On April 1st, 2017, Nvidia uploaded a video on their Youtube channel called ‘Introducing GeForce GTX G-Assist’. The video showed a USB stick, when plugged in, that could learn how you play the game, and an AI would play it for you, mimicking your playing habits. It was fresh, it was revolutionary, and it was also an April fools joke. There was no such thing as G-Assist that could play the game for you using artificial intelligence. However, AI is used widely in gaming nowadays. Starting from NPCs in Wolfenstein 3D in 1992, ‘Bots’ can be seen in popular games such as CS:GO and many MMORPG titles. 

nVIDIA cooperation logo

In 2016, Google Deepmind’s AI ‘AlphaGo’ created a sensation by playing and winning the game of ‘go’ against a world champion Lee Se-dol. This was so sensational because ‘go’ is a strategy game similar to chess, which has so many variables and number of possible cases, AI was previously thought to be too weak and incompetent to play this very complex game. AI won 4 out of 5 games against the world champion, and as a result of this, many professional go leagues were cancelled and Lee Se-dol recently retired from professional go playing. How has AI gotten to the point where it is so advanced, and how will it affect the gaming industry? In order to find out, let’s take a look at the history of AI in games.

The early days of game AI were mostly focused around NPCs. Although they might seem intelligent, they are only made to look this way. Finite State Machine or FSM is a prime example of this. They are programmed to react in certain ways to human player’s actions. For example, an NPC soldier would shoot at a player when he moves right in front of it and retreats when its health points are low. Battlefield, CoD and Tomb Raider are examples of it. A more advanced method is called the Monte Carlo Search or MCST algorithm. This algorithm, also used to defeat a human chess champion in 1997, considers all the possible moves it could make, then considers the possible human player moves, and so on – to create a tree of possible moves from the starting point. Then it will select the best move out of all, and repeat the process all over. Methods like FSM and MCST are useful in making characters look intelligent and beating humans at chess, but they lack the ability to learn. Neural networks are a set of algorithms, modeled loosely after the human brain, that is designed to recognize patterns and learn through it. Generational Neural Networks employ this method to create better ‘generations’ of AI players. After a series of inputs are passed through the hidden layers where random movements are generated, the ones with the best outputs are selected, and they are inputted back into the neural network, again to generate random outputs and best ones selected from them. After a thousand or even a few million generations, the AI becomes very refined and intelligent. AlphaGo, the AI that beat the human champion utilizes this method. By using this method, ‘deep learning’ is possible, where AI can teach itself to do almost anything from mastering Pac-man to driving Tesla’s autonomous cars. As we can see, the AI in games has evolved greatly over the last few decades from basic decision trees to complex neural networks which can learn by itself. This can have both positive and negative effects on the gaming industry.

The most recent case of a high-profile human defeat against AI happened in April 2019, when OpenAI Five won against a professional Dota 2 team. Now, playing the game well isn’t good enough unless you are better than the AIs who can train for thousands of gameplay hours in a short amount of time. AI is changing the ecosystem of the professional gaming industry drastically, and the space for professional game players are getting smaller and smaller. On the bright side, AIs are helping to make developing games more efficient and economically viable. A building in detail-oriented games like hitman or GTA with full-fledged features would take humans days and thousands of dollars to make, while an AI trained to build virtual buildings would take much less time, money and resources to do it. 

AI and its advancement is undoubtedly a double-edged sword, and it will have drastic impacts on gaming, like it has on other parts of our lives and technologies. We should take full advantage of its benefits, and at the same time strive to adapt to the gaming landscape that AI has created.

<strong>Seong Hun (Shaun) Lee</strong>
Seong Hun (Shaun) Lee

Student of NLCS Jeju
Member of NLCS Jeju Computer Science Society

You May Also Like
Read More

Self Driving

From the 1920s, there were experiments for automated driving systems. The trials began in the 1950s. First automated…