exploits | Client and server bots for Tom Clancy's The Division [75], EA SEED's reinforcement-learning test agents [24] |
| Game design assistance | AI tools that help developers plan features, set up scenes and generate placeholder assets | Unity's Agentic Assistant for Unity 6, in open beta from May 2026 [44] |
| Difficulty adjustment | Systems that change the challenge during play based on player performance | Left 4 Dead's AI Director (2008) [25]; Resident Evil 4 (2005), whose reported adaptive difficulty Capcom had not confirmed as of 2015 [76] |
| Narrative generation | AI-generated dialogue, storylines, and quest content | AI Dungeon, whose October 2026 model list includes [DeepSeek](https://aiwiki.ai/wiki/deepseek) and [GLM](https://aiwiki.ai/wiki/glm) models [98]; the dialogue and narrative toolset Xbox and Inworld AI announced in November 2023 [79] |
| Voice acting and synthesis | AI-generated voice lines for characters using [text-to-speech](https://aiwiki.ai/wiki/text_to_speech_ai) models | [ElevenLabs](https://aiwiki.ai/wiki/elevenlabs) voices for Fortnite's Darth Vader and Fortnite creator NPCs [89][91], NVIDIA ACE speech integration [13] |
| Art and asset generation | AI tools for creating textures, concept art, 3D models, and environments | Scenario (scenario.com) for game assets [99]; AI-generated placeholder images that shipped in Clair Obscur: Expedition 33 and were later patched out [102] |
| Music and sound design | AI-generated adaptive soundtracks and sound effects | AIVA [100] |

In the Game Developers Conference's 2025 State of the Game Industry survey of more than 3,000 developers and industry professionals, 52% said they worked at companies that had implemented [generative AI](https://aiwiki.ai/wiki/generative_ai), and 36% said they used it personally [11]. The 2026 survey's figures are summarized under "AI-assisted development (2025-2026)" below.

### Procedural content generation in detail

Procedural generation predates machine learning in games. Rogue, created in the early 1980s by Michael Toy and Glenn Wichman, randomized its dungeons on every play, which co-creator Wichman credited for its popularity [72]. Elite (1984) procedurally generated its universe, including planetary positions, names, politics, and descriptions; the technique could have produced far more galaxies, but publisher Acornsoft limited the game to eight galaxies of 256 planets each [73]. Researchers have also explored procedural content generation via machine learning (PCGML), which a 2018 survey defined as "the generation of game content using machine learning models trained on existing content" and described as a relatively new paradigm, in contrast with search-based, solver-based, and constructive methods. The survey focused on functional content such as platformer levels, game maps, interactive fiction stories, and cards in collectible card games [74].

| Game | What is generated | Technique | Scale |
|---|---|---|---|
| No Man's Sky (2016-present) | Planets, flora, fauna, terrain | Deterministic algorithms with seed-based generation | 18,446,744,073,709,551,616 possible planets [23] |
| Minecraft (2011-present) | Terrain | 2D, then 3D Perlin noise, as creator Markus "Notch" Persson described it in 2011 [104] | Generated in chunks as players explore; Persson wrote that the maps are "not infinite, but there's no hard limit either" [104] |
| Hades (2020) | The order of combat chambers in each run | Randomized assembly of designer-built chambers [105] | Smaller chambers tend to appear early in a region and larger ones later, according to designer Ed Gorinstein [105] |
| Spelunky 2 (2020) | Level layouts | A "safe" path to the exit built from room templates, as in the first Spelunky, according to Derek Yu [106] | Randomly generated levels [106] |
| Dwarf Fortress (2006-present) | Worlds with civilizations, histories, creatures and cultures | Simulation-based world generation | Its Steam page advertises the "generated rise and fall of civilizations, personalities, creatures, cultures" [107] |

### AI playtesting and quality assurance

Studios and researchers have applied AI to automated game testing and quality assurance. Examples include:

- **Tom Clancy's The Division:** Massive Entertainment and Ubisoft Reflections built two kinds of test bots for the franchise. Server bots run natively on the server to stress-test it, while client bots take control of a game client and play by the same rules as a human player [75].
- **EA SEED** researchers reported in a paper at the 2020 IEEE Conference on Games that [agents](https://aiwiki.ai/wiki/ai_agents) trained with deep reinforcement learning could increase test coverage, find exploits, test map difficulty, and detect common problems in first-person shooter games [24].

### Dynamic difficulty adjustment

Dynamic difficulty adjustment (DDA) changes a game's difficulty during play based on how the player is performing. Robin Hunicke and Vernell Chapman designed Hamlet, a DDA system built on Valve's Half-Life engine, to keep the player in the "flow channel" of M. Csikszentmihalyi's flow model, away from states where the game is far too challenging or too easy [97]. Commercial examples include:

| Game | DDA mechanism | How it works |
|---|---|---|
| Left 4 Dead (2008) | "AI Director" | Estimates each survivor's "emotional intensity" and alternates periods of full and minimal enemy populations; also chooses which designer-placed weapon caches and item groups appear [25] |
| Resident Evil 4 (2005) | Hidden adaptive difficulty | Reported to raise enemy damage and aggression when the player is progressing easily; as of Engadget's 2015 report, Capcom had never confirmed the feature existed [76] |
| Crash Bandicoot 2 and 3; N. Sane Trilogy (2017) | Dynamic difficulty adjustment | Gives the player extra Aku Aku masks and checkpoints after a certain number of failures in a level [77] |
| Forza Motorsport 5 (2013) | Drivatar AI | AI opponents learn to drive like real people from players' driving data [78] |

Left 4 Dead's AI Director is one of the best-known DDA systems in commercial gaming. In a 2009 talk, Valve's Michael Booth explained that the Director raises each survivor's intensity value when the survivor is injured, incapacitated, or pulled off a ledge, or when an enemy dies nearby. It builds up a full threat population until intensity peaks, sustains it for a few seconds, and then switches to a minimal population for a 30-45 second "Relax" period, producing peaks and valleys of intensity modeled on the pacing of Counter-Strike [25].

## AI-assisted development (2025-2026)

In the 2026 State of the Game Industry survey, 36% of game industry professionals said they used generative AI tools as part of their job, but only 30% of respondents at game studios did, compared with 58% at publishers, support companies, and marketing or PR firms. The most-used tools were [large language model](https://aiwiki.ai/wiki/large_language_model) chatbots: [ChatGPT](https://aiwiki.ai/wiki/chatgpt) (74%), Google Gemini (37%), and Microsoft Copilot (22%). The most common uses were research or brainstorming (81%), everyday tasks such as writing emails and code assistance (47% each), and prototyping (35%) [21].

Game engines have also gained agent features. In May 2026 Unity released a suite of AI tools in open beta for developers on Unity 6 or later. Its core feature was an "Agentic Assistant" that works inside a project, with a "Plan" mode for designing implementation steps, "Skills" for areas such as UI creation and scene setup, placeholder-asset generation, and a [Model Context Protocol](https://aiwiki.ai/wiki/model_context_protocol) server that lets external coding tools and language models connect to the editor [44]. Community projects apply the same approach to asset tools. [Blender MCP](https://aiwiki.ai/wiki/blender_mcp), for example, is a community plugin "to control Blender 3D with any LLM of your choice" [45]. Individual developers have also built games by directing general-purpose [coding agents](https://aiwiki.ai/wiki/coding_agent), sometimes in the iterative style called [vibe coding](https://aiwiki.ai/wiki/vibe_coding). [Thariq Shihipar](https://aiwiki.ai/wiki/thariq_shihipar), who describes himself as formerly of [Claude Code](https://aiwiki.ai/wiki/claude_code) at Anthropic, wrote in September 2026 that he was prototyping a video game as a side project and advised that "the process of making a game is deeply satisfying and creative, don't outsource that to AI. Use AI to work with you to bring your vision to life" [46]. The [AI game development](https://aiwiki.ai/wiki/ai_game_development) overview covers these workflows, tools, and their limits in detail, and [Developing Games with AI](https://aiwiki.ai/wiki/developing_games_with_ai) is a step-by-step how-to guide.

## AI NPCs and conversational characters

Language models have also been used to build NPCs that converse with players in natural language, remember earlier interactions and keep consistent personalities.

### Generative Agents

A research precursor to these products is "Generative Agents: Interactive Simulacra of Human Behavior," posted to arXiv in April 2023 (arXiv:2304.03442) by Joon Sung Park, Michael Bernstein, Percy Liang and colleagues from Stanford University and Google, and presented at the ACM UIST conference in October 2023. The authors extended a large language model with an architecture that stores a complete record of each agent's experiences in natural language, synthesizes those memories into higher-level reflections, and retrieves them to plan behavior. They populated a sandbox environment inspired by The Sims with 25 agents. Starting from a single user-specified idea that one agent wanted to throw a Valentine's Day party, the agents spread invitations over the next two days, asked each other out on dates to the party, and coordinated to show up together [85].

### Inworld AI

[Inworld AI](https://aiwiki.ai/wiki/inworld_ai) developed a character engine and language-model technology for AI-powered NPCs. It has since retired its legacy Character Studio; its documentation says Studio's workflows moved to the Inworld Portal and the Inworld Agent Runtime, which provides Character templates for Unity and Unreal Engine [86]. As of 30 October 2026, Inworld's website lists text-to-speech, speech-to-text and speech-to-speech models and an inference service, with games and media as one market [103]. Ubisoft's NEO NPC prototype used Inworld's Character Engine and LLM technology to give each character a background, knowledge base, and conversational style [12]. In November 2023, Xbox announced a multi-year partnership with Inworld to build "AI game dialogue & narrative tools at scale," including an AI design copilot that turns prompts into scripts, dialogue trees and quests, and an AI character runtime engine that can be integrated into a game client [79].

### Convai

Convai provides similar AI NPC capabilities, with a focus on voice-based interactions and real-time character animation. It offers a Unity SDK and an Unreal Engine plugin [80]. NVIDIA used Convai's framework in a version of its Kairos tech demo shown at CES 2024, in which NPCs could converse among themselves, pick up and deliver objects, and lead players to objectives [13].

### NVIDIA ACE

NVIDIA's Avatar Cloud Engine (ACE) is a suite of AI microservices for creating lifelike digital characters. In the pipeline NVIDIA describes, Riva automatic speech recognition transcribes what a player says, a large language model handles the [natural language processing](https://aiwiki.ai/wiki/natural_language_processing) and writes a text response, Riva text-to-speech voices it, and Audio2Face syncs facial expressions to the dialogue [13]. NVIDIA introduced production ACE microservices on January 8, 2024, starting with Audio2Face, which generates facial animation from audio, and Riva automatic speech recognition; it named Charisma.AI, Convai, Inworld, miHoYo, NetEase Games, Ourpalm, Tencent, Ubisoft and UneeQ as developers embracing ACE [81]. At CES 2025, NVIDIA announced ACE "autonomous game characters." KRAFTON's PUBG Ally co-playable character uses the Mistral-Nemo-Minitron-8B-128k-instruct small language model to communicate with game-specific lingo, give strategic advice, find and share loot, drive vehicles, and fight, and NetEase announced a locally run AI teammate built with ACE for NARAKA: BLADEPOINT [82].

KRAFTON opened PUBG Ally to players in a two-week global beta of an "Ally Duo" arcade mode in PUBG: BATTLEGROUNDS from June 17 to July 1, 2026, in which players team up with an AI character named Ella and communicate with her by voice in Korean, Chinese, or English [83].

### Ubisoft NEO NPCs

Ubisoft unveiled its NEO NPC prototype at GDC in March 2024. It was developed by an R&D team at Ubisoft's Paris studio with Inworld AI and NVIDIA: Inworld's technology supplied the characters' dialogue, and NVIDIA's Audio2Face provided real-time facial animation [12][14]. The demo's narrative team created two NEO NPCs, Bloom and Iron, each with its own background story, knowledge base, and conversational style [84]. Players had to talk with the NPCs through voice-to-chat to reach their objectives, and Ubisoft said the characters could react in real time, unlock new story elements, or generate new quests on the fly [12].

### Conversational NPC examples

Unlike traditional NPCs, which KRAFTON describes as reacting "based on fixed conditions" [83], the characters in these released games generate replies to what players type or say:

| Game | Developer | Released | What the sources document |
|---|---|---|---|
| Fortnite | Epic Games | Darth Vader event, May 2025 | Vader imitated James Earl Jones's voice, with his estate's permission, using Google's [Gemini 2.0 Flash](https://aiwiki.ai/wiki/gemini_2_0_flash) and ElevenLabs Flash v2.5 [89]. After players got it to say slurs, Epic pushed a hotfix and said its filters "did not catch a specific variation of an expletive" [90]. As of October 2026, Epic documents creator-built persona NPCs that use Gemini 3.1 Flash-Lite and ElevenLabs' Eleven v3 and respond only in English [91] |
| Where Winds Meet | Everstone Studio (NetEase Games) | November 2025 (worldwide) | Some NPCs chat through generative AI; one player won over an NPC who was searching for her two brothers by typing a bracketed narration in which the brothers suddenly appear [92] |
| Whispers from the Star | Anuttacon | August 14, 2025 | Players talk with a stranded astronaut, Stella; the Steam disclosure describes "fully voiced, AI-powered conversations and performances created in collaboration with an actor" [93] |

## Text-to-game generation

A newer frontier in game AI is the generation of entire playable environments from text or image prompts.

### Google Genie

Google DeepMind's Genie models are [world models](https://aiwiki.ai/wiki/world_model) that generate interactive environments from text or image prompts. The original Genie, described in a February 2024 paper, was trained on a curated dataset of more than 30,000 hours of 2D platformer gameplay video [28]. [Genie 3](https://aiwiki.ai/wiki/genie_3), announced in August 2025, generates worlds that users can navigate in real time at 24 frames per second and 720p resolution, staying consistent for a few minutes [15]. Unlike video generation models such as [Sora](https://aiwiki.ai/wiki/sora) or [Veo](https://aiwiki.ai/wiki/veo) that produce passive clips, Genie creates environments that respond to user actions in real time.

On January 29, 2026, Google opened Project Genie, an experimental web prototype powered by Genie 3, Google's Nano Banana Pro image model, and [Gemini](https://aiwiki.ai/wiki/gemini), to Google AI Ultra subscribers in the United States aged 18 and over [26]. Ultra was priced at $249.99 a month when Google introduced it in May 2025 [29]. Google limited generations to 60 seconds, and some Genie 3 capabilities, such as promptable events that change the world during exploration, were not included at launch [26]. 9to5Google reported that the prototype's worlds ran at 720p and 20-24 frames per second [27]. Shares of several game companies fell the day after the launch, with Unity hit hardest; Unity CEO Matt Bromberg responded that world-model outputs were "unsuitable on their own for games that require consistent, repeatable player experiences" [33].

[Genie 4](https://aiwiki.ai/wiki/genie_4), announced on October 27, 2026, generates ambient sound, music, and interactive dialogue alongside the visuals, which DeepMind describes as a first for its real-time world models. It was released as a research preview for trusted testers, with access to follow for Google AI Ultra subscribers in Project Genie and early access partners of the Gemini API and Google AI Studio. DeepMind said game developers among its testers were using it to go "from an initial idea to an explorable prototype in seconds" before "writing a single line of engine code" [30].

### GameNGen

GameNGen, from researchers at Google Research, Google DeepMind, and Tel Aviv University, was introduced in August 2024 in a paper titled "Diffusion Models Are Real-Time Game Engines." Its authors describe it as "the first game engine powered entirely by a neural model that enables real-time interaction with a complex environment over long trajectories at high quality." It can interactively simulate the classic game DOOM at over 20 frames per second on a single TPU. The system is based on [Stable Diffusion](https://aiwiki.ai/wiki/stable_diffusion) v1.4, modified to generate each frame of gameplay from previous frames and an action input rather than from text prompts [16].

### Prompt-to-game platforms

Other 2026 products turn text prompts into playable games rather than streaming frames from a world model. On October 7, 2026, Google launched [Playground](https://aiwiki.ai/wiki/playground_google_labs), an experimental Google Labs platform on which users aged 18 and over in the United States describe a game in text prompts, then play it and share it without coding experience [31]. The same day, Google and Unity announced a strategic partnership under which they would introduce [Unity Spark](https://aiwiki.ai/wiki/unity_spark) later in 2026, an expanded creation experience connected to Playground that adds "professional-level mechanics, high-fidelity 3D capabilities" and the Unity runtime [32]. Google described Spark as in testing with a closed beta coming soon [31].

## AI art in game development

Studios have used generative image tools for concept exploration and placeholder art, and disclosures of that use have drawn backlash. Larian Studios drew criticism after CEO Swen Vincke discussed the studio's experiments with AI in a Bloomberg interview, in terms that seemed to include concept art; in a Reddit AMA on January 9, 2026, he said Divinity would contain no generative AI art and that Larian would refrain from using such tools during concept art development [101]. AI-generated placeholder images that shipped with Clair Obscur: Expedition 33 in April 2025 were patched out after release, and Sandfall Interactive called them the product of "brief" experimentation in 2022; in December 2025 the Indie Game Awards, whose rules make games developed using generative AI ineligible, withdrew the game's Debut Game and Indie Game of the Year awards [102]. Scenario markets image, video, 3D and audio generation as "AI for creative production and game assets" [99].

In the 2026 State of the Game Industry survey, workers in visual and technical art held the most unfavorable views of generative AI of any discipline, with 64% saying it was having a negative impact on the industry [21].

## Controversy and ethical concerns

### Voice actor displacement

AI-generated voices have drawn criticism and disputes in several games:

- **Arc Raiders (2025-2026):** When Embark Studios' Arc Raiders launched in October 2025, many in-game voice lines, such as those heard when pinging a location, were created using AI text-to-speech built from audio provided by actors who were paid a licensing fee. The use of AI voices drew criticism, and in March 2026 Embark CEO Patrick Söderlund said the studio had re-recorded some lines with human actors after launch, saying "a real professional actor is better than AI" [17]. Junghun Lee, CEO of publisher Nexon, said in a November 2025 interview that "it's important to assume that every game company is now using AI" [34].
- **Apex Legends (2025):** In February 2025 Pascale Chemin, the French voice of Wraith in Apex Legends, said a publisher had added a contract clause that would allow actors' recordings to be used to train generative AI, and that 31 other voice actors on the project had refused to sign it. She did not name the game or the publisher, but press coverage identified them as Apex Legends and EA [18][35].
- **Tomb Raider IV-VI Remastered (2025):** An August 2025 patch for Aspyr's remaster collection added French voice lines that, according to Françoise Cadol, the French voice actress of Lara Croft, had been generated with AI from her voice; she sent Aspyr a cease-and-desist letter [36][37]. Aspyr later said the patch "contained unauthorized generative AI voiceover content" added by an external partner without its knowledge or approval, released an update removing all AI voiceover content, and thanked Cadol and the Tomb Raider community for raising the issue [19][37].

### The SAG-AFTRA video game strike

On July 26, 2024, SAG-AFTRA began a strike against major video game publishers [38]; according to NBC News, about 2,600 performers who do voice-over and motion-capture work were on strike [94]. Protections against the use of AI were the central dispute [20][38]; the union said the previous Interactive Media Agreement had no AI protections [94].

The strike lasted nearly 11 months. On June 9, 2025, SAG-AFTRA announced a tentative agreement, and it suspended the strike on June 11 [38]. Members ratified the 2025 Interactive Media Agreement on July 9, 2025, by a vote of 95.04% to 4.96% [20][38]. The contract included compounded increases in performer compensation of 15.17% upon ratification, plus additional 3% increases in November 2025, 2026, and 2027. On AI, the agreement established consent and disclosure requirements for AI digital replica use and gave performers the ability to suspend consent for the generation of new material during a strike [20].

### AI-generated game content

Beyond voice acting, objections to [AI-generated content](https://aiwiki.ai/wiki/ai_generated_content) in games come from several disciplines: in the 2026 GDC survey, 63% of respondents in game design and narrative and 59% in game programming said generative AI was having a negative impact on the industry [21]. Respondents to the 2025 survey named intellectual property theft, energy consumption, lower quality in AI-generated content, potential bias in AI programs and regulatory issues as the main reasons for their discontent [11]. The [AI in art](https://aiwiki.ai/wiki/ai_in_art) article covers the related debate over image generators.

Developer sentiment has turned more negative over time. In the GDC State of the Game Industry surveys, the share of respondents who said generative AI was having a negative impact on the industry rose from 18% in the 2024 survey to 30% in 2025 and 52% in 2026, while the share who saw a positive impact fell from 13% to about 7% between the 2025 and 2026 surveys [21]. The 2026 survey of students found that 74% were concerned about their job prospects in the industry, citing the lack of entry-level jobs, competition from laid-off workers, and AI-led displacement [21].

### Store disclosure rules

In January 2024 Valve added an AI section to the content survey that developers complete when submitting a game to Steam. It separates "Pre-Generated" content made with AI tools during development from "Live-Generated" content created while the game runs, and Valve said much of each disclosure would appear on the game's store page [39]. In January 2026 Valve revised the form's wording to say that "efficiency gains" from AI-powered development tools are "not the focus" of the section, which instead concerns AI used to create content "that ships with your game, and is consumed by players" [40]. Epic Games CEO Tim Sweeney had argued in November 2025 that AI tags make "no sense for game stores, where AI will be involved in nearly all future production" [43], and on October 23, 2026, he wrote that "forcing developers to write a political statement to be used against them by the hate mob they're helping organize is abhorrent" [41][42]. The [Steam AI disclosure](https://aiwiki.ai/wiki/steam_ai_disclosure) article covers the policy's history.

## Current state (2025-2026)

Characters driven by language models appeared in released games in 2025 and 2026, often as experimental or time-limited features. KRAFTON's inZOI entered early access in March 2025 with an experimental, RTX-only "Smart Zoi" setting built on NVIDIA ACE [87][88]; Epic Games added a voice-interactive Darth Vader to Fortnite in May 2025 [89]; NetEase's Where Winds Meet launched worldwide in November 2025 with generative AI chat for some NPCs [92]; and KRAFTON ran a two-week beta of its PUBG Ally companion from June 17 to July 1, 2026 [83]. Generative world models and prompt-to-game tools began reaching consumers in 2026: Project Genie opened in January and Google launched Playground in October, while Unity Spark was announced as Playground's more advanced tier and DeepMind previewed Genie 4 to trusted testers at the end of October [26][30][31][32]. All four remained experimental or limited in access as of 30 October 2026.

Adoption in development is uneven. In GDC's 2025 survey, 52% of respondents worked at companies that had implemented generative AI [11]; in the 2026 survey, 36% of industry professionals, and 30% of those at game studios, said they used generative AI tools in their own work, mostly language models for research, brainstorming, everyday tasks, and code assistance [21]. Engine vendors such as Unity have added agentic assistants to their editors [44], and AI test bots, such as those built for Tom Clancy's The Division, are described under [AI playtesting and quality assurance](#ai-playtesting-and-quality-assurance) [75][24]. Procedural generation research has added language models to the PCGML methods described above: MarioGPT, a fine-tuned GPT-2 model described in 2023, generates Super Mario Bros. levels from text prompts [95], and a survey presented at AIIDE 2024 wrote that the arrival of large language models "truly disrupted the trajectory of PCG advancement" [96].

On the labor side, the Interactive Media Agreement that SAG-AFTRA members ratified in July 2025 added consent and disclosure requirements for AI digital replicas [20]. Developer sentiment has grown more negative: in the 2026 GDC survey, 52% of respondents said generative AI was having a negative impact on the industry [21]. Disclosure is also contested, with Valve revising Steam's AI disclosure form in January 2026 and Epic Games CEO Tim Sweeney describing mandatory disclosure statements as "abhorrent" in October 2026 [40][41].

## See also

- [AI game development](https://aiwiki.ai/wiki/ai_game_development)
- [Developing Games with AI](https://aiwiki.ai/wiki/developing_games_with_ai)
- [Adding Generative AI Features to Games](https://aiwiki.ai/wiki/adding_generative_ai_features_to_games)
- [Text-to-image models](https://aiwiki.ai/wiki/text-to-image_models)

## References

1. Pittman, Jamey. "The Pac-Man Dossier." Gamasutra (Game Developer), February 23, 2009. https://www.gamedeveloper.com/design/the-pac-man-dossier
2. Orkin, Jeff. "Three States and a Plan: The A.I. of F.E.A.R." Game Developers Conference, 2006. https://www.gamedevs.org/uploads/three-states-plan-ai-of-fear.pdf
3. "Finite State Machine in Game Development." ResearchGate. https://www.researchgate.net/publication/355518086_Finite_State_Machine_in_Game_Development
4. Isla, Damian. "GDC 2005 Proceeding: Handling Complexity in the Halo 2 AI." Gamasutra (Game Developer), March 11, 2005. https://www.gamedeveloper.com/programming/gdc-2005-proceeding-handling-complexity-in-the-i-halo-2-i-ai
5. Hart, Peter E.; Nilsson, Nils J.; Raphael, Bertram. "A Formal Basis for the Heuristic Determination of Minimum Cost Paths." *IEEE Transactions on Systems Science and Cybernetics* 4(2): 100-107, 1968. https://doi.org/10.1109/TSSC.1968.300136
6. Chaslot, Guillaume; Bakkes, Sander; Szita, Istvan; Spronck, Pieter. "Monte-Carlo Tree Search: A New Framework for Game AI." *Proceedings of the Fourth Artificial Intelligence and Interactive Digital Entertainment Conference* (AIIDE), 2008. https://sander.landofsand.com/publications/AIIDE08_Chaslot.pdf
7. Silver, David; Huang, Aja; et al. "Mastering the game of Go with deep neural networks and tree search." *Nature* 529: 484-489, January 27, 2016. https://doi.org/10.1038/nature16961
8. Silver, David; Hubert, Thomas; Schrittwieser, Julian; et al. "A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play." *Science* 362(6419): 1140-1144, December 7, 2018. https://doi.org/10.1126/science.aar6404 (open-access author version: https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/alphazero-shedding-new-light-on-chess-shogi-and-go/alphazero_preprint.pdf)
9. Vinyals, Oriol; Babuschkin, Igor; Czarnecki, Wojciech M.; et al. "Grandmaster level in StarCraft II using multi-agent reinforcement learning." *Nature* 575: 350-354, October 30, 2019. https://doi.org/10.1038/s41586-019-1724-z
10. OpenAI: Berner, Christopher; et al. "Dota 2 with Large Scale Deep Reinforcement Learning." arXiv:1912.06680, December 2019. https://arxiv.org/abs/1912.06680
11. Elderkin, Beth. "GDC 2025 State of the Game Industry: Devs weigh in on layoffs, AI, and more." GDC, January 21, 2025. https://gdconf.com/article/gdc-2025-state-of-the-game-industry-devs-weigh-in-on-layoffs-ai-and-more/
12. "Ubisoft Unveils 'NEO NPCs', its First Prototype For GenAI-powered Characters and Gameplay." Ubisoft press release, March 19, 2024. https://staticctf.ubisoft.com/8aefmxkxpxwl/Mw2s4KjssknqHHh1VIf8V/720296810ecc3deae51778a219732c7f/PRESS_RELEASE_GDC_UbisoftUnveilsNEONPC_190324.pdf
13. "Coming Up ACEs: Decoding the AI Technology That's Enhancing Games." NVIDIA Blog. https://blogs.nvidia.com/blog/ai-decoded-ace-microservices-digital-humans/
14. "How Ubisoft's New Generative AI Prototype Changes the Narrative for NPCs." Ubisoft News. https://news.ubisoft.com/en-us/article/5qXdxhshJBXoanFZApdG3L/how-ubisofts-new-generative-ai-prototype-changes-the-narrative-for-npcs
15. "Genie 3: A new frontier for world models." Google DeepMind. https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/
16. Valevski, Dani; Leviathan, Yaniv; Arar, Moab; Fruchter, Shlomi. "Diffusion Models Are Real-Time Game Engines" (GameNGen). Project page, https://gamengen.github.io/; arXiv:2408.14837, August 2024. https://arxiv.org/abs/2408.14837
17. "Arc Raiders Replacing AI-Generated Voices With Human Actors." Kotaku. https://kotaku.com/arc-raiders-replaced-ai-generated-content-human-recorded-dialogue-voices-2000678774
18. "Apex Legends AI Voice Contract Sparks French Voice Cast Protest." OECD.AI. https://oecd.ai/en/incidents/2025-02-27-037e
19. "Aspyr Pulls Unauthorized AI Voice from Tomb Raider Remaster Amid Backlash." WebProNews. https://www.webpronews.com/aspyr-pulls-unauthorized-ai-voice-from-tomb-raider-remaster-amid-backlash/
20. "Video Game Actors Contract Ratified." Variety. https://variety.com/2025/gaming/news/video-game-actors-strike-contract-ratified-sag-aftra-1236451291/
21. Elderkin, Beth. "GDC 2026 State of the Game Industry Reveals Impact of Layoffs, Generative AI, and More." GDC Festival of Gaming, 2026. https://gdconf.com/article/gdc-2026-state-of-the-game-industry-reveals-impact-of-layoffs-generative-ai-and-more/
22. Wurman, Peter R.; Barrett, Samuel; Kawamoto, Kenta; et al. "Outracing champion Gran Turismo drivers with deep reinforcement learning." *Nature* 602: 223-228, February 9, 2022. https://doi.org/10.1038/s41586-021-04357-7
23. "Most explorable planets in a videogame." Guinness World Records. https://www.guinnessworldrecords.com/world-records/413048-most-explorable-planets-in-a-videogame
24. Bergdahl, Joakim; Gordillo, Camilo; Tollmar, Konrad; Gisslén, Linus. "Augmenting Automated Game Testing with Deep Reinforcement Learning." arXiv:2103.15819, March 2021 (SEED, Electronic Arts). https://arxiv.org/abs/2103.15819
25. Booth, Michael. "The AI Systems of Left 4 Dead." Keynote, Artificial Intelligence and Interactive Digital Entertainment Conference (AIIDE), Valve, 2009. https://steamcdn-a.akamaihd.net/apps/valve/2009/ai_systems_of_l4d_mike_booth.pdf
26. "Project Genie: AI world model now available for Ultra users in U.S." Google, January 29, 2026. https://blog.google/innovation-and-ai/models-and-research/google-deepmind/project-genie/
27. Li, Abner. "Google rolling out 'Project Genie' to generate playable worlds." 9to5Google, January 29, 2026. https://9to5google.com/2026/01/29/google-project-genie/
28. Bruce, Jake; et al. "Genie: Generative Interactive Environments." arXiv:2402.15391, February 2024. https://arxiv.org/abs/2402.15391
29. "Google AI Ultra: You'll have to pay $249.99 per month for Google's best AI." TechCrunch, May 20, 2025. https://techcrunch.com/2025/05/20/google-ai-ultra-youll-have-to-pay-249-99-per-month-for-googles-best-ai/
30. Hadsell, Raia; Rivas, Diego; Qi, Hang. "Genie 4: A new generation of world simulation." Google DeepMind, October 27, 2026. https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/
31. Karimzadehgan, Maryam. "Introducing Playground: Create and play custom games." Google, October 7, 2026. https://blog.google/innovation-and-ai/technology/ai/playground-experimental-gaming-platform/
32. "Google and Unity Partner on New AI Gaming Platform for the Next Era of Interactive Entertainment." Unity Technologies press release, October 7, 2026. https://unity.com/news/google-and-unity-partner-on-new-ai-gaming-platform-for-the-next-era-of-interactive-entertainment
33. "Unity boss Bromberg moves to calm markets as Google's Genie causes gaming stock slide." Mobilegamer.biz, February 2, 2026. https://mobilegamer.biz/unity-boss-bromberg-moves-to-calm-markets
34. Amber V. "'It's important to assume every game company is now using AI.' Nexon CEO comments following Arc Raiders launch." Automaton, November 11, 2025. https://automaton-media.com/en/news/its-important-to-assume-every-game-company-is-now-using-ai-nexon-ceo-comments-on-role-of-human-creativity-as-ai-tools-become-the-norm/
35. "French Apex Legends voice cast goes to war over AI." PC Games Insider, February 28, 2025. https://www.pcgamesinsider.biz/news/75006/french-apex-legends-voice-cast-goes-to-war-over-ai/
36. Herbig, Daniel. "'Tomb Raider Remastered': French voice of Lara Croft created with AI." heise online, September 11, 2025. https://www.heise.de/en/news/Tomb-Raider-Remastered-French-voice-of-Lara-Croft-created-with-AI-10641331.html
37. "A Message to the Tomb Raider Community." Aspyr Support. https://support.aspyr.com/hc/en-us/articles/40068858882189-A-Message-to-the-Tomb-Raider-Community
38. "2024-2025 SAG-AFTRA video game strike." Wikipedia. https://en.wikipedia.org/wiki/2024%E2%80%932025_SAG-AFTRA_video_game_strike
39. "AI Content on Steam." Steamworks Development, Valve, January 2024. https://store.steampowered.com/news/group/4145017/view/3862463747997849618
40. Squires-Hand, Liam. "Valve tweak Steam AI disclosure form for developers to clarify it's for content consumed by players." GamingOnLinux, January 17, 2026. https://www.gamingonlinux.com/2026/01/valve-tweak-steam-ai-disclosure-form-for-developers-to-clarify-its-for-content-consumed-by-players/
41. Sweeney, Tim (@TimSweeneyEpic). Post on X, October 23, 2026. https://x.com/TimSweeneyEpic/status/2113639717690638483
42. Walker, John. "Epic Boss Tim Sweeney Calls Steam's AI Declarations 'Political' And 'Abhorrent'." Kotaku, October 26, 2026. https://kotaku.com/epic-steam-valve-tim-sweeney-gen-ai-declaration-political-abhorrent-2000749706
43. Sweeney, Tim (@TimSweeneyEpic). Post on X, November 26, 2025. https://x.com/TimSweeneyEpic/status/1993687499621728312
44. McEvoy, Sophie. "Unity launches suite of AI tools to help creators accelerate game development." GamesIndustry.biz, May 7, 2026. https://www.gamesindustry.biz/unity-launches-suite-of-ai-tools-to-help-creators-accelerate-game-development
45. "mcp-for-blender." GitHub repository, ahujasid. https://github.com/ahujasid/mcp-for-blender
46. Shihipar, Thariq (@trq212). Post on X, September 30, 2026. https://x.com/trq212/status/2105333496768319969
47. "AlphaGo." Google DeepMind. https://deepmind.google/research/alphago/
48. Silver, David; Hubert, Thomas; Schrittwieser, Julian; et al. "Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm." arXiv:1712.01815, December 5, 2017. https://arxiv.org/abs/1712.01815
49. "AlphaStar: Mastering the real-time strategy game StarCraft II." Google DeepMind, January 24, 2019. https://deepmind.google/discover/blog/alphastar-mastering-the-real-time-strategy-game-starcraft-ii/
50. "OpenAI Five Defeats Dota 2 World Champions." OpenAI, April 15, 2019. https://openai.com/index/openai-five-defeats-dota-2-world-champions/
51. Condon, Edward U.; Tawney, Gereld L.; Derr, Willard A. "Machine to play game of nim." U.S. Patent 2,215,544, filed April 26, 1940, issued September 24, 1940. https://patents.google.com/patent/US2215544A/en
52. "NIMATRON: An Early Electromechanical Machine to Play the Game of Nim." HistoryofInformation.com. https://historyofinformation.com/detail.php?id=4012
53. Samuel, Arthur L. "Some Studies in Machine Learning Using the Game of Checkers." *IBM Journal of Research and Development* 3(3): 210-229, July 1959. https://doi.org/10.1147/rd.33.0210
54. "Deep Blue." IBM Heritage. https://www.ibm.com/history/deep-blue
55. Tesauro, Gerald. "Temporal Difference Learning and TD-Gammon." *Communications of the ACM* 38(3): 58-68, March 1995. https://doi.org/10.1145/203330.203343 (reprint: https://www.bkgm.com/articles/tesauro/tdl.html)
56. Schaeffer, Jonathan; Burch, Neil; Björnsson, Yngvi; et al. "Checkers Is Solved." *Science* 317(5844): 1518-1522, September 14, 2007. https://doi.org/10.1126/science.1144079
57. "Chinook: World Man-Machine Checkers Champion." University of Alberta, Department of Computing Science. https://webdocs.cs.ualberta.ca/~chinook/ (archived January 10, 2024: https://web.archive.org/web/20240110090453/http://webdocs.cs.ualberta.ca/~chinook/)
58. Brown, Noam; Sandholm, Tuomas. "Superhuman AI for heads-up no-limit poker: Libratus beats top professionals." *Science* 359(6374): 418-424, 2018. https://doi.org/10.1126/science.aao1733
59. Spice, Byron. "Carnegie Mellon Artificial Intelligence Beats Top Poker Pros." Carnegie Mellon University, January 31, 2017. https://www.cmu.edu/news/stories/archives/2017/january/AI-beats-poker-pros.html
60. Brown, Noam; Sandholm, Tuomas. "Superhuman AI for multiplayer poker." *Science* 365(6456): 885-890, 2019. https://doi.org/10.1126/science.aay2400
61. Meta Fundamental AI Research Diplomacy Team (FAIR); Bakhtin, Anton; Brown, Noam; et al. "Human-level play in the game of Diplomacy by combining language models with strategic reasoning." *Science* 378(6624): 1067-1074, 2022. https://doi.org/10.1126/science.ade9097
62. Schrittwieser, Julian; Antonoglou, Ioannis; Hubert, Thomas; et al. "Mastering Atari, Go, chess and shogi by planning with a learned model." *Nature* 588: 604-609, December 23, 2020. https://doi.org/10.1038/s41586-020-03051-4
63. Mnih, Volodymyr; Kavukcuoglu, Koray; Silver, David; et al. "Human-level control through deep reinforcement learning." *Nature* 518: 529-533, February 2015. https://doi.org/10.1038/nature14236
64. Badia, Adrià Puigdomènech; Piot, Bilal; Kapturowski, Steven; et al. "Agent57: Outperforming the Atari Human Benchmark." arXiv:2003.13350, March 2020. https://arxiv.org/abs/2003.13350
65. Baker, Bowen; Akkaya, Ilge; Zhokhov, Peter; et al. "Video PreTraining (VPT): Learning to Act by Watching Unlabeled Online Videos." arXiv:2206.11795, June 2022. https://arxiv.org/abs/2206.11795
66. Dorrier, Jason. "Watch This Open Source AI Learn to Dominate Super Mario World in Just 24 Hours." Singularity Hub, June 18, 2015. https://singularityhub.com/2015/06/18/watch-this-open-source-ai-learn-to-dominate-super-mario-world-in-just-24-hours/
67. "Sony AI Announces Full Global Release of AI Driver Gran Turismo Sophy for Gran Turismo 7 Players in Latest Update." Sony AI, November 1, 2023. https://ai.sony/articles/sonyai022/
68. Thompson, Tommy. "Revolutionary Warfare | The AI of Total War (Part 3)." Game Developer, February 12, 2018. https://www.gamedeveloper.com/design/revolutionary-warfare-the-ai-of-total-war-part-3-
69. Graham, David "Rez." "An Introduction to Utility Theory." In *Game AI Pro*, ed. Steve Rabin, chapter 9. CRC Press, 2013. http://www.gameaipro.com/GameAIPro/GameAIPro_Chapter09_An_Introduction_to_Utility_Theory.pdf
70. "Behavior Trees in Unreal Engine." Unreal Engine documentation, Epic Games. https://dev.epicgames.com/documentation/en-us/unreal-engine/behavior-trees-in-unreal-engine
71. Wallace, Chris. "Warner Bros secures patent for the Nemesis System from Shadow of Mordor." MCV/Develop, February 8, 2021. https://www.mcvuk.com/business-news/warner-bros-secures-patent-for-the-nemesis-system-from-shadow-of-mordor/
72. Loguidice, Bill; Barton, Matt. "The History of Rogue: Have @ You, You Deadly Zs." Gamasutra (Game Developer), May 5, 2009. https://www.gamedeveloper.com/design/the-history-of-rogue-have-you-you-deadly-zs
73. "Feature: The History of Elite." Gamasutra (Game Developer), April 7, 2009. https://www.gamedeveloper.com/game-platforms/feature-the-history-of-i-elite-i-
74. Summerville, Adam; Snodgrass, Sam; Guzdial, Matthew; et al. "Procedural Content Generation via Machine Learning (PCGML)." *IEEE Transactions on Games* 10(3): 257-270, 2018. arXiv:1702.00539. https://arxiv.org/abs/1702.00539
75. Thompson, Tommy. "The Secret AI Testers inside Tom Clancy's The Division." Game Developer, March 4, 2020. https://www.gamedeveloper.com/design/the-secret-ai-testers-inside-tom-clancy-s-the-division
76. Summers, Nick. "'Resident Evil 4' secretly adjusted its difficulty for you." Engadget, June 3, 2015. https://www.engadget.com/2015-06-03-resident-evil-4-difficulty.html
77. Brash, Mario. "Why Crash Bandicoot N. Sane Trilogy Is So Difficult, According to Devs." COGconnected, July 19, 2017. https://cogconnected.com/2017/07/crash-bandicoot-difficulty-vv/
78. Lowe, Mike. "Forza 5 for Xbox One: 'Drivatar' cloud-sourced AI learns how to drive like real people." Pocket-lint, June 10, 2013. https://www.pocket-lint.com/games/news/xbox/121606-forza-5-for-xbox-one-drivatar-cloud-sourced-ai-learns-how-to-drive-like-real-people
79. "Xbox and Inworld AI partner to empower game creators with the potential of Generative AI." Microsoft Game Dev, November 2023. https://developer.microsoft.com/en-us/games/articles/2023/11/xbox-and-inworld-ai-partnership-announcement/
80. "Welcome." Convai Documentation. https://docs.convai.com/api-docs
81. "NVIDIA and Developers Pioneer Lifelike Digital Characters for Games and Applications With NVIDIA Avatar Cloud Engine." NVIDIA Newsroom, January 8, 2024. https://nvidianews.nvidia.com/news/ace-avatar-cloud-engine-microservices
82. Burnes, Andrew; Schneider, Seth; Paul, Jason. "NVIDIA Redefines Game AI With ACE Autonomous Game Characters." NVIDIA GeForce News, January 6, 2025. https://www.nvidia.com/en-us/geforce/news/nvidia-ace-autonomous-ai-companions-pubg-naraka-bladepoint/
83. "KRAFTON Introduces PUBG Ally Beta Test." KRAFTON press release via Game Developer, June 16, 2026. https://www.gamedeveloper.com/press-release/krafton-introduces-pubg-ally-beta-test
84. "NVIDIA Digital Human Technologies Bring AI Game Characters To Life." NVIDIA GeForce News, March 19, 2024. https://www.nvidia.com/en-gb/geforce/news/nvidia-ace-gdc-gtc-2024-ai-character-game-and-app-demo-videos/
85. Park, Joon Sung; O'Brien, Joseph C.; Cai, Carrie J.; Morris, Meredith Ringel; Liang, Percy; Bernstein, Michael S. "Generative Agents: Interactive Simulacra of Human Behavior." *Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology* (UIST '23), October 2023. arXiv:2304.03442. https://arxiv.org/abs/2304.03442
86. "Character Studio deprecation." Inworld AI documentation, updated October 26, 2026; accessed October 30, 2026. https://docs.inworld.ai/character-studio-deprecation
87. Burnes, Andrew. "NVIDIA ACE Autonomous Game Characters Debut This Month In inZOI & NARAKA: BLADEPOINT MOBILE PC VERSION." NVIDIA GeForce News, March 13, 2025. https://www.nvidia.com/en-us/geforce/news/nvidia-ace-naraka-bladepoint-inzoi-launch-this-month/
88. "inZOI." Steam store page, inZOI Studio and KRAFTON (early access release date). Accessed October 30, 2026. https://store.steampowered.com/app/2456740/
89. Kerr, Chris. "Voice actor union strikes back after Fortnite debuts controversial AI-powered Darth Vader." Game Developer, May 20, 2025. https://www.gamedeveloper.com/business/voice-actor-union-hits-back-after-fortnite-debuts-controversial-ai-powered-darth-vader
90. Nelson, Jason. "'Fortnite' Fixes AI-Powered Darth Vader After It Starts Saying Slurs." Decrypt, May 16, 2025. https://decrypt.co/320500/fortnite-ai-powered-darth-vader-saying-slurs
91. "LLM Conversations in Unreal Editor for Fortnite." Fortnite documentation, Epic Games. Accessed October 30, 2026. https://dev.epicgames.com/documentation/en-us/fortnite/llm-conversations-in-unreal-editor-for-fortnite
92. Fujiwara, Hideaki. [Where Winds Meet: a trick for easily fooling its free-conversation AI NPCs] (in Japanese). AUTOMATON, November 28, 2025. https://automaton-media.com/articles/newsjp/where-winds-meet-20251128-368771/
93. "Whispers from the Star." Steam store page, Anuttacon (AI Generated Content Disclosure). Accessed October 30, 2026. https://store.steampowered.com/app/3730100/
94. Rosenblatt, Kalhan. "Video game actors' strike ends after tentative deal is reached." NBC News, June 11, 2025. https://www.nbcnews.com/tech/tech-news/video-game-actors-strike-ends-sag-aftra-tentative-deal-reached-rcna177738
95. Sudhakaran, Shyam; González-Duque, Miguel; Glanois, Claire; et al. "MarioGPT: Open-Ended Text2Level Generation through Large Language Models." arXiv:2302.05981, February 2023. https://arxiv.org/abs/2302.05981
96. Maleki, Mahdi Farrokhi; Zhao, Richard. "Procedural Content Generation in Games: A Survey with Insights on Emerging LLM Integration." AIIDE-24, November 2024. arXiv:2410.15644. https://arxiv.org/abs/2410.15644
97. Hunicke, Robin; Chapman, Vernell. "AI for Dynamic Difficulty Adjustment in Games." Northwestern University. https://users.cs.northwestern.edu/~hunicke/pubs/Hamlet.pdf
98. "AI Models and their Differences." AI Dungeon Guidebook, Latitude. Accessed October 30, 2026. https://help.aidungeon.com/ai-model-differences
99. "Scenario: AI for Creative Production." Scenario. Archived October 17, 2026. https://web.archive.org/web/20261017133615/https://www.scenario.com/
100. "AIVA, the AI Music Generation Assistant." AIVA. Accessed October 30, 2026. https://www.aiva.ai/
101. Madsen, Hayes. "Larian Says It Won't Use Generative AI For Divinity Concept Art." CGMagazine, January 9, 2026. https://www.cgmagonline.com/news/larian-generative-ai-divinity-reddit
102. Masters, Catherine. "Clair Obscur: Expedition 33 loses Indie Game Awards due to gen AI." The A.V. Club, December 22, 2025. https://www.avclub.com/clair-obscur-genai-iga-awards-rescinded
103. "Inworld AI." Company website. Accessed October 30, 2026. https://inworld.ai/
104. Persson, Markus. "Terrain generation, Part 1." The Word of Notch (blog), March 9, 2011. Archived: https://web.archive.org/web/20120129194417/http://notch.tumblr.com/post/3746989361/terrain-generation-part-1
105. Grayson, Nathan. "Hades' Level Design Is Less Random Than It Seems." Kotaku, October 2, 2020. https://kotaku.com/hades-level-design-is-less-random-than-it-seems-1845254545
106. "First look at Spelunky 2 gameplay, plus new Derek Yu interview." PlayStation Blog, August 29, 2018. https://blog.playstation.com/archive/2018/08/29/first-look-at-spelunky-2-gameplay-plus-new-derek-yu-interview
107. "Dwarf Fortress." Steam store page, Bay 12 Games and Kitfox Games. Accessed October 30, 2026. https://store.steampowered.com/app/975370/

