You have dealt with artificial intelligence if you’ve ever played a video game. You’ll still find elements powered by AI, regardless of whether you prefer race-car games like Need for Speed, strategy games like Civilization, or shooting games like Counter-Strike. AIs, like enemy creeps, neutral traders, or even animals, are always behind the characters you usually don’t pay much attention to. But how does AI connect to the AI that technological giants talk about every day in gaming? Nowadays Artificial Intelligence in video games is used to generate intelligent behavior similar to the intelligence of humans or animals.
Playing against an AI:
Elon Musk has recently warned the world that the accelerated production of AI by Google and Facebook with learning capabilities will place civilization at risk. This statement has brought a great deal of media attention to the subject of AI. The dazzling vision AI mentioned by these tech giants seems to be a software that, when fed more data, can teach itself and get stronger and stronger. For AI such as AlphaGo, which is known for beating the best human Go matches, this is true to some degree. Through watching millions of historical Go matches, AlphaGo was learned and still learns by playing online with human players. The word ‘AI’ in the sense of video games, however, is not limited to this self-teaching AI.
In video games, AI is programmed to improve human players’ gameplay experience instead of learning how best to defeat human players. Controlling non-player characters is the most common role for AI in video games (NPCs). To make these NPCs look wise, programmers also use tricks. In the 1990s, one of the most commonly known tricks, called the algorithm of the Finite State Machine (FSM), was applied to video game production. A planner generalizes all potential scenarios that an AI might face in an FSM and then schedules a particular response for each scenario. Basically, with its pre-programmed actions, an FSM AI will respond rapidly to the human player’s action.
An obvious downside of FSM architecture is its predictability. All NPCs’ activities are pre-programmed, so after playing an FSM-based game a few times, a player can lose interest.
The Monte Carlo Search Tree (MCST) algorithm is a more sophisticated approach used to boost your personal gaming experience. MCST encompasses the technique of the randomized trial for addressing a dilemma. This is the tactic used to beat a chess champion in 1997 by Deep Blue, the first computer program. Deep Blue will take the MCST to consider all potential movements in each position of the game, then take into consideration all possible human player moves, and then take all possible reaction moves and so on. Both future steps can be pictured as the branches expand out of a stem—why that’s we name the tree “search tree.” If this step has been replicated several times, the AI can determine the incentive and then select the appropriate branch to pursue. The AI will replay the search tree after a real step, depending on potential results.
In several military games, a related algorithm was also implemented. However, as the movements are much more than in chess, any of them cannot be taken into account. In these games, instead, the MCST will select randomly to launch any moves. As a result, the results for human players are much more unpredictable. For eg, it is not possible to preprogram any move for AI in the Civilization game, in which players engage in building a city in competition with an AI that does the same thing.
Instead of taking action based solely on the present state of FSM, the MCST AI evaluates any potential next steps, for example, the advancement of technology.’ The AI then executes the MCST to determine the total reimbursement on both of these movements.
The following figure shows a simplified flow chart of how to use MCST in such a game (Figure 1). Complicated open-world games such as Society use MCST to have multiple AI compartments at any round. In these games, the situation is never predetermined, supplying human players with a fresh gameplay experience at all times.
Learning to become a smarter AI:
While in the nineties AI designers tried very hard to make the NPCs look clever, one very significant attribute was missing: the desire to understand. In the majority of video games, the action habits of NPCs are coded, and they cannot learn from players, for example, because of the feedback of humans. Not just that program machines are challenging to study, the explanation that most NPCs do not demonstrate the potential to learn is because most programmers tend to prevent any unintended NPC actions that might affect a human player’s experience.
Petz, the digital pet game, was one of the first video spiel AIs to use NPCs with learning power. In a game, a player will just like a real dog or cat train a digital pet. As the training style varies between the players, the behavior of their pets is often customized and thus the relationship between pet and player is good. Having learned in this game would also mean that game makers are unable to monitor the game environment as a whole, meaning that this approach doesn’t make it very popular with designers.
With the aid of shooting again, a human player would turn up in one place intentionally over and again, and the AI can eventually strike this place without exploration. The player will then use AI’s memory to stop meeting or embushing the AI. This technique is beyond the influence of the designers. Up to now, the only category of the gaming industry that is continuously using AIs with the potential to learn is virtual pet play.
Playing against or living with an AI?
Following AlphaGo’s achievements, some raised the issue of whether AIs could also defeat human players in computer games like StarCraft, WarCraft, or FIFA in a real-time strategy (RTS). The quick reaction is yes, it does. RTS games are much more complex than basic games such as Go in terms of possible moves and the number of units to be controlled. In RTS playing, AI has big advantages relative to human competitors, such as multitasking and inhuman reactions. In reality, AI programmers had to intentionally reduce the ability of AI to enhance the experience of human players in some games.
In the future, video spiel production by AI would possibly not depend on making better NPCs to beat human players more effectively. The development would now concentrate on how a better and more unique user interface can be created. The distinctions between the virtual and real world have started to merge as virtual reality (VR, which presents an interactive visual experience with the help of a screen) and increased reality (AR, which incorporates the true view of the world of human beings with virtual elements). Last year’s AR game, Pokémon Go, displayed the convincing force to first merge the earth with the world of video games
Electronic Arts CEO, Andrew Wilson, famously projected “your life’s going to be a video game.” As AI-VR/AR technology evolves and allows us to plunge into a more and more immersive environment, his vision can be achieved. Do you think you’d like to play with an AI or a real human in this case? That will become an ever more important topic.