Artificial Intelligence (AI) is becoming a key part of the modern technological world, including in video games, despite some not even realizing how common it is. In the bigger picture, AI is the subject of great debate. Some (such as technology rock star Elon Musk) believe its advancement could have apocalyptic consequences. There are some scary stories, such as the Facebook experiment shut down after two separate AI bots developed their own language to communicate more effectively with one another.
These days Facebook’s mission of giving personalized content from user data has led to some bigger problems than bot experiments. With the dawn of personalized experiences driving engagement, with toys such as Tamagotchi’s or in RPGs where your choices impact your user experience, we’ve seen the illusion of intelligence be a core element of how gaming has developed. The difference now in 2018 is we’re seeing a shift from that illusion of intelligence to actual intelligence. It more than ever is worth exploring the connection between Artificial Intelligence and Video Games.
Competitive gaming is fueling AI development.
OpenAI is an organization that developed a bot that mastered the game of DotA (Defense of the Ancients). The bot learned the game by self-play, learning through its own mistakes, and not using imitation learning or tree search. In simple terms, it taught itself how to be good through trial and error. Their bot has developed now to the point to where it has beat both SumaiL (the top 1v1 player in the world) and Arteezy (the top overall player in the world).
Obviously, this is groundbreaking, but there were some limitations. For one, the player and the bot used the same hero (the Shadow Fiend), and the game was limited to a 1v1 match. However, there’s no dismissing the significance of OpenAI’s achievement. They are now developing a team of bots that can work together to beat the top teams. Games such as DotA have many variables and hidden information from players. There’s a lot of strategic gameplay and planning that goes into being successful in matches. Video games provide the perfect testing ground for not only AI to learn strategic problem-solving skills, but we could also look at them as a perfect place to learn social skills as well.
Video games could help AI learn more realistic social skills.
Tay.AI was Microsoft’s “conversational understanding” AI twitter bot. Some might recall shortly after launch Microsoft had to shut Tay down. Why? She, unfortunately, started going on racist rants fueled by content given by Twitter users. It was a disaster for Microsoft. A definite PR black eye that, at the least, raised questions as to how dangerous AI could be. There is a lesson to be learned here.
Tay gathered a vast amount of (poisonous) data it received from users, and it certainly affected its output very quickly. Using a game such as World of Warcraft (WoW) as an example, you could argue this would be a much better testing ground for Artificial Intelligence looking to learn realistic social interaction. This is because in-game relationships are built on trying to accomplish a goal and working together, rather than a site such as Twitter where you’re simply making a statement. The game provides a context, as it is a simulation of a world. WoW has been studied in the past to understand human behavior, such as the infamous Corrupted Blood incident.
Players in WoW are united with different groups and work together to achieve hard goals. Can people be toxic? Absolutely, but if you have a negative experience with someone there’s a chance you could see them again in the world or gain a reputation in these tight-knit communities as being someone to avoid.
What could the future hold?
With changes in the WoW MMORPG landscape, this social element is less significant than it was a decade ago. With that said, it still represents a closer-to-reality experience than a purely social site such as Twitter. Could a bot develop enough social intelligence to be able to work together with a team to accomplish a task, along with learning complex problem-solving strategies? Could a team consist of both humans and AI bots alike? The answers to this could have implications much more far-reaching than just video game gameplay.
In the upcoming WoW expansion, we’re going to be seeing the most advanced NPC (non-player characters) in the game for island expedition scenarios. These bots will be strategic and will operate at a higher skill level than any previous NPC included in the game. Is it that far to imagine a game where if you speak to these characters they will see you later in say Orgrimmar (one main game city) and discuss the match scenario you played together recently like any human player would? How about asking if you’d like to join them on another mission in 20 minutes or so? Could they decide if they’d want to join a guild in-game? Go raid with the team, or compete in Player VS Player (PVP) with them? What would this require from a research standpoint? If projects like OpenAI are any indication of the development of AI, we might not be too far off.