# AI vs. Video Games: Claude-3.7 Demonstrates Advancement in Playing Super Mario Bros.
Artificial intelligence has advanced remarkably in recent years, addressing complicated issues in areas such as healthcare, finance, and creative fields. But how proficient is AI at playing classic video games? A recent investigation by Hao Zhang, an assistant professor at UC San Diego, delves into this inquiry by evaluating AI models on titles like *Super Mario Bros.*, *Pokémon Red*, *Tetris*, and *2048*. The findings indicate that although AI encounters difficulties with certain tasks, newer models such as Claude-3.7 are indicating encouraging progress.
## AI’s Challenges and Achievements in Gaming
Recently, *BGR* reported on Claude’s efforts to engage with *Pokémon Red*. While the AI has yet to master Mt. Moon—a fairly early segment of the game—it has proven some advantages over rivals such as Gemini-1.5 Pro and GPT-4o. Nevertheless, the true breakthrough may emerge from AI’s capacity to participate in live, real-time gaming.
According to Zhang’s findings, Claude-3.7 has showcased enhanced performance in *Super Mario Bros.* relative to earlier AI models. In contrast to turn-based games like *Pokémon Red*, *Super Mario Bros.* demands quick reactions, real-time decision-making, and accurate platforming abilities. The research team discovered that Claude-3.7 was capable of outperforming other AI models by utilizing straightforward heuristics, establishing it as the most proficient AI examined to date.
## The Challenge of Real-Time Games
Engaging with a game like *Super Mario Bros.* presents a considerable challenge for AI. Unlike chess or Go, where AI has already outperformed human players, platformers necessitate a blend of spatial awareness, timing, and flexibility. AI must interpret visual stimuli, comprehend the game situation, and execute rapid decisions—all in the blink of an eye.
Although Claude-3.7 has exhibited enhancements, it still faces difficulties with certain gameplay elements. For instance, it may miscalculate jumps, delay responses to opponents, or become trapped in cycles of repetitive actions. However, in comparison to previous AI models, it shows a more refined grasp of objectives and level advancement.
## Testing AI in Additional Games
In addition to *Super Mario Bros.*, researchers are broadening their assessments to encompass other titles like *Tetris* and *2048*. These games necessitate distinct cognitive abilities—such as pattern recognition and strategic thinking—offering deeper insights into AI’s potential.
Moreover, the team has created a unique *Roblox* game titled *AI Space Escape*. This game aims to evaluate AI reasoning capabilities in real-world contexts, extending beyond conventional math and coding evaluations. The data gathered from these experiments could enhance AI’s decision-making proficiencies across various applications outside of gaming.
## The Future of AI in Gaming
Although AI is still a long way from mastering video games at a human level, these experiments underscore its potential. Future developments could lead to AI that can engage in games with human-like intuition, adjust to new challenges, and even learn from previous errors. This research could also have wider implications, such as advancing AI for robotics, autonomous systems, and real-time decision-making tasks.
Currently, Claude-3.7 might not claim the title of ultimate *Super Mario Bros.* champion, but it marks a significant advancement in AI’s ability to interact with dynamic environments. As AI continues to develop, we may someday witness models that can effortlessly tackle even the most demanding video games.
Would you tune in to watch an AI play *Super Mario Bros.* live? Share your thoughts with us!
