对人工智能AI的一些胡思乱想：
1.目前阶段的机器学习过程是一个数据拟合过程，是超高维空间的超级数据拟合；
2.AI模型能够破解部分事物的运行机理/规律。模型有了，基于模型的预测是有效的，从而说明破解是有效的。但破解是“暴力”的，即便其模型尚无完备的数学证明，能用海量数据和超强算力锤炼出来就行。故而破解还不是“全破解”，而只是“半破解”；
3.AI并非全能，大千世界中具有计算不可简化特性（或计算不可约性，computational irreducibility，数学大神Stephen Wolfram提出的概念）的现象，用目前的AI去处理可能就意义不大。换言之，没有一丁点儿内在结构化特征的事物，可能并无必要去穿一件AI外衣。如果非要用AI处理计算不可约，也将是在黑盒里拼凑和堆砌计算不可约的砖块。但问题是，大部分情况下，人类并不知道眼前的事物是否具备结构化特征。
4.机理/规律的“半破解”状态并不影响AI在工程上的应用。工程意义上的探索和应用先行，理论慢慢跟上。先“形而下”，再“形而上”。理论即使跟不上也不打紧，实用主义的AI工程应用已经在创造价值；
5.让机理/规律获得完备的数学证明，从而让其变得透明，依然是科学家的追求。AI的应用，反过来应该也能对这样的追求提供线索；
6.AI当前的演进道路和未来的演进道路，可能并不是同一条；
7.爱因斯坦大脑里的神经元连接和电化学信号作用，产生出了质能方程、相对论…。AI是否能够演化出爱因斯坦的大脑？不得而知。也许上帝知道。
瞎琢磨一通，见笑☺️。请路过的方家朋友留言指教。
Some Random Thoughts on Artificial Intelligence (AI):
1. At the current stage, the machine learning process is essentially a data fitting process, a super data fitting in a high-dimensional space.
2. AI models can decipher the operational mechanisms/patterns of certain phenomena. Once a model is established, predictions based on it are effective, indicating that the deciphering is successful. However, this deciphering is “brute force” — even if the model lacks complete mathematical proof, it can still be refined using massive data and immense computational power. Thus, this deciphering is not a “full deciphering” but rather a “partial deciphering.”
3. AI is not omnipotent. Phenomena in the vast world that possess the characteristic of computational irreducibility (a concept introduced by the renowned mathematician Stephen Wolfram) may not be meaningfully addressed with current AI. In other words, for things that lack any intrinsic structured features, it might not be necessary to wrap them in an AI “shell.” If we insist on using AI to deal with computational irreducibility, it would be akin to assembling and stacking bricks of irreducibility inside a black box.. The problem is, however, that in most cases, humans do not know whether the things in front of them have structured characteristics.
4. The “partial deciphering” state of mechanisms/patterns does not affect the application of AI in engineering. Exploration and application in engineering take the lead, while theory gradually follows. Practical implementation comes first, followed by theoretical understanding. Even if theory cannot keep up, it does not matter, as pragmatic AI engineering applications are already creating value.
5. Achieving complete mathematical proof of mechanisms/patterns to make them transparent remains the pursuit of scientists. Conversely, the application of AI should also provide clues to such pursuits.
6. The current and future evolutionary paths of AI may not be the same.
7. The neural connections and electrochemical signals in Einstein’s brain produced the mass-energy equation, relativity, etc. Can AI evolve to the level of Einstein’s brain? It is unknown. Maybe only God knows.
These are just some random musings, please feel free to leave your comments and suggestions. ☺️
