The term machine learning describes a very specific branch of information technology. For scientists within this field, the final objective is the creation of AIs that can autonomously react to external stimuli. Bear in mind that we’re not talking about giving computers a conscience. Rather, these machines should have the ability to improve their skills in order to reach a predetermined objective.
One of Facebook’s latest experiments gives us a clear idea of what machine learning can achieve. The site’s RnD department put two bots in front of each other, asking them to carry out a commercial transaction. Their conversation, originally in English, soon moved to a completely new language; one which the subjects found more efficient for the task at hand.
Like most modern computing, the birth of this technology can be traced back to the early 50s. By the end of WWII, cryptographer Alan Turing theorized the idea of a “machine that can learn”. Interest in his work finally grew in the 90s, when more powerful hardware finally allowed researchers to put these notions into practice.
Still, how exactly can a machine push beyond its limits and evolve? Experts study machine learning through “artificial neural networks”, mathematical models that can mimic the way neurons in the human brain operate. This branch of IT doesn’t only give us access to the most technologically-advanced AIs, but also to several tools through which we can learn more about ourselves.
Machine Learning And Video Games
Our first example of how machine learning and video games can coexist is MarI/O, an artificial neural network that tries to complete the popular Super Mario World. The software struggles at first but, thanks to a complex series of automated adjustments, it progressively learns how to avoid death and take full advantage of the environment around the character. Eventually, the machine will manage to get to the end on its own.
Smashbot, an AI developed to play Super Smash Bros Melee, also achieves impressive results. Once it gained a touch of familiarity with the game’s rules, the computer was able to come up with its very own strategies to become invincible. If you have the time, take a look at this video where Smashbot challenges several pro players, even managing to embarrass a few of them!
Finally, Silent Hill: Shattered Memories is an example of how – by collecting information and storing data – machine learning can be used to sketch a psychological profile of the player. The game uses every resource at its disposal to reshape the experience it offers, so that it is best suited for each one of us. Left 4 Dead also relies on some form of AI: the campaign’s difficulty will dynamically shift as we go through the levels, adapting to the human team’s behavior as well as their tactics.
Automated Learning In Abbatoir Intergrade
Jared Bagley is a University of Helsinki student who’s about to get a master degree in “Machine Learning, Algorithms and Data Analysis”. For his final assignment, he decided to develop Abbattoir Intergrade; a tower defense indie game with visual novel element that heavily relies on adaptive AIs.
Exactly like we would expect, our job is that of placing traps and various weapons along the path, with the final goal of stopping the approaching waves of enemies. As we clear each of the 10 available levels, we’ll get access to new devices and traps; all of which we can also independently upgrade as we see fit.
Aesthetically speaking, Abbattoir Intergrade features a full set of hand-drawn graphical assets. There’s no denying how Bagley spent a considerable amount of time on his project. The soundtrack also deserves a mention, mainly due to how much it fits and enriches the experience.
What really makes the difference, though, is the way Abbatoir Intergrade handles enemies. At the end of each round, the software takes note of how effective the last wave was as well as how the players reacted to it. This complex set of calculations allows the title to offer a level of challenge that is always in tune with the person in front of the screen.
As more people play, they anonymously provide the developer with additional information. Jared will then collect it all and use it to demonstrate his master thesis. Being The Indie Toaster a community built around collaboration, we would love to do our part and invite you all to go through a few matches.
If you like what you’ve read so far, consider trying the game. Abbatoir Intergrade is completely free and can be downloaded from IndieExpo. Aside from its attempt to break new ground in the field of machine learning, this indie title remains an extremely pleasant one that you won’t regret spending some time on!