人工智能与世界的本质 Artificial Intelligence and the Nature of the World

最近刚从墨尔本搬家回北京，各种事情忙到不亦乐乎，后台放的好多文章许久没发出来。实在抱歉:) 现在逐渐回归每一周到两周一篇文章的概率啦~~

Moved back to Beijing a few weeks ago. Busy with so many things and didn't keep my promise updating my blogs once or twice fortnightly. Sorry:) Now I'm getting back to the normal~ For English version, please scroll down.

今天是以准科学家的身份写一篇文章（我认为自己还称不上“科学家”的大名，至少博士头衔还没到手），内容到最后略带科幻和神学色，希望大家喜欢。最近在恶补机器学习和人工智能（AI）的知识，同时看了几本关于人工智能的书，感觉大受启发，于是想跟大家分享。我也不算内行，如有不准确请各路大神指正。文章最后分享了我阅读观看的一些书籍和视频的链接。

01

—

从对人工智能的不屑一顾到兴趣爆发

首先这里跟大家交代个背景。就在三个月之前，我其实还不怎么对“人工智能”感冒，对“大数据”也有一样的态度。有几个原因，其一，企业和媒体对这个词炒得太火了，让我对它有点心生厌恶。其二，学术圈80%（这个数字是我自己yy的不好意思）的人估计都是在用人工智能的概念给自己加加料，然而都是追着AI大牛跑，调用一下软件包，换汤不换药，也没什么创新的本质——对这样的研究我也不怎么感冒。我不知道说这个话会不经意得罪多少人，但是各位我还想跟大家做朋友，所以别走心吧，对事不对人。

但就是这几个礼拜，由于几次实习和工作申请碰壁，我开始考虑大数据和人工智能的价值。各行各业对其趋之若鹜并非空穴来风吧。我于是开始看牛津一个大牛帅锅科学家Nando de Freitas的教学视频，学习了一些基础。看看AI几十年的发展历史就知道，为数不多的科学家在AI两次冰河期里，在别人都放弃希望的日子里依然坚持着（这才是真正的科学精神！），今天都成了AI的领军人物。其中一位是华盛顿大学教授Pedro Domingos。这几天刚好在读他的书《终极算法：机器学习和人工智能如何重塑世界》(The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World)。读了不到四分之一我就已经上瘾了，因为我看到研究人工智能和机器学习（是AI的子领域）的本质是发现信息时代的“大统一理论”，完成爱因斯坦的未竟事业。我觉得人类又一次要去完成探索宇宙终极问题的挑战，而这次不是在物理学，是在认知科学和信息科学。

我突然好鸡冻！！！

02

—

渐渐地，我看到了数字化的“大统一理论”：

我们离宇宙的终极又进了一步？

《终极算法》这本书，不言自明，就是要寻求一个能解决所有问题的通用算法。如作者言：

如果存在终极算法，那么它可以通过数据学得包括过去的、现在的以及未来的所有知识。创造终极算法将是科学历史上最伟大的进步之一。

上一次和宇宙的终极走得这么近，估计是爱因斯坦探索大统一理论的时候。近代物理学飞速发展的时候，人们发现了可以解释宇宙所有现象的四种作用力。最开始，科学家都是在自己的小分支里发现小规律，一沟通才意识到，“啊！原来贵圈儿也有此种规律！” 然后小规律就渐渐变成大规律。爱因斯坦就痴迷于把已经很大的规律再拔高一层，发现宇宙本质的终极规律。然而他有生之年没成，至今也没有科学家办成。

再之前，估计是人类琢磨上帝造人的时候。其实，上帝造的不是人，而是宇宙万物的运行规律。人类想知道一个“为什么”背后的另一个“为什么”，然而发现有无穷的“为什么”。到底有没有一个根源呢？解释不清了，因为那是神的创造。

《终极算法》让我想到一步步接近宇宙最本质的层面，就像剥洋葱——用作者更优雅的方式来讲，像俄罗斯套娃。这个纷繁复杂的世界，很多时候背后只有那么一个简单而美丽的公式。然而在多大程度上人类能认识世界的规律呢？傅里叶变换（不用非要懂这是什么）告诉我们，这个世界的信息是由无穷种频率的波叠加而成的。我的理解，宇宙拥有无穷多维的信息，拥有无穷多维度的组合，会产生无穷多的可能性——请问怎么能得到全部？如果观测总是狭隘的，又如何推演规律呢？至少站在人类的角度看问题，我认为人类永远都无法知道宇宙终极理论。人类也许可以不停增加观测，和终极之间的距离可以缩短到无穷小，然而中间依然隔着无穷大的距离。从这个角度看，人类所发现的规律永远都是真实世界的一个子集，真的能教给机器的也只是这个子集或其中的一部分。所以我想，终极算法也许存在，但就像终极理论，可能人类永远也解译不出来。简单说就是下边这个图。

图1 (Figure 1)

03

—

探索终极理论有什么意义

从比较实际的角度看，我认为探索终极理论或算法是一个简化过程。之前需要对每个问题做一次解决方案，可现在只需要做一次就搞定全部。生活领域里很多方面的问题看似不相关，比如俄罗斯方块儿和邮递员怎么送货的问题看似毫不相关，其背后的数学本质是一样的。知识的发展和衍生就是这样吧，只有了解到更深层次的内容，才能往下演绎到具体案例中。一个更抽象的解决方案可以解决多个具体问题，很方便啊！这就跟用word模板一样，不用每次都编辑。

从更精神层面的角度看，至少相信终极理论的存在让我对探索世界还充满希望。看过《三体》的朋友应该知道，故事里的世界一直无规律变化，太阳有时候一天升起三次有时候好多天都不出来。如果世界的运行完全没有规律，我或许会觉得活着或探索根本没什么意义，我真的会坐立不安。然而，我相信终极理论存在，却不相信人类可以发现它！那天跟室友讨论这个问题，她问我说“那探不探索终极理论又有什么意义呢？”我说：“有！是信仰！”

As a quasi-scientist (well, not yet got my "Dr" title officially), I'm writing an article today on my thoughts of humans, machines, and deity based on my understanding of artificial intelligence (AI). It brings a little science fiction vibe at some points. Starting to develop my interest in AI, I'm inspired a lot by reading books on AI these days. Not an expert in AI, so please feel free to correct me. At the end of the article, I will share some links and interesting books on AI.

01

—

From being dismissive of AI to highly interested

I was actually not so into AI a few months ago, the same for big data. There are mainly two reasons. One is that entreprises and media are overselling the concept - so much that repels me from it. Another reason is that maybe 80% of academics (sorry that the number is just a random guess by me) are leveraging the term as gimmick for their work rather than really knowing what to do for innovation. I'm not pointing finger to any particular person, so don't take it personal:)

But it was just these few weeks - when I got some obstacles in applying for jobs - that I started to ponder on the value of AI and big data. There must be some reasons why everyone is going crazy for it. I started to learn the basics by watching the online course videos by the Oxford professor Nando de Freitas. Meanwhile, reading the history of AI development, I grew so much awe of the real AI scientists who insisted working on AI and believed in the potential of AI though everyone else gave it up at those toughest times. One of those AI avant-gardes is Petro Domingos, a professor at University of Washington. I just finished reading his book calledThe Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World. I was so absorbed reading it becauseI suddenly realised that the nature of studying AI is to find the Grand Unified Theories (GUTs) in the era of information technology - some super exciting things Einstein brought up but never ended up finishing. Human beings are once again bringing on the challenge questing the ultimate issue of the universe: not in Physics this time, but in Neuroscience and Information Science.

I got so excited!

02

—

Are we entering into the digital Grand Unified Theory and getting closer to the ultimiate end of the universe?

The Master Algorithm,self-evidently, is seeking for a universal algorithm that can solve everything.As the author put it:

If it exists, the Master Algorithm can derive all knowledge in the world—past,present, and future—from data. Inventing it would be one of the greatestadvances in the history of science.

Last time when humans were so close to the ultimate theory of the universewas probably the time when Einstein brought up the GUTs. Many scientificcoincidences took place when people from different disciplines communicatedwith each other to find the “Eureka!” That’s where small and specific rules aggregatedinto big general rules. Unfortunately, in Einstein’s lifetime, he didn’t manageto figure that out, nor has any of the scientists so far.

Fast backwards, the first time when humans were so close to the ultimatetheory of the universe was probably when human beings were thinking about how everythingwas created by the God (I’m saying this in a neutral standpoint). To be honest,I think the God (if there were one) created the rules of the universe ratherthan the beings themselves. Human beings are eager to know “whys” after “whys”.But which “why” is the ultimate WHY at all? Humans use deity as theexplanation.

The Master Algorithmreminded me ofthe analogy between cracking the ultimate rules of the universe and peelingonions layer by layer. To put it in a more elegant way as the author did, the setof rules of the universe is like Matryoshka dolls. The complex world can all bedrawn to a few simple beautiful formulas. But how much at all can humans reallyknow about the world? Fourier Transform (FT, well, you don’t have to know whatit is) tells that the world is made up of infinite frequencies (dimensions) of information.In my understanding, that meanstheuniverse encapsulates infinite combinations of possibilities, and we need tohave infinite observations in different aspects. That indicates the INFINITYbetween the ultimate rules of the universe and human beings.No matter howmany more observations we make, the gap between us and the ultimate is still smallinfinity, which is an endless INFINITY. We will always be observing only asmall subset of the reality, be learning and understanding a small subset ofthe rules, and therefore be able to teach machines only a subset of theknowledge learnt by us (as shown by the Figure 1).Maybe the ultimate rules really exist, but that’s never for us todetect them.So I never worry that machines will take over human beings.

03

—

The meaning to explore the ultimate rules

From a mundaneperspective, I think the quest for the ultimate theories is a process forsimplification. Without the general theories or the master algorithms, we mighthave to work on similar solutions multiple times targeting each specificproblem. However, the master algorithm allows us to crack multiple (seeminglyirrelevant) problems at the same time, e.g., the Tetris game and the travelsalesman problem. The nature of the mathematical rules residing behind theseproblems is the same. The deduction from an abstract solution to each specificsolution is like applying a word template for different articles.

From a more spiritual perspective, at least my belief in theultimate theories ensures me the hope for the world. If you’ve ever readThe Three-Body Problem, you probably canstill recall the scene when the world is changing irregularly. Sometimes theSun rises three times a day while other times the Sun doesn’t rise for a fewdays. If the world runs without any rule, I would probably live in such a panicway feeling nothing except the meaningless of life. I’d rather still believe inthe existence of ultimate theories, but don’t believe that humans can findthem. Is it still meaningful in this way? I think YES, because that is belief!

——————————————————————

以上纯属个人脑洞大开歪歪的结果，并不专业，班门弄斧了，欢迎大家留言讨论:))

推荐一些与话题相关的书籍和视频吧：

《终极算法：机器学习和人工智能如何重塑世界》by Pedro Domingos https://www.amazon.cn/dp/B01N2RMYKL （翻译不好，建议看英文版）

《全息宇宙》 by Michael Talbot

《终极算法》谷歌演讲 by Pedro Domingos

https://www.youtube.com/watch?v=B8J4uefCQMc

Above is some random thoughts popping up into my mind. Not professional enough. Welcome any comment or discussion with me:))

Recommendation of some good materials:

The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World. By Pedro Domingos.

Holographic Universe. By Michael Talbot

The Master Algorithm, Google Talk. By Pedro Domingos. https://www.youtube.com/watch?v=B8J4uefCQMc

欢迎喜欢文章的亲们转发此文，更欢迎大家留评论~

Welcome your likes, shares, andcomments~

If you like my aritcles, please follow me by pressing and holding the QR code!!