Ever wondered if artificial intelligence could tackle the challenge of playing Pokémon?
Well, the answer is a resounding yes! Thanks to the tireless efforts of Seattle-based software engineer, Peter Whidden, a reinforcement learning algorithm has been trained to navigate the classic Pokémon Red game, clocking in over 50,000 hours of gameplay.
Whidden's fascinating journey of AI development has been documented in a 33-minute YouTube video that has already garnered over 2.2 million views in just nine days. The video showcases the AI's journey through the game, complete with charming failures and unexpected behaviours that have captured the hearts of viewers worldwide.
The AI's reinforcement model is based on a point-based incentive system, encouraging the AI to level up Pokémon, explore new areas, win battles, and beat gym leaders. However, sometimes these incentives don't perfectly align with progression in the game, leading to some unique and endearing AI behaviours.
In one instance, the AI simply stops to admire the animated water in Pallet Town, seemingly mesmerised by the pixelated scenery. In another, it experiences a form of 'trauma' at the Pokémon Centre, developing a negative association with the location after accidentally depositing a Pokémon into storage, reducing the total levels of its party.
Despite these quirks, the AI has made significant progress. Initially, it struggled with backtracking to deliver a parcel, a task required to progress in the game. To overcome this, Whidden adjusted the game to start after the delivery, with Squirtle as the AI's starter Pokémon. With these changes, the AI has managed to navigate its way to Cerulean City, a significant milestone.
Whidden's work is a testament to the potential of reinforcement learning in gaming, a field that has seen significant interest with projects like DeepMind's AlphaGo. However, it's Whidden's ability to explain complex concepts through the familiar medium of Pokémon that has truly captured the public's imagination.
So, whether you're a Pokémon fan, an AI enthusiast, or simply curious, this AI's journey through the Kanto region is sure to entertain and inspire. Who knows, maybe one day we'll see AI competing in Pokémon Championships!
Made with TRUST_AI - see the Charter: https://www.modelprop.co.uk/trust-ai
Comments