Can A Super Smart Computer Teach Game Characters To Be As Stupid As Humans?

Supercomputer Watson can best the best of us in Jeopardy, but can it make the games we play at home more challenging and fun?

The day after IBM's Watson beat Jeopardy champs Ken Jennings and Brad Rutter, game designer Graeme Devine wondered on Twitter when game makers would be getting access to the technology.

I asked Devine, former gaming guru at Apple and founder of Grl Games, what he thought about the supercomputer and how he would use it.

"For me at least, Watson made me realize that we've constructed a vast artificial intelligence by accident," Devine told me. "I look at Watson as a super fast Google machine that can sift through Wikipedia, TMZ, whatever pops up for any question, comprehend it and answer. We do the same thing every night watching TV: 'Oohh, what show is that actor from?' and we reach for our iPads and Google the answer.

"So this intelligence is now there, answers to everything, and it is the Internet. AI doesn't mean learning anymore, it just means comprehending this vast dataset."

While computers have been beating humanity at the games they play for decades, from Tic-Tac-Toe and Chess to Poker and Halo: Reach, Watson is a different sort of artificial intelligence, Devine said.

"We've cornered the way we think about computers, languages, and programs so much that we can't see anything else but the desktop, code, and that's not changed for 30 years now," Devine said. "Sure, it's awesome for chess and trivia, but we need to get out of that mode of thinking to see the next level."

Chris Butcher, engineering lead at the Bungie - the studio behind the Halo series, says that next level, that new type of artificial intelligence seen in Watson is a leap in natural language processing. Something that could have huge implications for gamers.

Watson operates on several levels, combining a number of technologies together, including the ability to understand and respond in natural speech, the ability to quickly sort through vast amounts of information and on some level, the ability to reason.

"Natural language processing and question engines have two main applications in gaming – firstly by simplifying the interface," Butcher said. "Games (and user experiences in general) will start to be able to interact with us in our own language. The combination of natural language processing, question engines and gestural interfaces will allow us to interact with synthetic experiences as if they were extensions of the real world."

Butcher thinks this new technology will bring with it new challenges, the need for computers to mimic human interaction so completely that those interactions with these computer-driven game characters isn't off-putting to players.

"Even after natural language processing starts to make inroads into the interface, characters that use natural language will take a long time to show up in games," he said. "Think about how long we've been struggling with the uncanny valley of graphical rendering for human faces. We haven't really begun dealing with the equivalent problem in AI, of creating synthetic characters that use and respond to natural language in an authentic fashion."

Devine points out it's not simply a matter of making the computer-controlled characters in a game smarter or faster, it's making sure they also show some of the foibles of players in play.

"It's important that a good AI needs to be dumb enough to look human," Devine said. "When we first started training bots in Quake III they were too good, always perfect, and that's the trouble with AI... The clever AIs don't always do the same moves. They learn and adapt. Does Watson learn and adapt? I don't think so."

Current artificial intelligence systems in video games always seem to have a flaw, Devine said. They may be too fast, too accurate, or just plain cheat.

"It would be fun to see some general new approach to AI that lets me teach it real-time strategy games in the morning and Call of Duty in the afternoon," he said.

A clever, but not too clever artificial intelligence is one of the most important factors in today's video games.

"Great AI responds to the player in ways that are expected but not perfectly predicted," Butcher said. "It's a subtle dance, requiring AI to sense the player's intentions and actions, and work together with them to create a unique experience every time. I love being surprised by AI, it can create a ton of interest and replayability. Without that element of surprise and improvisation, a game can feel like a sequence of canned pre-authored experiences. That turns it into one-time consumable content, a challenge to be beaten and then put on the shelf."

Devine isn't convinced that Watson is the answer to making that dance more human, more subtle, instead he hopes the supercomputer can lead to something greater in video game development than an improvement of the games we already play.

"Will it make the dude on accompany missions not run weird in World of Warcraft? Nope," Devine said. "Will it maybe define a new category of game down the road? Maybe. Just maybe a good artificial intelligence can boot games that depend on data comprehension to a new level."

Well Played is a weekly news and opinion column about the big stories of the week in the gaming industry and its bigger impact on things to come. Feel free to join in the discussion.