adminAI in the games industry is not a new development, but the rapid rise of generative tools has made it harder to separate practical applications from inflated promises. The technology can help with specific production challenges, yet it also raises questions about quality, creative work, data rights and player trust. A more useful conversation starts by examining what each tool is designed to do—and whether it does that job responsibly.
Image credit: Jay Rembert
Dr Tommy Thompson is the founder and director of AI and Games. He previously worked as an AI researcher and university lecturer in game development. In 2024, he co-founded Game AI Events CIC, a non-profit organisation that hosts events for game AI professionals.
This article is part of AI Week.
AI in the games industry: separating hype from practical use
For more than 20 years, Thompson has worked around artificial intelligence and games. He describes the current debate as unusually difficult to navigate: AI has long offered useful ways to improve games and their development, but public discussion often treats very different technologies as if they were the same thing.
Since GPT-4 launched in 2023, generative AI has been promoted as a way to make development faster and less expensive, create more reactive characters, and tailor stories to individual players. Those possibilities are appealing to studios under pressure to control costs. But projected benefits should be weighed against real limitations, including unreliable outputs, intellectual property disputes, energy use and the risk of low-quality material reaching players.
Thompson also points to the broader effects of investment in AI infrastructure. Increased demand for memory and computing resources has contributed to supply pressures and higher costs. Valve has said rising memory prices led it to review the Steam Machine’s shipping schedule and pricing; Thompson questions whether the system could launch below $1,000. That figure is his expressed doubt, not a confirmed price.
Image credit: Valve
In Thompson’s view, generative AI companies often present uncertain tools as dependable solutions, despite the challenges of making them consistent, affordable and suitable for real production. Systems such as Google DeepMind’s Genie 3 represent notable progress in world models, but he argues that their cost, limited fidelity and difficulty maintaining consistency make them poorly suited to many current game-development needs. He sees potential in more focused efforts, including Microsoft’s Muse, while cautioning against claims that such systems will transform game creation overnight.
A long history of game AI
AI has been part of games since early non-player characters and procedural generation techniques in the 1980s. Techniques developed during the late 1990s and early 2000s helped shape NPC behaviour in games such as Quake III Arena, Half-Life, The Sims, Halo 2 and F.E.A.R.
Machine learning has also found valuable uses outside visible gameplay. Xbox Live’s TrueSkill matchmaking system, which launched with Halo 2, is one example. Motion Matching, pioneered in Hitman Absolution in 2012, uses AI to support animation and is now a feature of Unreal Engine 5.
Image credit: IO Interactive
Studios also apply machine learning to player analytics, automated testing, content workflows and community safety. The examples discussed by Thompson include:
- King’s tools for content generation, automated testing and player profiling.
- Machine-learning bots used by Ubisoft to stress test updates for Rainbow Six: Siege and For Honor.
- EA’s use of machine-learning bots for quality-assurance tasks in Battlefield, NHL 26 and EA Sports FC.
- Machine-learning systems used by Square Enix to test RPG balance at scale.
- Tools for detecting cheating and toxicity through text chat, voice chat and gameplay signals.
These applications illustrate a practical point: a tool can be valuable without generating story, art or dialogue. For studios assessing AI, it is useful to define the production problem first, then check whether a system improves accuracy, saves time or supports quality without creating unacceptable costs or risks.
Generative AI is not a simple yes-or-no question
The games industry is facing layoffs, market uncertainty and barriers for people trying to enter the profession. Against that backdrop, generative AI is often presented as a way to reduce hiring, lower production costs and bring products to market sooner. Thompson argues that neither blanket enthusiasm nor blanket rejection offers a sufficient answer.
Generative AI is a branch of machine learning that produces content from patterns learned in data. Its impact depends on how a system is designed, trained and used. A narrowly defined task, clear data choices and human review can make a tool more appropriate for a particular workflow. Those safeguards do not remove every concern, but they help studios evaluate a system on its actual purpose rather than on broad promises.
Image credit: CD Projekt Red
Some examples involve supporting existing creative and technical work. DLSS uses generative techniques to upscale images and has helped enable ports including Cyberpunk 2077 and Star Wars Outlaws on Switch 2. It has also been used to reduce texture sizes in God of War: Ragnarok on PlayStation 5. The team behind Mass Effect Legendary Edition used an earlier form of generative AI for an initial texture-upscaling pass, with artists refining the results by hand.
Other examples include EA’s AgentMerge tool for consolidating duplicate JIRA tickets, Infinity Ward’s multimodal asset search engine and Twitch’s use of large language models to improve toxicity detection. Studios such as BitPart, Lingotion, Meaning Machine, Raw Power Labs and Studio Atelico are exploring gameplay experiences using ethically developed, fine-tuned models, often running on users’ devices rather than in the cloud.
Thompson maintains that generative AI does not match human quality in artistic work. He argues, however, that the debate should also consider whether a specific application solves a genuine problem, respects the people whose work may be affected and delivers a result that players value.
Evaluate both the gains and the costs
Responsible adoption requires more than announcing that a studio is using AI. Developers, publishers and industry organisations need to share evidence about successful uses as well as failures. Teams should assess a tool in realistic production conditions, including its output quality, reliability, operating costs, data provenance and the amount of human review it requires.
It is also important to consider who bears the consequences. Artists, designers, writers, voice actors, performers, community managers and programmers may worry that automation will replace their work or weaken their livelihoods. A production plan should openly address those concerns rather than treating staff impact as an afterthought.
For teams considering an AI system, practical questions include:
- What specific task is the tool meant to improve?
- What data was used to build or train it, and are there rights or consent concerns?
- How will people check, correct or reject its output?
- Does the tool meet quality and performance needs without creating hidden costs?
- Will players understand where AI is used, and does that use improve their experience?
Thompson co-founded a conference in 2024 to encourage more informed discussion and responsible practice. He also argues that players’ responses matter: if AI features fail to offer clear value, studios may face stronger consumer resistance. In his view, the industry should communicate honestly about how these systems are used and avoid allowing hype or cost-cutting alone to determine their role.
A more responsible future for AI in games
AI has already contributed to matchmaking, animation, testing, analytics and accessibility across games and their production pipelines. Generative AI adds new possibilities, but it does not remove the need for skilled people, careful evaluation or accountability. The technology is most useful when it addresses a defined need and its limitations are understood before it reaches a production workflow.
The future of AI in the games industry will depend on more than technical capability. Studios must weigh gains against costs, protect creative work, involve employees in decisions and show players how a feature benefits them. A clear, evidence-based approach can help the industry make room for useful tools without treating AI as a silver bullet.
