adminGenerative AI in video games has become a source of both excitement and concern across the industry. Dr Luke Dicken, Take-Two’s former head of AI, believes the current hype could overshadow more established AI techniques that may help developers make better games. His perspective draws on years of work exploring how technology can adapt game experiences to players.
Image credit: Valentin Lacoste on Unsplash
Take-Two laid off its AI team as part of a restructuring initiative. The move drew attention because the company had previously spoken positively about generative AI. But the team’s work began before tools such as ChatGPT brought generative AI into the public spotlight, and its broader focus was on applying AI to game development.
What generative AI in video games can learn from tabletop RPGs
The group was established at Zynga in 2019 as a research and development team. Zynga was acquired by Take-Two in 2022. Rather than concentrating on generative AI, the team explored a wider range of AI methods and how they might support games.
Dicken’s interest in the field is tied to the role of a human game master in tabletop role-playing games such as Dungeons & Dragons. A game master reads the group, considers what players want from the experience and adjusts the story or challenge accordingly. Dicken sees this as a useful model for AI: technology could help a game respond to different play styles without replacing the design principles that make the experience engaging.
For developers, the practical takeaway is to start with a clear player need. An AI system is most useful when it supports a specific design goal, rather than being added simply because the technology is popular.
Personalising a game through player behaviour
Dicken’s team tested ways to tailor games using machine-learning profiles built from around 40 measures of how someone played. The idea was to identify preferences and adapt elements of a game to suit them. For example, a game might change its visual theme while keeping its underlying structure and mechanics.
The team released Spell Forest in 2020, using the project to test whether adapting a player’s experience could affect business measures. The work helped bring the group into other teams’ projects at Zynga. This type of personalisation is different from generating new text or images: it relies on analysing player behaviour and adjusting a designed experience.
Spell Forest was released in 2020. Image credit: Zynga
How ChatGPT changed the conversation
After ChatGPT became publicly available in 2022, businesses faced new questions about generative AI tools, including how user inputs might be handled and whether they could become training data. Zynga’s AI team was asked to help govern the company’s use of these tools and educate staff about their risks.
The team had 25 members, but only three or four were focused on generative AI. The rest continued working across AI more broadly. Later, the group also oversaw technology across Take-Two. Its governance responsibilities were eventually transferred to other teams as part of the company’s restructuring.
Dicken describes generative AI as a highly polarising subject. He argues that companies need to assess individual uses rather than treating every application as equally acceptable or unacceptable. A useful first step for studios is to define who will review a tool, what information can be entered into it and what conditions must be met before it is used in production.
Ethical, technical and business concerns
Dicken raises ethical concerns about the material used to train large language models and image-generation systems. He points to legal proceedings that brought attention to claims about artists’ work being used to train models without their knowledge. He says that he knows artists and writers whose work may have been used in this way.
He also questions whether these systems reliably produce high-quality work. Large language models predict likely next words based on patterns in their training data. In Dicken’s view, this can lead to conventional or average-sounding results. He cautions that such tools may help someone with limited coding experience produce workable code, but could also encourage skilled developers to settle for less distinctive solutions.
Reliability is another consideration. A model or its training can change, and even a small adjustment may affect how it performs. A tool that once met a studio’s needs may not behave the same way later. Teams evaluating a system should therefore test it regularly, document the version and settings used, and keep human review in place for consequential work.
Dicken also questions the business case for the wider AI boom. Alongside questions of ethics and legality, studios need to consider whether a tool delivers enough value to justify its costs and risks. These issues can vary by project, company size and use case, so a clear assessment is more helpful than adopting a tool simply to follow a trend.
Why AI hype could affect traditional game technology
Generative AI has encouraged more people to consider how technology might improve games. That increased openness can create opportunities to revisit established techniques, including approaches to procedural or level-content generation that were previously dismissed.
However, Dicken worries that inflated expectations could create disappointment if generative AI fails to deliver what its supporters promise. If the technology’s reputation suffers, studios might also become less willing to explore other forms of AI that could provide practical benefits. This is why he distinguishes generative AI from the wider field of artificial intelligence: they are not interchangeable terms, and one technology’s limitations should not decide the value of all the others.
When assessing a proposed AI feature, developers can ask whether it improves the player experience, whether the output meets quality standards and whether the team can maintain control over the system. Those questions help keep the discussion focused on a game’s needs rather than the hype cycle.
How game studios can make informed decisions
Dicken says the choices around generative AI depend on context. A small studio facing serious financial pressure may weigh potential efficiency differently from a large publisher with complex legal and governance obligations. But in either case, decision-makers should consider three connected areas:
- Ethics: Consider the origins of training data and the effect a tool may have on creators and staff.
- Legal risk: Review relevant rights, contracts, privacy requirements and company policies before adopting a tool.
- Business value: Test whether the system solves a real problem and whether its ongoing costs and risks are justified.
These considerations may point in different directions, so studios need to decide which principles guide their work. A practical policy can set boundaries for approved uses, explain what information employees may share with external tools and outline how generated material will be checked.
Conclusion: a balanced view of generative AI in video games
Generative AI in video games may have useful applications, but its benefits should be judged separately from the excitement surrounding it. Dicken’s experience highlights the value of player-focused design, careful testing and clear standards for ethical, legal and commercial decisions. Studios that distinguish generative tools from other AI methods can assess each approach on its merits—and avoid letting hype determine what belongs in a game.
