The Evolution of AI: A Journey through Pokémon Red


Almost a decade ago, the internet was entranced by “Twitch Plays Pokémon,” a social experiment where over a million people played Pokémon Red simultaneously. It was an innovative and chaotic exploration of crowd-sourced gaming. Fast forward to today, we’re witnessing the next evolution in gaming: artificial intelligence.

“The future of AI in gaming is not just about winning, but understanding, exploring, and sometimes, just enjoying the view.”

Paving the Path with Pixels

Peter Whidden, a Seattle-based software engineer, has spent the last few years training a reinforcement learning algorithm to navigate the nostalgic terrain of Pokémon Red. Accumulating over 50,000 hours of gameplay, the AI has learned to level up characters, explore new areas, win battles, and even beat gym leaders. However, AI has exhibited unexpected, endearing, and intriguing behaviors beyond these game-centric goals.

AI’s Unexpected Behaviours

The AI’s behavior is not always aligned with the typical progression of the game. In one instance, it stopped moving to gaze at the animated water in Pallet Town, the game’s starting location. This seemingly insignificant act might seem like an error, but it offers a fresh perspective on the AI’s ability to ‘experience’ the game. It’s as if the AI, unburdened by the rush to “catch ’em all,” is simply appreciating the pixelated beauty of the Kanto region.

AI’s Emotional Quotient?

Despite being an algorithm, the AI has exhibited what can be interpreted as emotional responses. One such incident occurred in the Pokémon Center, where the AI accidentally deposited a Pokémon into storage. This action drastically reduced the sum of all levels, sending a strong negative signal to the AI. As a result, the AI formed a negative association with the Pokémon Center and avoided it altogether in future games. Though it doesn’t possess human emotions, such a reaction hints at the ability of AI to adapt based on its experiences.

Challenges and Progress

Despite these strides, challenges persist. For instance, the AI struggled to complete an early task in the game that required backtracking due to its inability to interpret in-game dialogues. Whidden had to modify the game to begin after this task, with Squirtle as the starter Pokémon. This adjustment allowed the AI to progress further. Finally, after some tweaks in the code and using a different learning algorithm, the AI successfully navigated the challenging terrain of Mt. Moon to arrive at Cerulean City.

The Future of AI in Gaming

Whidden’s exploration is not the first time AI has been used in gaming. DeepMind’s AlphaGo made headlines as the first AI to defeat a professional Go player. However, this Pokémon experiment stands out due to Whidden’s ability to explain complex AI concepts through the lens of a beloved game. It also highlights that the future of AI in gaming is not just about winning but understanding, exploring, and sometimes, just enjoying the view.



Source link