The future of AI in Gaming
Every so often an article or a video is posted showing a program playing a game, and touting the future of Artificial Intelligence (AI) as if playing through a game somehow makes it more intelligent. Of course, that flies in the face of decades' worth of my parents scolding that gaming would make me dumb...
But does playing a video game really demonstrate the power of AI? In most examples, the "AI" is really just a trained mathematical model that translates the pseudo-random input of a bot into the game, and uses evolutionary algorithms to improve on the outcome over hundreds or even thousands of iterations. While this is impressive, the AI doesn't really "understand" what it's doing; nor is it able to make decisions if the gameplay were more dynamic or non-linear.
Similar examples can be found in the OpenAI project, where the primary input for the bot is the image on the screen. You might say, "But GameDriver, the image on the screen is our only input as human players!" This is true, but humans understand something that these bots do not; context. To a machine learning algorithm, there is only the fitness of the current run compared to all previous runs. There is no decision-making taking place based on what's happening in the game. No strategy.
Purpose-built computers like IBM's Deep Blue were created with a single objective, and could evaluate millions of possible outcomes per second at 10-11 moves ahead of it's competitors. While this is a step in the right direction, such simple rule-based games don't easily translate to the types of interactive experiences played out in modern video games.
Take the much-anticipated Cyberpunk 2077, for example. This isn't a simple game with 6 pieces on an 8x8 board like chess, but an immersive human-like experience that changes based on your actions.
Warning: Intense scenes
This type of game couldn't be played by the aforementioned Machine Learning algorithms alone, or even the multi-million dollar deep learning machines. Conversely, this is the direction video games will continue to take. More complexity and detail, higher levels of immersion, and greater levels of decision making required by the player. So what does that mean for AI in terms of playing video games?
At GameDriver, we believe AI is critical to the advancement of video games, both in terms of proving the state of Artificial Intelligence in comparison to human intelligence, and also as a way to drive forward games and other immersive experiences themselves. As these mediums become more rich and complex, they will become increasingly difficult for machines to navigate. And with so many possible outcomes, how would you even begin to test these games?
In order for AI to truly flex it's muscles in the world of truly immersive games, it will need to have insight over what is happening in the game. It needs to understand the objectives. Without this, there's simply no way for the AI to contextualize the decisions that are being made. This is where the GameDriver approach will be key. By embedding GameDriver within the game itself, we provide an interface to control game from the inside; providing the context needed to tie decisions and actions to outcomes.
It is our vision to drive games forward through better automation, and perhaps help to move AI forward in the process.