King is leveraging artificial intelligence to accelerate the creation of new levels for Candy Crush
The integration of AI in enhancing game development is a familiar territory for King.
For more than five years, the company behind Candy Crush has been at the forefront of utilizing AI and machine learning technologies, which predates its 2022 acquisition of the Swedish AI software firm Peltarion.
King's AI Labs, overseen by Sahar Asadi, has developed AI tools already being utilized in various games, including Candy Crush, profoundly impacting the development process of these popular titles.
"The playtesting bot we've created enables designers to gauge the game experience before releasing a level," says Asadi. "It allows them to determine if a level meets the desired experience and refine it if necessary.
"Additionally, we've added an enhancement feature. If designers identify levels that miss the mark, the bot can automatically make adjustments based on set criteria. The best solutions generated by the bot are then reviewed by the designers."
"The creative aspect of designing levels is enjoyable for designers, but refining and testing levels repetitively can be tedious. The playtesting bot significantly reduces this manual labor, granting designers more time for innovation."
The concept of computers playing games is not novel. However, the AI developed by King aims to emulate rather than outperform humans, necessitating a distinct approach.
"Crucially, our playtesting aims to be human-like. Unlike DeepMind's AlphaGo, which aimed to defeat a Go master, our goal is to mimic real players. The bot assesses the game scenario and opts for the move most likely chosen by a human," remarks Asadi.
This emulation is based on extensive data collected from player behavior.
"We have data on millions of game states and their corresponding player actions. The bot learns this pattern and can predict a human-like response to new situations. It may not always choose the optimal move, but it reflects typical human decisions."
With access to vast amounts of data, the bot estimates level difficulty and player challenges, consistently aligning with human gameplay patterns. Ongoing development focuses on integrating player skills and preferences for even more realistic emulation.
According to Asadi, the playtesting bot has cut down manual level adjustments by 95%, accelerating overall level tweaks by 50%. Beyond speed, it ensures release quality, monitoring factors like playability and challenge appropriateness.
"The work dynamics will evolve, and the skills required daily will shift."
Concerns naturally arise about job security for designers as AI tools become more effective. However, Asadi maintains that designers remain indispensable.
"Designers are crucial," asserts Asadi. "We view these tools as co-pilots, offering gameplay insights pre-release. Ultimately, designers determine fun and overall experience, deciding based on these insights whether to release or refine levels further."
She emphasizes, "Qualities like fun and good gameplay can't be mathematically defined, making designers' roles irreplaceable."
Asadi argues that these tools are born from the collaborative efforts of designers, eager to streamline mundane tasks and focus on innovation.
"Progressing from paper and pencil to tools like Photoshop, and now to advanced UI interfaces, designers continually adapt. This new toolset empowers them to focus on their core creative skills, potentially resulting in more exciting and engaging content."
Even as AI assumes more repetitive tasks, Asadi acknowledges that such tasks often serve as entry points for newcomers in the game industry.
"Work processes will evolve, skill requirements will shift. We are continually building new technologies and products, changing the roles of everyone, including mine," she concludes.
"Thanks to machine learning, many tasks that engineers once needed to code manually are no longer necessary. When I interview candidates nowadays, the criteria have changed from what they were two years ago. However, a thorough understanding of machine learning remains essential."
"For other companies, AI is progressing rapidly, but how should they integrate it in terms of technology, culture, and deriving value?"
Looking ahead, King's AI Labs is exploring how to better understand what players want at different times, enhancing their gaming experience.
"For example, if I’m on a bus and have five minutes, I want to resume my game level quickly," she elaborates. "But if I’m at home with half an hour to spare, I might prefer to engage in various quests. Understanding this context and delivering joy through gameplay is something that could significantly benefit players. Part of our research involves using foundational models and the latest in machine learning to better interpret player behavior and integrate this insight into the game, thereby enhancing the overall experience."
AI is evolving swiftly, and within a year or two, King's current research could be reflected in their games. This rapid progress is due to early investments in AI. For other game studios, seizing AI opportunities will hinge on their preparedness to leverage this technology.
"The landscape of AI development is shifting quickly," Asadi observes. "It’s an exhilarating time because it opens up new avenues for innovation, keeping my team and me constantly engaged. These advancements present substantial opportunities to incorporate research into games. Crucial to this process is the readiness of games to adopt these technologies. At King, we have the advantage of having started early with internal research and the acquisition of Peltarion, providing us with significant expertise. Our strong connection with the game facilitates smooth integration of AI solutions, enabling us to move swiftly."
"Other companies need to carefully consider this. AI is advancing rapidly, but integrating it technically, culturally, and in terms of value is essential. Frequently, my team works on projects that make it into games within one or two years. I hope the new solutions we’re discussing can be implemented even faster."
