The relationship between games and AI: Mechanisms, history and constant evolution

• AI Guys CTO Ken Noland was among the first to speak on stage at PGC London 2026.
• He presented "a history of intelligence", comparing game mechanisms to generative AI.

AI Guys CTO Ken Noland appeared on stage during Pocket Gamer Connects London last month to explore the shared technology between artificial intelligence and the games industry.

He spoke from a technical perspective, having begun with a definition of modern generative AI as a state machine that operates over a probabilistic space. This means large models are "just a bunch of numbers and vectors".

"It has an optimisation factor which is introduced through feedback mechanisms," Noland added. "Every single token that’s generated is searching through a massive data set, running through various algorithms and generating the next token."

He highlighted the four core "primitives" that every generative AI has as state, action, memory and feedback, all of which "have been prevalent in games for ages".

AI in retrospect

"Games matter because we are developing worlds that AI can work within. We’re developing constraints, we’re developing worlds that they can explore, we’re developing experiences that AI can use to better understand the actual world around them. And we’re doing it in a way that allows the AI to actually be smarter about the decisions it’s making," Noland said.

He suggested generative AI’s four key elements have all been present in games as far back as Pac-Man, at which time the state aspect was "critical" because ghosts changed behaviour as the player moved around collecting nuggets.

He highlighted chess games as an example where the AI needs to look forward, thinking multiple moves ahead to anticipate outcomes. This sense of planning has long given players the feeling of playing against an intelligent opponent.

From there, Noland proceeded to real-time strategy games.He suggested the genre has "always had a core AI problem", facing numerous difficulties on the technical side. Between pathfinding, real-time feedback, and tactical and strategic layers, there were various challenges to overcome.

Pathfinding improved considerably in the 2000s, Noland said, and he noted that real-time feedback especially - having to constantly evaluate the state of a world and plan its next move - is "exactly the same" as generative AI.

"It’s constantly evaluating the tokens based on its context."

And, Noland’s talk went on to cover procedural worlds - comparing their generative environments to the modern use of AI, and defining rules like the probabilistic sense of a river appearing in a jungle versus a desert, for example.

"At its core, it’s the same kind of mechanisms under the hood for procedurally generated content as it is for generative models, generative environments," he said.

Learning behaviours and neural nets aren’t new technologies either, Noland added.

Finally, he predicted that the next frontiers will be embodiment, simulation and performance.

Pocket Gamer Connects San Francisco is taking place on March 9th, 2026.