AI is Having Problems Beating Pokemon Red

Summary

AI is playing Pokemon Red, taking 80 hours to secure 2 badges, still progressing through Rock Tunnel after 6+ days.
The AI model is reasoning its moves, analyzing maps visually, and saving to memory, not a quick speedrun.
OpenAI's o3 AI isn't the first to play Pokemon; Google's AI beat Pokemon Blue after 800 hours, showing evolution in AI.

In order to test its capabilities, AI is playing through Pokemon Red, but it's not having an easy time of it. The Pokemon Red playthrough is something that anyone can watch via streaming platform Twitch, but given the progress made so far, it may not be the most exciting thing to do.

While plenty of Pokemon Red and Blue's target audience - humans - have enjoyed the games over the decades, strangely, the same titles have also seen non-human players go viral in recent times. One famous example of this is fish playing Pokemon, which had a surprising outcome, as the fish actually beat Pokemon Sapphire. Some fans may also remember the Twitch Plays Pokemon experience, where players would bombard the game with commands via the chat until progress was made. In this case, though, it's neither a living entity nor a group playing through Pokemon Red.

pokemon-scarlet-and-violet-video-shows-switch-2-gameplay-60-fps-no-skips-free-update

Nintendo Shows Off Pokemon Scarlet and Violet Switch 2 Upgraded Performance

Days before the launch of the Switch 2, Nintendo demonstrates how Pokemon Scarlet and Violet will perform on the upcoming system.

Posts

By James Ratcliff

OpenAI, the organization behind ChatGPT, has another AI model called o3, and one community member is putting it through its paces by having it play Pokemon Red. However, it took the AI model a whopping 80 hours in order to just get the first two badges of the game. At the time of writing, the AI model is attempting to forge its way through the game's Rock Tunnel after successfully securing the Thunder Badge from Lt. Surge. According to the stream, it's been playing for over six days, and between winning the Thunder Badge and making its way to its current location, it spent another whole day and 19 hours. It's obvious that the AI model isn't going to be speedrunning the games anytime soon. However, for the sake of comparison, the original Twitch Plays Pokemon attempt took roughly 400 hours to complete the game.

AI Is Evolving, Long Pokemon Playtime or Not

While it's certainly taking some time, it's nonetheless shocking to see an AI system figuring out how to get through Pokemon on its own. The AI was given some basic objectives to get through the areas it's in, but otherwise, it's trying to figure out how to achieve the goals on its own, and offers a thorough report on every decision it made. With that said, it's actually not the first AI model to play Pokemon, and it has some catching up to do. During a recent Google I/O event, the company announced that its own AI had beaten Pokemon Blue. However, it wasn't quick about it either, taking 800 hours to complete the game by beating the Elite Four.

At the moment, it remains to be seen if the o3 model can successfully beat Pokemon Red, and whether it will do so faster than Google Gemini could the counterpart. As ChatGPT, o3, and Google Gemini continue to develop new capabilities, it's possible that gamers could one day see these AI or similar systems playing more complex games. Given the speed at which they're finishing Pokemon, though, it will probably be quite some time yet.

Rating block community and brand ratings Image