Trackmania, a popular and wildly unpredictable racing game abundant with user-generated tracks peppered with outrageous corners, corkscrews, and other surreal obstacles, is the digital playground for a player named Yosh who spent three years perfecting an AI seemingly superior to him. Known for its high-bar driving model and competitive nature, Trackmania stands as an ideal experimental lab for AI training. Propelling it all was Yosh's relentless quest to construct an AI capable of outdoing his personal best timings.
Yosh began by creating a simplistic track with gentle curves resembling a mildly tilted waveform. After multiple iterations, the AI successfully learned to complete the course with respectable timelines, albeit lagging behind Yosh's completion time. Yosh describes the experience as a repetitive cycle of trial and error, constantly tweaking the AI and re-running the training, with more failures than successes.
This iterative process eventually bore fruit when the AI began beating Yosh's record, improving its timing gradually, shaving off fractions of a second from each lap. However, this victory was limited to a simple racetrack. Yosh's new challenge for the AI came in the form of a boundary-less colossal maze. This new test surface demanded around 35 training hours from the AI to eventually best Yosh's time, an impressive achievement considering its human competitor had 17 years of experience.
However, another hurdle awaited the AI—it hadn't been trained to use brakes. Yosh, to maintain equality, hadn't applied brakes either. But with the AI's successive triumphs, Yosh decided to introduce brakes into the equation. Unleashing a new record almost nine seconds faster by employing brakes and drifting around the corners, he presented a new challenge to his AI. Impressively, the AI defeated this record as well, despite suffering from a no-brake disadvantage.
Once the AI mastered the art of braking and drifting, it appeared unstoppable. On a massive maze-style racetrack, the AI's mechanical consistency shone as it surpassed small errors a human opponent might make. It demonstrated its prowess by beating Yosh on even shorter variants of the track.
The crux of this intriguing tale isn't just about an AI eclipsing a human's timing in a computer game. It provides an enlightening view into how machine learning operates and the considerable effort needed to train an AI algorithm for specific purposes. For instance, there was a recent gaming AI that required a whopping 7,000 hours to overcome the first gym in Pokemon Red.
AI-generated content continues to spark debates, and with valid reasons. However, advancements in machine learning technology suggest AI presence in the gaming world is on an upward trajectory. Undertakings such as Yosh's paves the way for more understanding of how this technology really works. Meanwhile, let's cherish our victories in the realm of racing games before potent AIs decide to contest our presence there.
You must be logged in to post a comment!
