Game makers tapping artificial intelligence to keep players hooked
In the game industry of today, titles like Clash Royale and Pokemon Go are free for most people because there’s a small number of players who pay for extras like special weapons and more lives. Game developers have to strike a delicate balance in this free-toplay model between drawing the masses and encouraging big spenders — and they need both for a successful title.
Silicon Studio Corp. is trying to help by providing game makers with deep-learning algorithms to create what amounts to a psychological profile of each player. The Tokyo-based company’s software predicts how long people will play, what levels they might achieve, how much money they might spend and on what. Even more important, the technology lets game creators mould player behaviour to keep them hooked.
“Game data is perfect for studying human behaviour,” said Africa Perianez, chief data scientist at Silicon Studio and a former nuclear physicist at the European nuclear research organization CERN. “It’s going to change the industry, change the direction of personalized games.”
The machine-learning software, called Yokozuna Data after the highest rank in sumo wrestling, is drawing customers. Three publicly traded Japanese publishers and a South Korean developer have signed up to use the product, Perianez said, declining to give their names because of confidentiality agreements. The company is also in talks with large European publishers of massive multiplayer online role-playing games, Perianez said.
Japanese and South Korean game publishers pioneered the art of making money from freeto-play titles. For years, they employed so-called live ops teams that use events, competitions and limited-time offers to get people to pay up. As those techniques mature, companies are turning to artificial intelligence and data-mining to influence players — strategies similar to those Google and Facebook use for targeted advertising.
Silicon Studio was founded in 1999 as a unit of Silicon Graphics, the U.S. maker of high performance computers used for special effects in “Jurassic Park.” The company was spun off the following year to focus on software tools for other game makers, like Yokozuna, and develops its own games.
Yokozuna, which was in development for two years, can tailor promotions to specific groups or individuals. For example, users at risk of quitting a game like GungHo Online Entertainment’s Puzzle & Dragons may find it easier to win rare monsters or faster to advance through game levels. For Niantic’s Pokemon Go, Yokozuna could help schedule extra events for a holiday weekend — and customize walking distances based on fitness.
A key challenge in free-toplay gaming is maintaining a healthy ecosystem of players who spent a lot (called whales) and those who never pay (krill). Industry insiders, who favour marine-biology terms, call casual spenders dolphins. Whales usually comprise one per cent of all players, but generate half of total revenue. Though krill may seem irrelevant for game developers since they don’t pay, they are essential because paying users need competition from others to hand over their money. The whales need something to eat.
As more smartphone games became available for free, the industry adopted microtransactions to generate revenue, selling digital trinkets and tokens. Even the simplest games operate virtual economies, with startups such as Scientific Revenue and Gondola offering analytics and dynamic pricing tools.
Enticing players to pay with custom incentives is tricky, though. In July, fans of Zynga’s CSR Racing 2 game revolted when they discovered some gamers paid $35 for content that others got for $5. The company apologized and offered compensation.
Silicon Studio’s Yokozuna software is part of a bigger trend of artificial intelligence researchers looking to video games for complex challenges beyond chess and Go. Many of the recent advances in naturallanguage processing, and image and speech recognition, have come from deep learning, an AI subdiscipline that requires human-labeled data to work. Video-game environments are a good source of data because every interaction is recorded.
“There is no other field that has better data,” said Perianez, who previously worked on predictions of mobile subscriptions and Coca-Cola sales. “You can measure habits continuously for years.”