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.

To evaluate its skills, AI is attempting to conquer the game known as “Pokemon Red”. However, it’s proving to be quite a challenge for the AI. This Pokemon Red gameplay can be viewed by anyone on the streaming platform Twitch. But considering the current progress, it might not offer the most thrilling experience.

For years, people have been captivated by the games of Pokemon Red and Blue, but surprisingly, these same titles have gained attention lately among non-human players. A notable instance is that of fish playing Pokemon, which led to an unexpected victory over Pokemon Sapphire. You might recall the Twitch Plays Pokemon event too, where participants flooded the game with commands via chat until advancement was achieved. However, in this scenario, it’s not a living creature or group that is playing Pokemon Red.

The company responsible for ChatGPT, OpenAI, has an additional AI model called o3, which one enthusiast is testing out by having it play through ‘Pokemon Red’. But, surprisingly, the AI needed a massive 80 hours just to earn the first two badges of the game. At this moment, the AI is trying to advance through the Rock Tunnel following its acquisition of the Thunder Badge from Lt. Surge. As per the live broadcast, it’s been playing for more than six days now, and between getting the Thunder Badge and reaching its current position, it took another full day and 19 hours. It seems that this AI model won’t be breaking speedrunning records in Pokemon games any time soon. However, for context, the initial Twitch Plays Pokemon run took approximately 400 hours to complete the game.

AI Is Evolving, Long Pokemon Playtime or Not

It’s quite astounding to witness an AI system learning to navigate Pokemon on its own accord, even though it requires some time. The AI was equipped with fundamental objectives for moving through the regions it’s in, but it’s left to itself to devise strategies for achieving these goals. Interestingly, it provides a detailed account of every move it made along the way. However, it should be noted that this isn’t the first AI model to engage with Pokemon, and it has some ground to cover. In fact, during a Google I/O event, Google revealed that their AI had conquered Pokemon Blue. But just like our new AI friend, it didn’t rush things either, taking 800 hours to finish the game by overcoming the Elite Four.

Currently, it’s uncertain if the O3 model can outperform Pokemon Red, nor is it clear whether it will do so quicker than Google Gemini could with its counterpart. As both O3, ChatGPT, and Google Gemini progress in their abilities, there’s a possibility that we might witness AI or comparable systems competing in more intricate games in the future. However, considering how swiftly they’re completing Pokemon games, it might still be quite a while before we see such developments.

Read More

2025-06-03 02:05