BrainyBeard | Trends In Game AI 2017 (January - April) - BrainyBeard

Trends In Game AI 2017 (January – April)

Posted by  on 
28th April 2017

Final Fantasy screenshot

It has been an impressive start to 2017

Introduction

The world of game AI moves quickly and at BrainyBeard, we summarise all of the latest AI news every week. In this article, we outline the top three trends in game AI in 2017.  This will help to provide you with an overview of what has happened so far.

 

Game AI Mastering More Complex Games

Last year, DeepMind shocked the world when it’s AlphaGO AI managed to defeat a world-class GO player. In the first four months of 2017, AI has continued to master a number of high-profile games. Along the way to victory, AI has claimed the scalps of some serious professionals.

Poker fell in February when an AI developed at Carnegie Mellon University defeated players in a marathon of games. Poker is significant because of its use of hidden information. The AI does not know the opponent’s hand and must consider if the player is bluffing or not. Later that month, Super Smash Bro’s Melee was also conquered. The Nintendo classic demise was slightly less impressive as it’s a reaction based game, something computers are traditionally superior at! In surprising news from China, in April StarCraft took one step closer to biting the dust. A team from giant Alibaba published a paper outlining a system that successfully organised combat using strategies learnt from pros.

Real-time Strategy games have often been seen as one of the hardest genres for game AI to conquer. These games contain a large amount of data, which is also hidden via a fog of war. Understandably it has long been thought that it would take some time for AI to master the genre. But a team in China have proved game AI is moving fast, and what title will falter next is anyone’s guess!

 

The scale of Ultimate Battle Simulator is breathtaking

Technical Achievements Involving Game AI

As processing power improves and games become easier to create, game AI is being used more innovatively. In the last 4 months, games from AAA to indie have provided us with a glimpse into the future.

Most recent is Ultimate Battle Simulator. The early access game pits thousands of AI agents against each other in a battle to the death. Although gameplay must be further improved, the technical innovation of the game is inspiring. In March we highlighted Everything, the game that lets you play as…. everything. From bacteria to entire galaxies, the world is yours to control! We have seen impressive simulation games in the past such as SimCity, yet the scale of Everything is astonishing. In February we brought Blitzkrieg 3 to your attention, the unique RTS that utilises neural networks extensively. Whether it’s ensuring soldiers avoid obstacles or the AI taking control of your armies whilst you’re offline, it’s seriously impressive. In a recent article, a gamer commented on how it felt like they were playing a different human during every playthrough.

Grand Theft Auto 5 is one of the best examples of AI in games. It’s driving and traffic systems are excellent, and it’s civilians and police force bring the world to life. Games are becoming realistic enough that researchers are more frequently using them to cheaply test real life AI systems. For example, we recently discovered that Grand Theft Auto is being used to train self-driving cars!

Everything is a peculiar, spectacular game!

Huge Investments In Game AI Globally

The potential of games being used to train AI has created a boom in investment across the globe. Google’s DeepMind team appear to have their hands in almost every AI pie, but there is now increasing pressure however from other teams ready to crush the British-based giants. A team from one of Elon Musks projects OpenAI, recently defeated classic Atari games in a time substantially better that DeepMinds.

In the last month, the aforementioned poker AI from Carnegie Mellon University took part in a poker competition in China. Over 5 days, the AI once again managed to defeat a number of human players, earning its creators a cool $290,000. In May the AlphaGo team will also travel to China in order to take on who many professional GO players deem as the true world’s best, Ke Jie. AlphaGo could soon be facing stiff competition however from Chinese giant Tencent who have recently set up a staggering 250 strong AI workforce and have had success with GO themselves. Elsewhere in Asia, South Korea has held StarCraft competitions to showcase game beating AI systems, and in India, the team behind the Norah AI engine have recently acquired a large amount of investment. The competition is fierce.

Blitzkrieg 3 screenshot

Blitzkrieg 3 uses neural networks to power it’s AI

Conclusion

As you can see in just 4 months so much is already happening in the world of game AI. To keep yourself up to date we publish a quarterly report of game AI as part of our newsletter that you can sign up to below. You can also keep up to date with the weekly game AI news every Monday posted here on our blog. Thanks for reading, and here is to another exciting 4 months.