Artificial Intelligence In Games News – Week 18 & 19
Welcome to the weekly roundup, where we draw your attention to everything newsworthy in artificial intelligence in games! If you’ve found or shared something that you think belongs here, contact us on Twitter! Last issue, we spoke about an Oscar nominated game trailer and the shadow AI in Killer Instict. You can find that here.
AI Mobile Money
When we think of AI in games, we think of the bad guys or procedurally generated levels. Rarely do we think about the boring, but incredibly useful application of AI to games as a whole. However, Changsu Lee has highlighted 6 ways AI is being used to drive monetisation and engagement. From keeping players interested with dynamic difficulty adjustment and churn reduction techniques, to the more serious fraud detection systems that cost developers a lot of money. Perhaps one of its most interesting uses could be in predicting the in-app purchases that the player is likely to be interested in. There are also product recommendations, from our experiences playing games such as League Of Legends we have no idea what items are best to purchase mid-battle, but having an AI to guide us could effectively level the playing field.
Grow Up, Alpha GO
Alpha GO has officially retired. Upon beating the best GO players in the world, DeepMind needs to sink its teeth into a new, major challenge. Yet it’s not as simple as playing games anymore, life offers serious consequences. This article effectively summarises the progress that has been made in the last few years with AI playing games and where we go from here. Games aren’t that appealing for the experts at DeepMind, most boardgames have teams conquering them across the world, digital games such as Super Smash Bros and StarCraft II are being defeated and even Poker suffered a major defeat earlier this year. So, what should AI do next? The article points out that DeepMind has recently explored AI in healthcare but faced controversy for being given accession to millions of medical records by the NHS. As AI begins to have a real impact on our day to day lives, there will be more and more consequences, something games do not offer.
AI video game titles
In this article, a neural-network-powered AI creates really bad pickup lines and recipes, as well as some interesting game titles. Andrew Pierno from Zuma Ventors is the developer responsible for the tool and even has the process documented on his blog. If you’re looking for pickup lines such as: “I love you like an English loves charming, but I’ll be all the studies on your eyes”, or recipes such as Cabbage Pot Cookies then this is the AI for you. Its game titles are created from a 6.5mb text file created years back and has produced classic such as Mutant Tycoon and Alien Pro Baseball.. I’d play those!
AI is moving fast and it is full of technical jargon, which can easily leave beginners within the field feeling confused. Fear not, for this author helps deconstruct three key terms within the field, artificial intelligence, machine learning and deep learning to provide the perfect starting point.
Game playing AI giant
Yet another contender has entered the race to create a general AI, starting with games. Vicarious, a tech firm invested in by giants such as Musk and Zuckerberg were made famous by their captcha-busting program and are now using schema networks to outperform standard deep reinforcement powered AI. Their latest paper demonstrates their schema algorithms learning much faster than deep reinforcement techniques when playing classic Atari game Breakout.
What’s interesting about Vicarious is that they have been heavily criticised and haven’t produced papers that are particularly well cited. Their AI also only managed to beat Breakout, something Deeepmind had defeated as far back as 2013, along with 6 other games! Beating games is no longer the challenge, beating Alpha GO is!
Microsoft AI Winning at Ms Pacman
One day somebody is going to create a fascinating timeline of when each classic arcade game crumbled at the hands of AI, and last month that game was Ms Pacman. Ms Pacman was supposedly built to get arcade goers to put forward as much money as possible and was therefore easy to play but challenging to master. For this reason, the game has been subject of several research papers in the past when testing AI. Canadian AI experts Maluuba, a Microsoft-owned company, have managed to conquer the game however, achieving a perfect score of 999,990.
It achieved this by using 150 agents working in parallel to work with a senior manager to decide the best action for Ms Pacman to perform. This can become complicated with the AI needing to decide whether to pick up an extra pellet or start moving away from a ghost. The research has been published in a paper called Hybrid Reward Architecture. Many may think that this brings us one step closer to general AI but senior researchers are quick to remind us that this is like climbing Everest, vs travelling through galaxies.
Microsoft Pig Wrangling Challenge Update
A few months ago, we drew your attention to a challenge from Microsoft whereby using Minecraft they wanted to encourage researchers to explore cooperation. With a cool $20,000 up for grabs as well as a place at Microsoft’s Research AI summer school, it’s certainly an appealing proposition. We’ve now had some updates from the competition, with participants using both traditional hard coded approaches, to more modern machine learning techniques.
Ultimately, a team from Oxford took home the virtual bacon by using reinforcement learning. For those not in the know, the challenge focused on participants working with AI to capture Minecraft pigs. The goal was to build better cooperation with bots to help one day apply these techniques to the real world. Unfortunately, researchers are quick to point out that human cooperation is far more complex that what is being seen in this game environment, and how much of an impact this challenge will make is questionable.
Age Of Empires Gets a Revamp
At E3 this year, Microsoft announced Age Of Empires: Definitive Edition in order to celebrate the games 20-year anniversary. The game has truly been pulled into the 21st century with polished assets, fresh sounds, online multiplayer and new and improved AI! Although details remain thin at this moment in time, one aspect of AI we do know has been improved is pathfinding. AI agents walking through the enemy bases, getting lost or just taking obscure routes across the battlefield is an issue that plagues many RTS games and any improvement is always welcomed. This announcement is fantastic news for AI as RTS games provide developers to really build clever AI unlike what is seen in any other genre – and in recent years, RTS games have decreased in popularity.
In some BrainyBeard news we have just launched two pages on Facebook aimed at highlighting the latest developments in AI being scary as well as AI for good. We’re aiming to link to 2-3 articles per day and these will certainly be interesting for those of you that have an interest in AI outside of gaming. Come over and say hello on Facebook!
Need More News?
BrainyBeard keeps you up to date with all of the latest artificial intelligence news in games. Sign up to our newsletter below to ensure that you don’t miss out on new articles.