Could the secret to developing truly "intelligent" AI be role-play gaming? Rather than the ability to destroy a human player at chess, true AI should be able to convince other players it is really able to invent a narrative.
If this could be achieved, we might well be on the true path to real artificial intelligence.
How is AI used in gaming?
Artificial Intelligence has actually been an integral part of gaming since it's inception in the 1950s. One of the very first examples was a game called Nim that was released in 1952.
The game was very basic compared to modern games, but could regularly beat skilled human players.
AI tends to provide various crucial roles in games and its remit has expanded considerably in recent years,
But, it should be noted that although the term AI is used, they tend to consist of highly sophisticated algorithms and generally do not meet the generally agreed definition of "intelligence".
In most video games, AI tends to be responsible for generating a response, adaptive and intelligent behaviors for non-player characters (NPCs). The more sophisticated applications can almost replicate an artificial human player to provide a challenging gaming experience for players.
Notable examples are non-player civilizations in the Sid Meier's Civilisation Series and, more recently, the Alien in Alien: Isolation.
Most modern games integrate techniques from the discipline at large, like pathfinding and decision trees, to improve the 'intelligence' and responsiveness of NPCs. Additionally, AI is often used in mechanisms which are not immediately visible to the user, such as data mining and procedural content generation.
As you'd expect, the ability of an AI to provide a challenging opponent to a human player is not without criticism from gamers. So much so, that the term "NPC" has come to be used as a pejorative in the gaming community.
This is often because most AI's that control NPC characters tend to follow a pre-determined decision tree that can, after some time, become very predictable indeed.
As these tend to be handwritten they often result in "artificial stupidity" such as repetitive behavior, loss of immersion, or abnormal behavior in situations the developers did not plan for.
Should AI play games rather than run them?
Whilst AI has been an integral part of games since the very beginning, the many examples of NPC "fails" reveals it is far from perfect. Especially for providing a truly realistic experience for players.
But as any gamer will tell you, it is part of the fun.
Backgammon, Go and Liubo, for example, are very ancient indeed. Another board game, called "Senet" is thought to originate from predynastic Egypt and could be as old as 5000 years.
With video games being the natural progression of our innate desire to play, many have seen this ability as a mark of intelligence. In computer science, games are often also used as a benchmark to check to see if an AI can pass muster.
When computer scientists were trying to figure out who the smartest people they knew were, they soon realized, rather modestly, that is was of course themselves.
‘They were all essentially mathematicians by training, and mathematicians do two things – they prove theorems and play chess. And they said, hey if it proves a theorem or plays chess, it must be smart.’ Wilensky told the authors of the 1985 book Compulsive Technology: Computers as Culture.
This might explain why games like Chess have mainly been the focus for proving an AI can beat some of the best human players in the world.
Whilst such achievements are impressive, games like Chess and Go (which is the test game for Google's DeepMind), tend to be highly restrictive with set objectives and clear paths to victory.
That's where role-playing games, like Dungeons and Dragons, might be fertile ground for further developments...
Improving AI through RPGs
Many role-playing games, like D & D, tend to be open-ended, almost sandbox games. Perhaps a proper test for an AI's intelligence is not its ability to thrash a human player in a set number of moves, but rather assess its ability to tell a compelling story.
Could an AI, instead of using the Turing test, be able to 'fool' a human into thinking it is 'real' through games like Dungeons and Dragons?
In Dungeons and Dragons, players are required to inhabit many different characters through many games. Each individual player is often also required to switch between roles from warrior to thief to the healer.
AI, on the other hand, seems to struggle to apply its well-trained algorithm to domains even slightly different from what they are used to. Humans are able to perform this mental "trick" without any real difficulty.
It's probably fair to say that this is a benefit of being the product of several million years of evolution. Not to mention being an intelligence within a physical body.
Video game players, for example, often have very real "physical" interactions with games, from visual, audio and tactile stimuli. Especially with more immersive experiences like VR.
Yet even without sophisticated immersion add-ons, human players can have very real biological responses to games including stress and fear (think Alien: Isolation).
RPG's, like D & D, are also highly social games and so too is intelligence. AI, on the other hand, tends to play games in a purely combative manner.
AI algorithms will go head to head with winning strategies reinforced and rewarded. Whilst certainly effective, this will only make an AI more and more specialized and not well-rounded.
Social animals, like humans, tend to learn socially. Collaboration and mentorship form an important factor in each individuals development.
For open-ended sandbox games, like RPGs, this often requires the player to think creatively and cooperatively with other players.
If AI could also learn to tell a story, as humans can, rather than just beat us at chess, it would make them more multifunctional and adaptable.
This could be the key to developing cleverer, and more believable AI in the future.