03.《为什么》why

This time I'm going to talk about a very difficult book called 'Why'.
 The author of this book, Perle, is very interesting. He was almost in his late teens, in his seventies and eighties, when he suddenly had a big epiphany and took his life's work and turned it on its head. At one time, Perl was the originator of the technology of machine learning. If you compare today's artificial intelligence to a building, Perle is the man who built the foundation. But in his old age, he has become the technology's fiercest critic.
 All artificial intelligence now, he felt, should be called, artificially unintelligent, because they don't understand cause and effect. They have to be taught to do cause-and-effect thinking, and that is what is called intelligence. But the problem is a bit of a problem up to that point. Because all the scientific language nowadays, does not depict cause and effect. It's like one plus one equals two. It's just an equation that holds just as well backwards; it doesn't describe which is the head and which is the tail, which is the cause and which is the effect. For a machine to learn cause and effect, you have to invent a scientific language that describes cause and effect. It's as if, back in the day, Newton wanted to study physics, found the existing mathematical tools inadequate, and invented calculus himself. That's what Perle is doing today. The old man, in one fell swoop, has actually come up with his own, scientific language for depicting cause and effect. It's this book, Why. It's the same as when Zhang Sanfeng invented Taijiquan in his old age. Don't you admire this boldness?
 But, going back to the book, this book actually presents a somewhat paradoxical phenomenon, that is, since the world does not speak of cause and effect, why do humans prefer to rely on cause and effect to think?
 The world has its laws and we have our norms. Science does not have cause and effect, humans have causal thinking, which may seem contradictory, but essentially just act on what is actually happening and abide by a variety of principles, regardless of pure right or wrong. As Phil Gerard said. Gerard said: first-class wisdom is the existence of two diametrically opposed ideas in the mind that can nevertheless go hand in hand.
Causality, too, is not completely absent, but it is difficult to observe clearly. For things to be relevant is for there to be a certain causality, except that the complexity of the universe makes the weight of a certain causality vary, some deterministic, some non-deterministic, and some outcomes are the result of many factors acting together, and discovering these causal relationships requires more scientific knowledge. Where scientific research is inadequate, only crude causal judgements can be made.
Correlation and causation, in fact, are not so clearly demarcated. Correlation is a low level of causation because there are so many factors in a complex event that a definitive outcome cannot be obtained without all factors working together.
This theory sums up very well and reminds me of an example. For example, if a student gets into a good university, this is a complex event, right? There are many factors in between, such as the student studying hard, the parents disciplining him well, the school educating him well, but if all these factors, do not work together, then it is not really possible to say whether he will get into a good university. Previous science has only been able to obtain correlations for one factor alone, but the combination of all the correlations can lead to a causal relationship. This is what the authors mean when they say that the more conditions that are identified as influences, the clearer the causality, which also means that the more certainty there is, the less uncertainty there is, and the more cause and effect is revealed.
 But there is one thing in particular that we need to be careful of when thinking in terms of cause and effect, and that is, not to over-explain. When we look back at history, it is often easy to fall into horse-trading and to explain away all past choices with as much sound logic as possible. This phenomenon is well worth noting. So, how do you avoid over-explaining? You can listen, to more people's explanations, because everyone will look at the past with a different logic, and you need to find the key elements in each of these reasonable logics, as far as possible, to analyse, and you will get the appropriate value.
译文：
这回要说的是一本难度非常高的书，叫《为什么》。
 这本书的作者珀尔很有意思。他差不多是到了晚年，七八十岁的时候，突然大彻大悟，把自己毕生的成果，来了个大翻盘。曾经，珀尔是机器学习这门技术的开山鼻祖。假如把今天的人工智能比作一幢大厦，那么珀尔就是这个造地基的人。但是，到老了，他却成了这门技术最激烈的批评者。
 他觉得，现在所有的人工智能，都应该叫，人工不智能，因为它们不懂因果关系。得教会它们做因果思考，这才叫智慧。但是，问题到这一步，就有点麻烦了。因为现在所有的科学语言，都不描绘因果。就好比一加一等于二。这只是个等式，倒过来一样成立，它不描述哪是头，哪是尾，哪是因，哪是果。要想让机器学会因果关系，你得先发明一套，描绘因果的科学语言。就好比当年，牛顿要研究物理，发现现有的数学工具不够用，自己发明了微积分一样。今天的珀尔就大有这个架势。老人家一鼓作气，居然自己搞出了一套，描绘因果的科学语言。就是这本书，《为什么》。就跟张三丰到老发明太极拳一样。这个气魄，你难道不佩服吗？
 但是，回到书里的内容，这本书其实提出了一个有点矛盾的现象，那就是，世界既然不讲因果，但是，人类为什么偏偏要依靠因果关系来思考呢？
 世界有它的规律，我们有我们的规范。科学没有因果，人类有因果思维，看似矛盾，本质就按实际情况办事，遵守各种各样的原则，不论纯粹的对错。就像菲尔.杰拉德说的：一流的智慧，就是头脑中存在两种截然相反的思想，却能并行不悖。
因果关系，也不能说完全不存在，只不过，很难被清晰地观察到。事物具有相关性就是有一定的因果性，只不过宇宙的复杂性，使得某种因果性的比重有所差异，有的是决定性因素，有的是非决定性因素，有的结果是许多因素共同作用产生的，发现这些因果关系需要更多科学的认知。在科学研究不够充分的情况下，只能做粗略的因果关系判断。
相关和因果，其实并没有那么明显的分界线。相关性是低等级的因果关系，因为一件复杂事件的因素很多，所有因素不共同起作用就无法获得确切的结果。
这个理论总结得非常好，让我想到了一个例子，比如说，一个学生考上了一个好大学，这是一个复杂事件吧，这中间的因素很多啊，比如学生努力学习、家长好好管教、学校好好教育，但是这些因素，不共同起作用，那他能不能考上好大学，还真不好说。以前的科学单纯针对一个因素只能获得相关性，但是所有相关性的结合却能够得出因果关系。这也就是作者说的，找出影响的条件越多，因果关系越明确，这也说明确定因素越多，不确定性就越少，因果便呼之欲出。
 但是，在用因果关系思考时，有一件事我们尤其需要注意，那就是，不要去过度解释。我们回看历史，往往容易陷入马后炮，用尽可能的合理的逻辑去解释过往一切选择。这个现象很值得注意。那么，怎么避免过度解释呢？你可以去倾听，更多人的解释，因为每个人看过去的逻辑都会不一样，你需要在这些各自合理的逻辑中，尽可能找到关键要素，进行分析，就会获得相应的价值。