Using Video Games in Developing AI

According to Henry Lowood, video games before the 1970s operated without a single program code or algorithm, but developed with a simple logic design and television technologies (5). The introduction of microprocessors and computer circuits in the early 1970s contributed enormously to the improvement of algorithms, computer software, artificial intelligence and video games. Moreover, early AI researchers at MIT and Stanford University not only assisted in developing programmed games but also grafted AI onto video games and foresaw the boundless capacity of these new technologies. Nowadays, many people from various fields, including IT, are developing cognitive technology and striving to implement AI to their business in diverse forms. Since I am majoring in technology management in the college, in the half of my classes, the professors and classmates all talk about new technologies, AI in particular. I have learned that there are many forms of AI in industries from simple process automation to relatively complicated cognitive engagement.

More importantly, the benefit that the cognitive technology will bring has a huge impact on manufacturing, logistics, marketing and sales. It is now easy to see automated robots in factories or communicate with an AI agent that provides customer service which is available any time even in late night. Nevertheless, cognitive technology that solves complex problems in real-world is not common yet. I remember when one of the professors asked in the class about the first experience on AI, most of the classmates answered the customer services or smart home devices. However, the first time I experienced AI was when I played against the AI program in Starcraft, a military real-time strategy video game from Blizzard Entertainment, in the summer of 1999. I did not know the significance of AI at that time, but the intelligent machines eventually become an important technology worldwide. I realized that the AI boom began with video games in the 1970s from the early popular adventure game Zork to the recent AI Go player AlphaGo as well as many others. It is already proven that AI and video games helped develop each other for decades. Adapting AI to video games is much easier than to the real world. Therefore, games could be an appropriate method to help AI adapt to the real world and solve physical problems that share similar properties to video games. I first became interested in the relationship between AI and games when Google DeepMind’s AlphaGo, an AI program that plays the board game Go, beat Sedol Lee, the 18-time world champion considered to be the best Go player in the last decade. It has become more common for AI players to win games against humans in chess, quiz shows, Othello, and video games. Unlike other simple board games, Go is played on a board that has a 19×19 grid of lines containing 361 points, and two players compete to surround more territory than their opponent with white and black stones in turns. Since there is a great number of possibilities that can be played on the board, DeepMind initially declared that the strongest computer programmed Go players before AlphaGo were only able to compete with amateur players despite decades of efforts (AlphaGo). Traditional AI methods, including the successful computer chess machine, built a search tree over all possible actions.

However, the old methods could not be successful in Go because it was difficult to determine the strength of possible board positions and moves (AlphaGo). Researchers at DeepMind integrated Advanced Tree Search with Deep Neural Networks in order to understand the aspect of the board positions. According to the article The story of AlphaGo so far, the neural networks contain millions of neuron-like connections to calculate all possible moves, the Policy Network selects the next move, and the Value Network predicts the winner of the game (AlphaGo). Furthermore, with large amounts of amateur games data, AlphaGo understood what reasonable human play would look like, and developed itself through a process called Reinforcement Learning (AlphaGo). In March 2016, AlphaGo defeated Sedol Lee by 4-1 and became the first computer programmed player to win over a professional human Go player.

DeepMind proposed that the goal of AlphaGo was not only to defeat professional human Go players, but also to implement AI in real-world situations. DeepMind believes that the approach of a cognitive programmed player will be applicable to real world structured problems sharing analogous features with a game like go (AlphaGo). It will be especially effective in problem-solving situations where the correct sequence is significant or planning tasks such as protein folding, reducing energy consumption, or searching for revolutionary new materials (AlphaGo). Likewise, the article Is chess the drosophila of artificial intelligence? A social history of an algorithm also has a similar viewpoint. Since the mid-1960s, computer scientists have referred to chess as the “drosophila”, or fruit fly, of AI because chess is a relatively familiar, accessible, and simple experimental technology that can be applied to create valid information about more complex technology (Ensmenger 5). Ensmenger showed that algorithms of chess acted as a bridge to new technologies for computer software and for computer games in particular, and he emphasized that it is important to develop tools and methodologies for learning new technology (5, 23). By helping researchers in computer science to enhance the capacity of AI, chess is a suitable example of how games help AI adapt to realities and solve real-world problems. While I was learning about AI in another course, I found a video on TEDx YouTube that strongly supports my thesis. It is called “Artificial intelligence, video games, and the mysteries of the mind by DeepMind senior research scientist Raia Hadsell, who earned her Ph.D. in Computer Science at New York University.

Her main thesis was Artificial intelligence could be the powerful tool we need to solve some of the biggest problems facing our world (TEDx Talks). There are many resources about AI, but what made me use this video as a primary source of my paper was that she worked with video games to prove her hypothesis. From the book, From counterculture to Cyberculture: Stewart Brand, the Whole Earth Network, and the Rise of Digital Utopianism, Turner appeals that people should understand and control over the computer program as a tool, and the program should instruct users for control (115). Activities with computer and video games had once treated as technologies of dehumanization and anti-society, and, in the other hands, counterculture had been formed. Individuals, who controlled over the technology and invented for pleasure, were considered as hackers. The hackers from AI labs contributed to invention of SpaceWar and recognized as countercultural pioneers (Turner 116). Is it just a coincidence that both Hadsell and hackers are (AI researchers) who wanted to expose the technology and improved the world. I believe Turner wanted to suggest that the hacker culture allows people to learn computer technology as they wish to learn something about the world at large. There are conservative views on new technology, particulary on AI, mostly for taking humans’ working place or not having an ability to make a judgment like human beings. I suppose, despite of the pessimistic view on AI, Hadsell shares the hacker ethics that she wanted to expose the capability of AI and control the technology to solve the structured problems. Moreover, according to her, games are useful to objectively evaluate the AI agent and to compare scores with other AI agents. More importantly, Hadsell emphasized that games are a microcosm of human culture, and diverse and ubiquitous (TEDx Talks). However, she did not explain enough about why video games are the most suitable medium. I believe the article, Anytime, Anywhere: Tetsuwan Atomu Stickers and the Emergence of Character Merchandizing, can help explain why Hadsell choose video games as a medium of her experiments.

Steinberg describes how media ubiquity with character stickers can be a communicative medium or technology of connection (128-129). He explains that characters that can be stuck anywhere and seen anytime have a capacity for interpersonal communication and persuasive force (Steinberg 128-129). Like Steinberg exemplified with the character sticker, video games are casually accessible, embedded in the environment, and connected to ubiquitous network structures. Therefore, the video games can be an effective medium of Hadsell’s presentation. To prove her hypothesis, her team tested 57 Atari games, and in most of them, the AI performed better than humans. She showed BreakOut and Pong as examples to prove how machines can quickly evolve with reinforcement learning (I believe Henry Lowood’s article could prove why she used Atari games to test her experiments). In the last part of the video, she defined three reasons why she wanted to research AI. The first reason is to gain intelligence through interactions with a complex world (TEDx Talks). With another video game, The Infamous Montezuma’s Revenge, she wanted to exemplify that AI can break down complicated problems to simpler problems in order to answer questions and learn from them. The second reason is that she was deeply curious about how we work, and AI may be exactly the platform we need to prove those mysteries (TEDx Talks). The way AI learns from a video game with reinforcement learning is similar to how humans learn. As Turner supposed, Hadsell wanted to study the technology to learn something about the world at large (135). The last reason is that AI may be the powerful tool that we need to solve our problems to cure disease, to ease inequality, and to save the warming planet (TEDx Talks). As she mentioned, video games could be perfect places for AI agents to practice real-world tasks. There is not yet a complete artificial intelligence, but she believes that it is possible we could see one develop in the next decade. Hadsell’s three reasons greatly match my thesis that video games help AI adapt to the real world and solve physical problems that share similar properties to video games. While Hadsell tested her experiment with relatively simple Atari games, Artur Filipowicz, an AI researcher at Princeton University, has been trying to build AI software for autonomous vehicles using Grand Theft Auto V. GTA V is an action-adventure video game published by Rockstar Games, and it allows players to freely roam in an artificial city based on Los Angeles. The problem Filipowicz attempts to solve is that autonomous vehicles AI should be able to recognize stop signs.

Since the failure of AI could cause a human fatality, it must perform the tasks without a single mistake. Video games that provides the virtual space could allow users and researchers to simulate situations that can be happened in the real-world or that is too dangerous to be happened or causing physical damages in reality. In addition, the cyber space of GTA V actually resembles the existing city. The virtual space that copies the real-world could act as an augmented reality, and provides more immersive experience to users. These reasons led Filipowicz to use GTA V to solve this real-world problem. I believe the book Hamlet on the Holodeck by Janet H. Murray also explains why GTA V is the perfect system for Filipowicz’s research. According to Murray, games have a spatial capacity to represent navigable space (79-80). Because the cyber city of GTA V offers an artificial environment of Los Angeles, the game allows Filipowicz to develop and test his software without damaging anything in the real world.

Furthermore, Murray also insists that video games have the encyclopedic capacity to simulate real-world-like situations. She used the video game SimCity to show that the software calculates the effects of each change by using models very like the ones used by social scientists and policymakers to study urban systems (Murray 87-88). Filipowicz’s software could not only navigate through virtual streets and roads but also react to stop signs as the software would drive on real roads.

Unlike the Atari games, video games, designed with encyclopedic capacity, provides more broad range of possible simulations that players can interact with greater freedom. Before researching for my thesis, I considered GTA V as a violent game that the users mainly play to rob, assassin, kidnap and so on. I did not also think about using GTA V for a scientific experiment. Moreover, playing such a high quality games require me to buy an expensive graphic card. These games providing immersive experience with ultra HD resolution contributed greatly to the improvement of graphic cards. The graphic processing unit turned out to be incredibly effective when the graphic processing unit is applied to the calculations needed for neural nets.

The GPU expeditiously manipulate memory to create images in a frame buffer. Because GPU was originally and mostly developed to dispay graphics for computer gaming, it decrease the unit costs and offered scale economies as well. This improvement also favors engineers to exam and produce better algorithms as well as allows computers to absorb larger data in a shorter period of time. For example, GPU enables the cryptocurrency miners to solve tough calculations in order to mine more cryptocurrencies. Video games like GTA V devoted to the improvement of technology both directly and indirectly. Particularly, GTA V provides the cyber space for researchers to test the hypothesis, and demands the development of hardware. In the actual test with the video game, the software could only detect 95.5% of the stop signs within 20 meters, and the software designers need to increase the accuracy in order for the application to be implemented (Filipowicz 24). The way Filipowicz used GTA V to simulate real-world problems in virtual space supports my thesis. GTA V acted as a medium to help the AI software adapt to the reality and solve structured problems that features analogous property to the game.

Nowadays, many people talk about AI, and big IT companies strive to develop and implement this technology. Nevertheless, it is still controversial whether AI is truly beneficial to humans or it is likely to be a risk. Cognitive technology can reduce time and cost for many product and services as well as have abilities to learn and compute to provide good advices to people. Despite of the advantages, there are cons about AI too. For instance, some people believe that AI can be programmed to do something devasting, lead to major job loss, or do not have an ability to make judgments like humans.

Not only the people from tech field, but also entire world are debating the aftereffect of the intelligent system. Some people might think that AI is years away from influencing our lives, but AI is actually affecting our daily life and our decisions such as smart phone applications, social media feeds, smart home devices, online advertising network, and so many others. In fact, the first time I experienced a cognitive program was when I played a video game. However, I was only aware of AI in video games as NPCs that either entertain players or instruct beginner and intermediate players. I now know that video games can be a great place where people and AI can practice physical problems. It has been proven that artificial intelligence and video games help develop each other. In addition, with the examples of DeepMind, Hadsell, and Filipowicz, I believe that video games can be a suitable medium that assists AI to adapt to the reality and solve problems facing the real world.

Cite this page