Pokémon No-Go: Claude’s advanced AI struggles to navigate Pokémon Red despite 3.7 upgrade

Anthropic‘s advanced AI agent Claude 3.7 Sonnet is struggling to complete the decades-old children’s game Pokémon Red, despite being one of the industry’s most sophisticated AI models. This experiment highlights the significant gap between current AI capabilities and true autonomous agency, as Claude’s difficulties with basic visual processing and navigation demonstrate that even advanced language models still face fundamental challenges when interacting with virtual environments.

The big picture: Anthropic is livestreaming “Claude Plays Pokémon” as a demonstration of AI agent capabilities, but progress has been painfully slow and inconsistent.

• Claude has managed to obtain three Gym badges and reach Cerulean City, but spent nearly 80 hours confused in Mt. Moon before finally escaping.
• The AI is currently stuck trying to find Route 5, repeatedly searching for a “gatehouse” while unable to recognize it needs to use the HM “Cut” ability on destructible trees.

Behind the struggles: Claude’s primary challenge stems from its inability to effectively process and interpret the game’s visual elements.

• According to Anthropic engineer David Hershey, “Claude’s still not particularly good at understanding what’s on the screen at all. You will see it attempt to walk into walls all the time.”
• The AI can access the game’s RAM for information like coordinates and excels at text-based portions like Pokémon battles, but struggles with the pixelated graphics of the Game Boy title.
• Ironically, Hershey suggests Claude might perform better with more visually realistic games rather than Pokémon’s low-resolution environment.

Promising signs: Despite its limitations, Claude occasionally demonstrates surprisingly human-like problem-solving abilities.

• The AI follows the same learning patterns humans would when encountering misleading in-game clues, such as being told to find Professor Oak next door only to discover he isn’t there.
• Claude 3.7 Sonnet has progressed significantly further than its predecessor, Claude 3.0 Sonnet, which couldn’t even leave the starting area of Pallet Town.

Why this matters: The experiment reveals both the progress and limitations of current AI agent technology.

• Despite rapid advances in language models, the ability to interpret visual environments and make appropriate decisions remains a significant challenge for even the most sophisticated AI systems.
• The gap between current capabilities and the industry’s goal of creating fully autonomous AI agents that can match or exceed human capabilities remains substantial.