How AI is shaping the future of interactive games

This article is part of a collaboration with iQ by Intel.

In the 2013 film Her, protagonist Theodore plays a videogame where he is surprised by a wild, swearing artificially intelligent cartoon character. The foul-mouthed little alien launches into a conversation with Theodore, remaining stubbornly unhelpful. Realizing it must be a test, Theodore curses back in a verbal brawl that ends with the alien showing him the way forward.

Her is science fiction, but that human-like interactive game featured in the movie may be closer to becoming real. Created by game designer David OReilly, the Alien Child game offers a glimpse into how artificial intelligence (AI) and machine learning could make games of the future more engaging and interactive than ever before.

The Suspect 1

“The most fascinating use of machine learning is where the AI learns about us, as well as learning about itself and the characters we write,” said Guy Gadney, director of The Suspect (2014), an interactive AI thriller. The Suspect used machine learning to put players in conversation with the fictional criminal mastermind Markus Winter. To save some of his hostages, players must use interrogation tactics to try and outwit him.

Similar to interacting with a chatbot, the experience makes players type responses to Winter, then the AI responds on-the-fly. Aiming to approximate the kind of wily, organic conversation seen in Her, Gadney’s use of machine learning technology adds an organic intensity and suspense. “We wanted to create surprise, which only happens when a machine steps out of common sense and presents a situation no one is expecting,” he said. “As humans, we communicate. It’s what makes something really believable.”

Traditional software is a set of strict rules that a machine follows to the letter. Usually, a computer just executes that chain of command, but machine learning allows the computer to discover the rules with time and experimentation. “Machine learning is defined as a class of programs where performance improves with time,” explained Pradeep Dubey, an Intel Fellow and Director of Parallel Computing Lab. “As the program crunches more data, it gets better and better.”

Machine learning and AI technologies open a world of creative possibilities

George Dolbier, IBM North America CTO of Interactive Media, describes machine learning as “a way of writing software with data rather than code.” “Software makes choices, perform actions based on statistics and change behavior based on new data,” he said. Because they are nimble, machine learning and AI technologies open a world of creative possibilities, according to Dolbier. It can be applied “to almost all aspects of the play experience in the future, whether deciding how to adjust difficulty and rewards for specific players, analyzing massive amounts of player data or creating entirely fictional personalities we can interact with,” he said.

Back in May, for example, YouTuber SethBling wrote a program called MarI/O that taught itself to play Super Mario World (1990). Unlike other AI, it started out with no prior directives but, like human players, learned to navigate the level through trial and error. MarI/O wasn’t necessarily “thinking” about how to beat the game. After each death, it repeated successes and randomly tried new things to replace failures. While approximating human faculties, MarI/O’s machine-like mentality became apparent through its hyper focus on reaching the end goal, ignoring the temptation of power-ups and coins along the way — something a human player likely wouldn’t do.

The Suspect 3

Machine learning may not be ready to fully replicate human behavior, but Gadney believes it can bring a more human element to the way people interact with technology. “As we grow up, we learn through experience and emotions,” he explained. “And if we can use machine learning to recreate that and put it into an interactive experience, then I think we’re on to something.” Machine learning carries The Suspect’s unique experience, but that doesn’t mean the game is writing itself or that the story is random. It was used in combination with other, more traditional forms of game design.

Gandey said the best way to think about machine learning in game design right now is as another tool, like a paintbrush. It’s a tool that’s far from perfect. In March, Microsoft’s Twitter bot Tay was supposed to learn how to have a casual conversation with everyone who interacted with her. But after 24 hours of coordinated trolling, she ended up spewing hate speech and was shut down. This emphasized the need to balance freedom and pointed rulesets — like accounting for trolls in an online forum.

One day, machine learning may look as seamless and personal as the artificial intelligence envisioned in Her. Games are far from that right now, but a whole new world appears to open up when players are able to have conversations and build more organic relationships with AI, according to Alien Child creator David OReilly. Machines are giving human intelligence a run for the money, according to Dubey. “It’s not just a theory anymore,” he said. “In practice, machines have actually delivered better accuracy than sometimes the best human.”