Alan Turing designed a test based on the imitation game in 1950 which tests the human behaviour of computers. The Turing Test has traditionally been text based because computer speech is still noticeably different from human. For the majority of games this would not be applicable as game AI involves a lot of actions, reactions, and speech meaning a graphical view would be needed.
The most recognisable game AI are in non playable characters (NPCs), usually as companions, or rivals. The game being tested will often differ from other games and experts of the particular activity will need to be the judge. A good example are football games, where the computer could be one team and the human player could be the other.
Using this example, how can we adapt the Turing Test so that computer games can be tested?
The main difference would be the judges, usually the only requirement is that they are human, however with games there would need to be a certain amount of knowledge and experience in games to allow them to recognise the difference. In this case they may also need a certain amount of knowledge in the rules and tactics of football. The judges would also have less control as they would no longer ask questions but possibly fill in a questionnaire of pre-defined criteria while they watch the gameplay.
Another difference would be the human players, how competent should they be at playing the game. An experienced player could seem like AI if they played too perfectly while at the other end of the scale novice players could seem like a terrible AI. The level of play would need to be tested and possibly randomized as no matter the human players experience they are still examples of human intelligence.
The games themselves would also need to be adapted to allow more anonymity, such as different camera views and removing defining features from human players.
These changes could work for most games with more possible changes as with our example, there might be changes needed as part of the human players team is also controlled by AI. This could hinder the judges decision, however if the judges have this knowledge they may be able to still judge correctly.
The Turing Test may be a good way to rate game AI however there may be simpler ways that could contain a scale similar to game rating.
