谷歌DeepMind近日推出了一项名为SIMA（可扩展、可指导、多世界智能体）的新型人工智能研究项目，旨在打造一种能像人类一样通过学习来掌握游戏技能的智能体。SIMA的目标是能够在各种游戏环境中展现出类似人类的适应性和策略，而不仅仅是一个单纯追求效率的强大力量。
据《The Verge》报道，SIMA目前仍处于研发阶段，但已经在包括《Valheim：英灵神殿》、《无人深空》和《模拟山羊》等多款不同类型的游戏上进行了训练。通过这些训练，SIMA旨在理解并掌握不同游戏的规则和策略，逐步形成跨游戏的通用学习能力。最终，DeepMind期望SIMA能够应对各种复杂的游戏环境，包括那些没有明确结束线或开放世界设定的游戏。
这一创新性的研究不仅展示了人工智能在游戏领域的潜力，也预示着未来AI在解决现实世界中复杂、非结构化问题时可能采取的学习模式。DeepMind的这一进展再次证明了他们在人工智能领域的领先地位，并可能为游戏设计、教育、甚至是科学研究带来革命性的变化。
英语如下：
News Title: “Google DeepMind Launches SIMA: A New AI Breakthrough for Gaming, Learning Like Humans”
Keywords: Google DeepMind, SIMA Agent, Game Training
News Content: Google DeepMind recently unveiled a cutting-edge AI research project called SIMA (Scalable, Instructable, Multi-World Agent), designed to create an intelligent agent capable of learning game skills like a human. The objective of SIMA is to exhibit human-like adaptability and strategy in various gaming environments, rather than simply being a force driven solely by efficiency.
As reported by The Verge, SIMA is still in development but has already undergone training across a range of diverse games, including ‘Valheim: The Norse Mythology-inspired Survival Game,’ ‘No Man’s Sky,’ and ‘ Goat Simulator.’ Through these sessions, SIMA aims to grasp the rules and strategies of different games, ultimately developing a cross-game general learning capability. Ultimately, DeepMind envisions SIMA tackling complex game environments, including those without clear endpoints or open-world settings.
This innovative research not only demonstrates the potential of AI in gaming but also hints at a possible learning model for AI when addressing real-world, complex, and unstructured problems. DeepMind’s advancement reasserts their leadership in the AI domain and might bring revolutionary changes to game design, education, and even scientific research.
【来源】https://www.theverge.com/2024/3/13/24099024/google-deepmind-ai-agent-sima-video-games
Views: 1