A duo of King product and research directors spoke on the Game Developers Conference today about how AI has transformed automated level creation in Candy Crush Saga.
King has had an interesting journey in implementing and leveraging AI to create business value; Investments are made in every thing from A/B testing to high-speed AI level testing to enhance the gaming experience.
Back in September, Activision Blizzard's King (now owned by Microsoft) announced that Candy Crush Saga had to this point generated $20 billion in revenue and five billion downloads since its debut on mobile devices in 2012. This is the equivalent of a wrecking ball swinging through the video game industry. In 2016, King only had 2,000 levels for the sport.
Sahar Asadi has been Director of AI Labs at King for five years and leads a team focused on applied research. They've been exploring add value to the corporate by exploring longer-term strategic initiatives and conducting applied research, working very closely with core technology organizations but in addition very closely with game developers, she said in an interview with GamesBeat.
As Director of AI Labs, she is chargeable for defining an AI research roadmap; Developing and integrating AI use cases that would ultimately have multiple business impacts. Asadi has a real interest in appearing at and co-founding industry conferences.
Prior to joining King, Sahar held plenty of key positions including data scientist for Spotify, research scientist in natural language processing and data retrieval at Meltwater, and lead machine learning engineer at Clusterone at two startups, where he was a Distributed deep learning platform built learning and dealing on mobile product discovery.
And one in all the core project areas Asadi began working on when she joined King was gameplay testing, in addition to level management and automation using AI. This is essential work considering that King has maintained such strong user interest in King for greater than a decade.
“Our players spend most of their time playing levels and it is key to supply them with a superb gaming experience,” she said. “We began developing playtest bots to envision and understand the standard of levels before release, and likewise developed tools to refine the degrees. It was a really close collaboration with Games.”
Asadi gave the talk alongside Anna Hernandelius, Product Director of Candy Crush Saga. Their goal was to elucidate the iterative process that enables designers to spend less time on mundane tasks and more time creating levels.
“This is especially vital for 2 reasons. One is that levels are a central art of the player experience. The second is that this game that we’ve got in Candy Crush Saga alone has greater than 16,000 levels and the variety of levels is growing,” Asaid said. “This scale is essential with a view to maintain quality, but in addition with a view to generally give you the option to create an increasing number of levels and to do justice to the usual or quality that we’ve got had to this point.”
“We will discuss how AI has played this transformative role in achieving this growth,” Asadi said.
AI can test whether a level is fun or not, whether it is simply too frustrating or too difficult, and get a general idea of its quality level. It allows designers to iterate faster and spend more time on creative tasks, Asadi said. AI brings the standard in less time with a human being within the loop. The goal is to cut back the “shuffle experience,” where the player essentially has to start out over.
Researchers are currently studying how much generative AI can contribute as a supporting tool for designers to concentrate on improving the gaming experience and increasing designer productivity. The goal of AI is to eliminate tedious work in any job and provides people the liberty to be creative, she said.
Asadi is amazed at how quickly players can race through the newly created levels and keep track of the Candy Crush Saga expansions as they arrive out.
“We have lots of players at the top of our progression and so they are really looking forward to mastering the brand new levels and ending them quickly,” she said.
The focus of the research is on the interface between AI and business.
“I personally have worked in many various firms. And my motto was to beat this challenge, to bring research to product, which is a difficult thing in the event you just wish to leverage existing solutions and take small steps,” she said. “It’s much easier than developing strategic, visionary solutions and turning them into products. And so we discuss how we cracked this: How will we create continuous innovation? How did we manage to get adopted, how did we work together and what led to this case? So that’s a part of the conversation.”
She may be very all for advancing the longer term of AI, not only for Candy Crush Saga, but for all things gaming and game testing.
On the progress made in AI to this point and the way much is to return, Asadi said: “It's a really exciting moment that AI is gaining momentum and innovation is having great momentum around the globe. And that opens up lots of latest opportunities.” , she said. “There is a big appetite (for AI) in society and all firms wish to explore and check out out AI. There are also good technologies that enable further innovation. That’s why I believe it’s time to innovate and find solutions.”
This implies that King's research solves more complex problems.
“The core (of the research) is to be sure that we offer players with an entertaining gaming experience. And we also be sure that we give our designers the space to work on the creative part and fewer on on a regular basis parts,” said Asadi. “And that makes me excited. We offer real added value in each features to our developers and designers in addition to our players. I would like to emphasise the undeniable fact that it takes an incredible effort of cross-collaboration to implement this in a corporation – to adopt it and really use it in production.”
I noticed that there are external services that King can now use for AI services. Asadi said the corporate has expanded its internal capabilities over time and is exploring external tools. But many tools must be application-specific, Asadi said.
“We publish lots about our research. And we also like to take a look at the issues that we’d like to resolve internally by developing AI solutions internally,” she said. “We don’t go into the technical details of the machine learning stuff we’re constructing. We are working on reinforcement learning. And we're AlphaGo, like everyone was talking about, which is all about getting essentially the most powerful agents that may beat one of the best human player. We wish to have a playtest agent that plays like a human because we would like to measure whether the gaming experience we’re developing is sweet for the player.”
And King has various kinds of players with different skills, styles and preferences. And it explores integrate all of this into its solutions.
