Since last night’s victory of IBM’s Watson over human Jeopardy opponents, the news has reverberated with stories of the triumph of machines over man (often ignoring, of course, the massive human effort that went into Watson’s creation).
It’s no surprise that a game was the format for this test of artificial intelligence. After all, games have a huge advantage over the real world: each game provides a simple bounding box, with a discrete set of rules and information. Before Watson, IBM became famous when Deep Blue beat world chess champion Gary Kasparov. Although Jeopardy has far more complexity than chess (at least in terms of the number of potential responses at each individual turn of the game), it’s still a lot simpler than real life.
Humans and all other biological organisms on the Earth have evolved for hundreds of millions of years to interact with our environment. The real world is hard. Really hard. That’s why things like independent robots and even “simpler” robots like automated cars have been so hard to work. Our brains are really good at handling edge conditions, unexpected situations, and learning from mistakes.
On the other hand, games provide a great landscape for developing intelligent computers–the problem space is much smaller. Watson is essentially a powerful search engine technology strapped to a voice synthesizer; it’s intelligence is more of an illusion brought about by the narrow “world” imposed by the Jeopardy game. But where will we go next? I imagine that 3D, immersive games and MMORPGs will be where we’ll see the first realistic simulacra of human players–where interesting characters might draw upon a repository of language, interactions and storytelling to create compelling experiences. These characters could become far more interesting than the cardboard cutouts that occupy most games today–and if properly trained, might be more fun than many human “roleplayers.”
I’m expecting that we’ll meet the first computers to pass the Turing Test inside a game.
Games are constantly getting more complex, and 3D landscapes will increasingly approach simulations of the real world. This will be a perfect environment to challenge machines to become more intelligent over time–preparing them to interact in the real world. Games are like a nurseries for emergent intelligence.
This raises another question: what exactly is intelligence anyway? We often seem to have a bias toward defining intelligence as intellectual problem solving. Of course, computers already surpassed us at certain problems (e.g., performing calculations) years ago. Things that deal with language, abstract concepts, etc. are harder–but don’t seem intractable. On the other hand, we’re often biased against other forms of intelligence:
- Emotional intelligence. It’s always been a trope within science fiction that computers would lack emotions. This strikes me as a huge shortcoming, and a big intelligence deficit. Our emotions are really optimizations–they give us physical feelings that help us make decisions that (hopefully) favor our survival. Empathy and interpreting other peoples’ emotions is also central to this form of intelligence. Computers have a long way to catch up to dogs in this area, let alone humans. Ultimately, this will mean that computers will need to interpret body language and other organic signals to develop emotional intelligence–but games might provide another simplified framework for early development, since online games separates us from these elements.
- Social intelligence. This goes right along with emotional intelligence; it’s understanding the subtleties of interactions between people. Some of this is hardwired, but it’s also something we learn through the course of our lives. A huge portion of our brain is dedicated to dealing with social situations, understanding social institutions and conventions, and developing expectations or theories about how other people behave. Again, online social games might provide a simpler place for computer-controlled characters to begin developing social skills that could someday be adapted to the real world.
- Physical intelligence. Our brains contain a huge number of neurons associated with learning, controlling and adjusting our physical movements. The cerebellum, where much of this intelligence is stored, might actually comprise most of the neurons in our brain! Isn’t a dancer, athlete or other physical performer expressing a form of intelligence that’s at least as impressive as anything involving intellectual problem-solving? Once again, games are providing a simpler context for refining physical intelligence–consider the RoboCup, a competition for robots to learn to play international football (soccer). Getting robots to deal with all the challenges that come up in the diversity of environments and situations that humans deal with regularly will take a long time–but games offer the perfect training ground.
What are some other forms of intelligence that might emerge from within games? I’d love to hear your comments!