Gamasutra recently reported that Google’s DeepMind AI won five games of Go against the world renowned Lee Sedol. This is a huge milestone for artificial intelligence technology because it marks the reality that AI like Data from Star Trek or the upcoming Detroit: Become Human will eventually live among us. Essentially, AI are becoming extremely intelligent due to the magic of computer programming and robotics. DeepMind AI developers are focused on creating a machine that can learn and perform increasingly complicated tasks. This is unlike an AI in a video game, which usually has scripted responses rather then learning and reacting naturally.
For example, DeepMind AI is not a glorified calculator, it learned the ancient game of Go much like a human does. As such, because computers are able to store knowledge more effectively than the human brain, DeepMind AI was able to mimic the sense of intuition crucial to playing Go. In addition, the AI that, for example, acts as an enemy in a video game is not learning anything from the player. A Big Daddy from Bioshock is triggered by the proximity of the player’s character and goes through a series of scripted attacks in an apparently random order. I believe that AI like Google’s DeepMind could be applied to video games in a constructive and exciting way.
However, an entirely new gaming industry would likely result from the marriage of DeepMind AI and game AI systems. Games will no longer be a relatively simple set up on your living room, but require entirely full immersion. For example, think of an NPC like Snow from The Wolf Among Us that reacted in real time to the player’s statements and even facial expressions. Suddenly, games that rely heavily on the player’s reaction would be a great deal more complicated and exciting. Not only will NPC’s remember the player’s actions, but retain the ability to react accordingly. Rage quitting will have a whole new meaning.
Another example for a DeepMind AI in a game would be in the enemies, like the Gaping Dragon from Dark Souls . If it learned your attack patterns and countered accordingly, the fight would nigh impossible. Instead of cycling through all of its attacks at random, the player’s ranged attacks would be countered with a rush and poison attacks. Game physics would certainly improve as well; if the Gaping Dragon learned from its environment, it would use the lack of hiding spaces to corner the player and attack them.
A Gaping Dragon with an AI like DeepMind introduces the possibility that an AI can become too smart. Not only will this create a less entertaining game, failing a quest or battle to an AI over and over would certainly lose its flavor quickly. However, if an AI is supposed to reflect human intelligence, then even the AI version of Sherlock Holmes will have flaws. I do not believe that “too smart” AI are what the gaming industry should strive towards, but the latter flawed AI. Any AI in a game needs to be believable and follow a certain amount of logic, even if it is just game logic. If you have ever played chess with a computer and went from easy to hard, you can definitely see the difference. On easy, the computer will purposely make mistakes and if it were not so obvious, you could almost pretend you were playing with a friend. Of course, AI like DeepMind are a lot more sophisticated, and the mistakes are more natural. If an NPC is too perfect, it is impossible to relate to or interact with in any entertaining way. If the AI is flawed, the player can more easily understand and communicate with it, on a human level, so to speak.
Using AI similar to DeepMind in games is a step towards a whole new game industry, requiring full immersion. We might already be heading that direction with virtual reality technology. Still, developers should tread carefully. The area between too smart and too flawed is difficult to find, but it is possible. Once that happy medium is found, the possibilities for games are endless. Until that day, we shall stick to Gaping Dragons that do not kill us in a matter of seconds.