Introduction:
The online sensation “Twitch Plays Pokémon” brought together over a million players to play Pokémon Red simultaneously AI Can Play Pokémon Red about ten years ago. Each player’s keystrokes were translated into orders for the single pixelated avatar. Now, when a Magikarp develops into a Gyarado, technology raises a new issue: Can AI play Pokémon?
The original Pokémon series’ initial game has been the subject of a few years of training by Seattle-based software engineer Peter Whidden’s reinforcement learning algorithm; during that period, the AI has played the game for more than 50,000 hours. Whidden published a 33-minute YouTube video detailing the creation of the AI; nine days later, the video had 2.2 million views.
The AI uses a Pavlovian reinforcement model and is rewarded with points for levelling up Pokémon, discovering new places, winning battles, and defeating gym leaders. The errors of the AI are oddly endearing, even though they occasionally don’t completely line up with game progression, which is presumably why Whidden’s movie has gained so much attention.
One of the AI’s attempts in Pallet Town, the first location you visit in the game, is to stand still and stare at the water. It becomes stranded in a place with moving grass, water, and NPCs that walk back and forth, so even if it is just sitting still and hasn’t caught its first Pokémon yet, every frame appears to the AI to be a unique experience. However, this AI is not rushing to “catch ’em all.” It’s just taking in the splendour of the Kanto region (or perhaps it’s adopting a moral stand against making these adorable little animals engage in combat… who knows).
In the movie, Whidden argues, “So, according to our own objective, just hanging out and admiring the scenery is more rewarding than exploring the rest of the world.” This is a dichotomy we experience in real life: our curiosity propels us to produce some of the most significant discoveries, yet it also leaves us open to distractions and puts us in danger.
The AI touches our hearts once more as it goes through what would be considered a terrible occurrence at the Pokémon Center. The combined levels of every Pokémon in your group serve as a gauge for the AI’s performance.
However, the sum of all levels rapidly decreases when an AI visits a Pokémon Center and presses enough buttons to put a Pokémon into storage, giving a powerful negative signal to the AI. The total number of levels in its party included Pidgey and an unknown monster known only by the moniker “AAAAAAAA,” which brought the total to 25.
AI Can Play Pokémon Red:
AI Can Play Pokémon Red (Image Source: techcrunch.com)
Despite the AI’s capacity for trauma and its admiration for Pallet Town’s lovely pixels, it is only a machine. In early editions, the AI would become stopped at a crucial junction in the game because it could not read and comprehend dialogue. In Pokémon Red, you receive an item to return to the Pokémon Professor in Pallet Town after you reach the second town.
However, the AI had difficulty backtracking to deliver the item, making it impossible to move further. As a result, Whidden stepped ahead to have each game start after delivering the box and with Squirtle as the AI’s starter Pokémon because having a water Pokémon at your disposal makes the early game generally easier.
According to Whidden, the AI travels the furthest in the film between the first and second gym, Mt. Moon. Even if you have a real human brain, early Pokémon games’ caves are infamously challenging. The AI finally left the cave and made it to Cerulean City after Whidden recently modified some of the rewards in his programming and tried a different learning process.
With DeepMind’s AlphaGo, the first computer program to defeat a skilled Go player, other researchers have employed reinforcement learning to examine the usage of AI in gaming. But Whidden’s film has received so much attention because he is so good at demystifying complex ideas using a well-known format: Pokémon.
