On an RSE webinar the Chairman of 4J Studios, Chris van der Kuyl, talks about Artificial Intelligence within video games.
How is AI used within video games?
At 00:12, Kuyl introduces the wide definition of AI.
He says game designers typically ‘cheat’ because if they use the perfect AI technology it would consume all the processing capacity and then the game will typically slow down, especially if it’s a massively multiplayer game. Typically game developers use an unsophisticated side of AI which utilizes basic rules as to what one should do in certain circumstances.
There is also a use of much more sophisticated AI, such as providing the intelligence for a computer controlled character for the player to fight against, so that it feels as much as possible like playing with a real human being.
What are some examples of outstanding use of AI within games?
At 01:45, Kuyl says that the ultimate accolade is when a player can’t tell whether it’s AI software on the other end or a real player, a passing of the Turing Test so to speak. When you look for specific examples, some of the work being done recently on very sophisticated AI is exemplified when you talk about games like chess and Go where they are remarkably good and are defeating the worlds best players. (An evolution documented in the AlphaGo movie).
Companies in the UK are also developing AI algorithms that keep getting better, you don’t tell the machine how to play but instead effectively tell them the rules. Because of the fast speed and processing of these algorithms, within hours and sometimes minutes, they turn in to the world’s best players without a human telling anything. Especially board games with fixed rules are where AI has been executed the best so far.
How has the gaming industry pushed AI advancements within other industries?
From 04:32, Kuyl says it’s difficult to state that games alone are advancing AI innovations, instead it is a simultaneous progress in parallel across multiple industries and open source communities.
However gaming and the gaming industry is an ideal environment for trying out new algorithms and feeding knowledge back into the communities. In particular Chris believes that what game development is the best at it is utilizing limited resources; as he mentioned developers ‘cheat’, referring to their approach of finding the optimal model for achieving game play functionality and also the huge amounts of user testing gaming delivers.
Thus they provide the best forum for making use of algorithms in the real world and feed that knowledge back into the academic community.
Is it possible for AI engines to adapt themselves using this player feedback?
At 06:43, Kuyl mentions how all good game developers put metrics into games to see data regarding how people genetically play the game, where the hotspots are and where improvements can be made in time. However a fine balance has to be struck between the improvements and game experience because you can’t significantly make changes like people stop defeating levels that they could easily defeat earlier.
Have hardware advancements allowed developers to push the boundaries of AI?
At 08:53, Kuyl talks about the integration of such sophisticated AI has been made possible because of the overall improvement in technology. He talks specifically about GPUs (Graphical Processing Units), that companies like Nvidia and AMD make, that may sound very expensive to people but the processing power it gives the user is almost inconceivable. Twenty or thirty years ago this was a power only available in Cray supercomputers.
He also talks about how in automotive industry, people have started using autonomous AI algorithms to drive cars. He also goes on to talk about cloud based AI in massively multiplayer games which, from a player’s perspective, is a player like any other player but it’s mainly to gather feedback. Some of that can also be challenging because they have to careful to not let the user experience be affected.
What are your views on the ‘Paper Clip Theory’?
At 13:29, he says that everyone has to realize that any computer program and any instructed machine will operate as the humans have made it to. However with AI, there’s no doubt that you have to be incredibly careful. He goes on to talk about the different schools of thought, the followers of one are extremely scared about the future of AI as it gets stronger while the other are very optimistic.
He talks about robots are never meant to harm the user so the same should be kept in mind while building systems with AI to rule out any negative implications. He says the optimization of anything without thinking of the consequences around it can naturally be harmful.
(The “paperclip maximizer” is a thought experiment described by Swedish philosopher Nick Bostrom in 2003. It illustrates the existential risk that an artificial general intelligence may pose to human beings when programmed to pursue even seemingly-harmless goals, and the necessity of incorporating machine ethics into artificial intelligence design.)
Do you think improved AI will hurt future developers or become another tool?
At 20:16, he says that the availability of solutions has really made a significant difference. He doesn’t think reinventing the wheel is necessary so a lot of record developments are optimized, shared and making life easier. He also believes that these developments should be democratized to avoid the alchemist paradox where only one person has all the knowledge so naturally they become more valuable.
He references the inventor of Minecraft (Markus Persson) as a great example. He didn’t overly concern himself with it being the best technology but rather that it delivered the best gaming experience for players, and then as the venture grew organizations like Chris’ 4J Studios could worry about the detail of platform optimizations. This is a guiding principle for the type of team members Chris is looking for.