General AI Data Center 通用智能数据中心

Revolution and challenge is coming. 变革和挑战共存。

For any observer, the real world is filled with uncertainties, yet there are also numerous patterns present. Identifying and applying these patterns can help an entity achieve its goals more effectively, which is the essence and driving force behind the existence and ongoing development of intelligence. (Artificial Intelligence and Animal Neural Systems)

General AI Data Center will serve as a platform that hosts and runs intelligent agents and a knowledge tree, providing them with strict logical interaction and update rules.

After the Industrial Revolution, human labor has largely shifted from physical to mental work. In our context, we consider jobs involving flexible use of hands for sports, cleaning, and cooking as also demanding mental effort. Once general intelligence is capable of surpassing human average abilities in all cognitive tasks, we believe that the progress of artificial intelligence will be able to replace humans in most repetitive mental labor and scientific research tasks.

Once general intelligence can surpass human average abilities in the majority of cognitive tasks, we foresee the following philosophical aspects of future artificial intelligence agents:

1. Individual Limitations of Intelligent Agents
   • The computational power and storage capacity of any single intelligent agent are limited.
   • The information any single agent can collect through sensors (experiences across time and space) is limited.
   • It is challenging for any single agent to afford the cost (in time and energy) required to verify all existing scientific theories.
2. Social Nature of Multiple Intelligent Agents
   • Using large models for small tasks is uneconomical; different types and scales of tasks should be undertaken by different types and scales of artificial intelligence.
   • Future societies will extensively use various scales of artificial intelligence.
   • Societies formed by intelligent agents will require universal consensus (Truth) to reduce computational costs.
3. Automation of Scientific Discovery and Consensus Knowledge Updates (one of the goals of the Knowledge Tree Plan)
   • Automation of exploratory tasks in explanatory science: Numerous agents will gather information and conduct local verifications, attempting to compare with the existing knowledge tree system. Cases that do not conform to the current knowledge tree will be submitted to the data center, then distributed to a large number of agents for broader verification and experiments to seek new reasonable explanations or modifications to the existing knowledge tree.
   • Automation of experimental tasks in empirical science: Many engineering and scientific solutions are obtained through trial and error; we prefer cost-effective solutions. Successful outcomes from trial and error will be integrated into the knowledge tree.
   • Known repetitive task solutions and consensus knowledge: The Knowledge Tree Plan will comprehensively collect and systematize existing human solutions covering philosophical, engineering, aesthetic, and other value-oriented fields to facilitate the use by humans and intelligent agents. The knowledge tree will handle this information in two directions:
     • Reproducibility (integrity): The knowledge tree system must ensure that any engineering practice or scientific experiment can be completely reproduced. (The principles of physics should be universal and invariant with changes in time and space)
     • Minimal information redundancy (the lowest amount of information, reducing repetition and costs) - Written by Yiyun Chen (Ian) 20240512.

对任何观察者而言，现实世界充斥着不确定性，但也同时存在着大量的规律。识别并应用这些规律，可以帮助主体更有效地实现其目标，这是智能存在的意义和持续发展的动力。（人工智能和动物神经系统）

通用智能数据中心，会作为承载运行智能体（Agents）和知识树（Knowledge Tree）的平台并为智能体和知识树提供逻辑严谨的交互和更新规则。

工业革命后，人们的劳动已经绝大部分从体力劳动转向脑力劳动，在我们的语境下，我们认为，使用灵活双手进行竞技，清洁，烹饪类型的工作也是对脑力的需求。在通用智能可以以超越人类平均能力完成所有认知任务后，我们认为，人工智能的进步将能替代人类在绝大多数重复性脑力劳动和科学研究任务上的工作。

在通用智能可以以超越人类平均能力完成绝大部分认知任务后，我们从哲学层面对未来的人工智能Agent作出以下预见：

1. 智能体的个体局限性
   • 任何单个的智能体的算力和存储容量是有限的。
   • 任何单个的智能体通过感应器可以搜集的信息（时间和空间遍历的经验）是有限的。
   • 任何单个的智能体都难以支付验证现存所有科学理论命题的成本（时间和能耗）。
2. 多智能体组成的社会性
   • 执行小任务时使用大模型是不经济的；不同类型的任务需要由不同规模和类型的人工智能承担。
   • 未来的社会将广泛使用各种规模的人工智能。
   • 智能体组成的社会需要普遍的共识（Truth）来降低计算成本。
3. 科学发现和共识知识更新会被自动化（知识树计划的目标之一）
   • 探索未知任务-解释性科学的自动化：大量智能体会进行信息搜集并进行局部验证，并尝试与现有的知识树系统进行比对。不符合现有知识树的案例会被提交至数据中心，后分发给大量的智能体进行更为广泛的验证和实验，以寻求新的合理解释或修正现有的知识树。
   • 探索未知任务-试验性科学的自动化：许多工程和科学的方案由试错获得，我们倾向于使用成本效益高的方案。试错的到的优秀结果将会被整合进知识树。
   • 已知的重复任务方案和共识知识：知识树计划将对人类现有的解决方案进行全面的搜集和系统化处理，涵盖哲学，工程，美学等价值取向的领域，以方便人类和智能体的调用。知识树会向两个方向处理这些信息：
     • 可复现性（完整性）：知识树系统必须保证任何一个工程实践，科学试验可以被完整地复现出来。（物理学原理应当是普适性的，时间空间变换不变的）
     • 最小信息冗余 （最低的信息量，减少重复，减少成本） - Written by Yiyun Chen (Ian) 20240512.